Merge remote-tracking branch 'origin/master' into libjpeg-turbo

* origin/master: (108 commits) Bump version number to 3.1. jpegyuv: fix memory leak when path is invalid jpegyuv: fix memory leak when @image_buffer allocation fails yuvjpeg: fix memory leak when @image_buffer allocation fails jpegtran: Do not leak the input and output buffers Fix previous commit Scan optimization: return error when unable to copy data buffer cjpeg option for baseline quant tables Fix #153 rdpng: convert 16-bit input to 8-bit Larger number of DC trellis candidates Fix overflow issue #157 Const on getters Const on simple getters and copy source Expanded .gitignore Add pkg-config requirement Re-order links. Declare inbuffer const Oops. Delete the duplicate copy of [lib]turbojpeg.dll in the binary directory when uninstalling the package. Get rid of changelog file that we don't update. ...
2016-04-27 23:35:38 +01:00
parent 1a830d52a4 5198654f73
commit a22ab9f3df
87 changed files with 5124 additions and 1775 deletions
--- a/.gitignore
+++ b/.gitignore
@@ -1,14 +1,44 @@
 Makefile.in
 Makefile
 /CMakeFiles
 /autom4te.cache
 /aclocal.m4
 /compile
 /configure
 /depcomp
 /install-sh
 /libtool
 /missing
-/stamp-h1
+/stamp-h*
 /java/classnoinst.stamp
 /pkgscripts/
 /jconfig.h
 /jconfigint.h
 /config.guess
 /config.h
 /config.h.in
 /config.log
 /config.status
 /config.sub
 /ltmain.sh
 /ar-lib
 /libjpeg.map
 /.libs/
 /simd/.libs/
 /simd/jsimdcfg.inc
 *.o
 *.lo
 *.la
 /cjpeg
 /djpeg
 /jcstest
 /jpegtran
 /jpegyuv
 /md5/md5cmp
 /rdjpgcom
 /test_enc_*
 /tjbench
 /tjbenchtest
 /tjunittest
 /wrjpgcom
 /yuvjpeg
--- a/BUILDING.txt
+++ b/BUILDING.txt
@@ -7,6 +7,7 @@
 Build Requirements
 ==================
 -- pkg-config
 -- autoconf 2.56 or later
 -- automake 1.7 or later
 -- libtool 1.4 or later
@@ -35,21 +36,9 @@ Build Requirements
   NOTE: the NASM build will fail if texinfo is not installed.
-- GCC v4.1 or later recommended for best performance
+-- GCC v4.1 (or later) or clang recommended
   * Beginning with Xcode 4, Apple stopped distributing GCC and switched to
     the LLVM compiler.  Xcode v4.0 through v4.6 provides a GCC front end
     called LLVM-GCC.  Unfortunately, as of this writing, neither LLVM-GCC nor
     the LLVM (clang) compiler produces optimal performance with libmozjpeg.
     Building libmozjpeg with LLVM-GCC v4.2 results in a 10% performance
     degradation when compressing using 64-bit code, relative to building
     libmozjpeg with GCC v4.2.  Building libmozjpeg with LLVM (clang)
     results in a 20% performance degradation when compressing using 64-bit
     code, relative to building libmozjpeg with GCC v4.2.  If you are
     running Snow Leopard or earlier, it is suggested that you continue to use
     Xcode v3.2.6, which provides GCC v4.2.  If you are using Lion or later, it
     is suggested that you install Apple GCC v4.2 through MacPorts.
-- If building the MozJPEG Java wrapper, JDK or OpenJDK 1.5 or later is
+-- If building the TurboJPEG Java wrapper, JDK or OpenJDK 1.5 or later is
   required.  Some systems, such as Solaris 10 and later and Red Hat Enterprise
   Linux 5 and later, have this pre-installed.  On OS X 10.5 and later, it will
   be necessary to install the Java Developer Package, which can be downloaded
@@ -64,30 +53,40 @@ Out-of-Tree Builds
 Binary objects, libraries, and executables are generated in the same directory
 from which configure was executed (the "binary directory"), and this directory
-need not necessarily be the same as the libmozjpeg source directory.  You
+need not necessarily be the same as the mozjpeg source directory.  You
 can create multiple independent binary directories, in which different versions
-of libmozjpeg can be built from the same source tree using different
+of mozjpeg can be built from the same source tree using different
 compilers or settings.  In the sections below, {build_directory} refers to the
-binary directory, whereas {source_directory} refers to the libmozjpeg source
+binary directory, whereas {source_directory} refers to the mozjpeg source
 directory.  For in-tree builds, these directories are the same.
-======================
+================
-Building libmozjpeg
+Building mozjpeg
-======================
+================
-The following procedure will build libmozjpeg on Linux, FreeBSD, Cygwin, and
+The following procedure will build mozjpeg on Linux, FreeBSD, Cygwin, and
 Solaris/x86 systems (on Solaris, this generates a 32-bit library.  See below
 for 64-bit build instructions.)
 Simple Release tar.gz Source Build
 ----------------------------------
  cd {source_directory}
  ./configure [additional configure flags]
  make
 Non-Release Source Build (e.g. GitHub clone)
 --------------------------------------------
  cd {source_directory}
  autoreconf -fiv
  cd {build_directory}
  sh {source_directory}/configure [additional configure flags]
  make
-NOTE: Running autoreconf in the source directory is usually only necessary if
+Results
-building libmozjpeg from the SVN repository.
+-------
 This will generate the following files under .libs/
@@ -110,28 +109,28 @@ This will generate the following files under .libs/
  libjpeg.dll.a (Cygwin)
      Import library for the libjpeg API
-  libmozjpeg.a
+  libturbojpeg.a
-      Static link library for the MozJPEG API
+      Static link library for the TurboJPEG API
-  libmozjpeg.so.0.0.0 (Linux, Unix)
+  libturbojpeg.so.0.1.0 (Linux, Unix)
-  libmozjpeg.0.0.0.dylib (OS X)
+  libturbojpeg.0.1.0.dylib (OS X)
-  cygmozjpeg-0.dll (Cygwin)
+  cygturbojpeg-0.dll (Cygwin)
-      Shared library for the MozJPEG API
+      Shared library for the TurboJPEG API
-  libmozjpeg.so (Linux, Unix)
+  libturbojpeg.so (Linux, Unix)
-  libmozjpeg.dylib (OS X)
+  libturbojpeg.dylib (OS X)
-      Development symlink for the MozJPEG API
+      Development symlink for the TurboJPEG API
-  libmozjpeg.dll.a (Cygwin)
+  libturbojpeg.dll.a (Cygwin)
-      Import library for the MozJPEG API
+      Import library for the TurboJPEG API
 libjpeg v7 or v8 API/ABI Emulation
 ----------------------------------
 Add --with-jpeg7 to the configure command line to build a version of
-libmozjpeg that is API/ABI-compatible with libjpeg v7.  Add --with-jpeg8 to
+mozjpeg that is API/ABI-compatible with libjpeg v7.  Add --with-jpeg8 to
-the configure command to build a version of libmozjpeg that is
+the configure command to build a version of mozjpeg that is
 API/ABI-compatible with libjpeg v8.  See README-turbo.txt for more information
 on libjpeg v7 and v8 emulation.
@@ -140,7 +139,7 @@ In-Memory Source/Destination Managers
 -------------------------------------
 When using libjpeg v6b or v7 API/ABI emulation, add --without-mem-srcdst to the
-configure command line to build a version of libmozjpeg that lacks the
+configure command line to build a version of mozjpeg that lacks the
 jpeg_mem_src() and jpeg_mem_dest() functions.  These functions were not part of
 the original libjpeg v6b and v7 APIs, so removing them ensures strict
 conformance with those APIs.  See README-turbo.txt for more information.
@@ -150,7 +149,7 @@ Arithmetic Coding Support
 -------------------------
 Since the patent on arithmetic coding has expired, this functionality has been
-included in this release of libmozjpeg.  libmozjpeg's implementation is
+included in this release of mozjpeg.  mozjpeg's implementation is
 based on the implementation in libjpeg v8, but it works when emulating libjpeg
 v7 or v6b as well.  The default is to enable both arithmetic encoding and
 decoding, but those who have philosophical objections to arithmetic coding can
@@ -158,11 +157,11 @@ add --without-arith-enc or --without-arith-dec to the configure command line to
 disable encoding or decoding (respectively.)
-MozJPEG Java Wrapper
+TurboJPEG Java Wrapper
 ----------------------
 Add --with-java to the configure command line to incorporate an optional Java
-Native Interface wrapper into the MozJPEG shared library and build the Java
+Native Interface wrapper into the TurboJPEG shared library and build the Java
-front-end classes to support it.  This allows the MozJPEG shared library to
+front-end classes to support it.  This allows the TurboJPEG shared library to
 be used directly from Java applications.  See java/README for more details.
 You can set the JAVAC, JAR, and JAVA configure variables to specify
@@ -173,9 +172,9 @@ to specify arguments that should be passed to the C compiler when building the
 JNI wrapper.  Run 'configure --help' for more details.
-========================
+==================
-Installing libmozjpeg
+Installing mozjpeg
-========================
+==================
 If you intend to install these libraries and the associated header files, then
 replace 'make' in the instructions above with
@@ -188,16 +187,16 @@ For example,
 will install the header files in /usr/local/include and the library files in
 /usr/local/lib64.  If 'prefix' and 'libdir' are not specified, then the default
-is to install the header files in /opt/libmozjpeg/include and the library
+is to install the header files in /opt/mozjpeg/include and the library
-files in /opt/libmozjpeg/lib32 (32-bit) or /opt/libmozjpeg/lib64
+files in /opt/mozjpeg/lib32 (32-bit) or /opt/mozjpeg/lib64
 (64-bit.)
 NOTE: You can specify a prefix of /usr and a libdir of, for instance,
 /usr/lib64 to overwrite the system's version of libjpeg.  If you do this,
 however, then be sure to BACK UP YOUR SYSTEM'S INSTALLATION OF LIBJPEG before
-overwriting it.  It is recommended that you instead install libmozjpeg into
+overwriting it.  It is recommended that you instead install mozjpeg into
 a non-system directory and manipulate the LD_LIBRARY_PATH or create symlinks
-to force applications to use libmozjpeg instead of libjpeg.  See
+to force applications to use mozjpeg instead of libjpeg.  See
 README-turbo.txt for more information.
@@ -284,10 +283,10 @@ Add
  CC=cc
-to the configure command line.  libmozjpeg will automatically be built with
+to the configure command line.  mozjpeg will automatically be built with
 the maximum optimization level (-xO5) unless you override CFLAGS.
-To build a 64-bit version of libmozjpeg using Oracle Solaris Studio, add
+To build a 64-bit version of mozjpeg using Oracle Solaris Studio, add
  --host x86_64-pc-solaris CC=cc CFLAGS='-xO5 -m64' LDFLAGS=-m64
@@ -304,19 +303,19 @@ Use CMake (see recipes below)
 ARM Support
 ===========
-This release of libmozjpeg can use ARM NEON SIMD instructions to accelerate
+This release of mozjpeg can use ARM NEON SIMD instructions to accelerate
 JPEG compression/decompression by approximately 2-4x on ARMv7 and later
-platforms.  If libmozjpeg is configured on an ARM Linux platform, then the
+platforms.  If mozjpeg is configured on an ARM Linux platform, then the
 build system will automatically include the NEON SIMD routines, if they are
 supported.  Build instructions for other ARM-based platforms follow.
-Building libmozjpeg for iOS
+Building mozjpeg for iOS
------------------------------
+------------------------
 iOS platforms, such as the iPhone and iPad, use ARM processors, some of which
 support NEON instructions.  Additional steps are required in order to build
-libmozjpeg for these platforms.
+mozjpeg for these platforms.
 Additional build requirements:
@@ -324,6 +323,9 @@ Additional build requirements:
  (https://sourceforge.net/p/libjpeg-turbo/code/HEAD/tree/gas-preprocessor)
  should be installed in your PATH.
 ARM 32-bit Build (Xcode 4.6.x and earlier, LLVM-GCC):
 Set the following shell variables for simplicity:
  Xcode 4.2 and earlier:
@@ -332,55 +334,82 @@ Set the following shell variables for simplicity:
  IOS_PLATFORMDIR=/Applications/Xcode.app/Contents/Developer/Platforms/iPhoneOS.platform
  IOS_SYSROOT=$IOS_PLATFORMDIR/Developer/SDKs/iPhoneOS*.sdk
  Xcode 4.6.x and earlier:
  IOS_GCC=$IOS_PLATFORMDIR/Developer/usr/bin/arm-apple-darwin10-llvm-gcc-4.2
  Xcode 5.0.x and later:
  IOS_GCC=/Applications/Xcode.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/clang
-  ARM v6 (code will run on all iOS devices, not SIMD-accelerated):
+  ARMv6 (code will run on all iOS devices, not SIMD-accelerated):
  [NOTE: Requires Xcode 4.4.x or earlier]
  IOS_CFLAGS="-march=armv6 -mcpu=arm1176jzf-s -mfpu=vfp"
-  ARM v7 (code will run on iPhone 3GS-4S/iPad 1st-3rd Generation and newer):
+  ARMv7 (code will run on iPhone 3GS-4S/iPad 1st-3rd Generation and newer):
  Xcode 4.6.x and earlier:
  IOS_CFLAGS="-march=armv7 -mcpu=cortex-a8 -mtune=cortex-a8 -mfpu=neon"
  Xcode 5.0.x and later:
  IOS_CFLAGS="-arch armv7"
-  ARM v7s (code will run on iPhone 5/iPad 4th Generation and newer):
+  ARMv7s (code will run on iPhone 5/iPad 4th Generation and newer):
  [NOTE: Requires Xcode 4.5 or later]
  Xcode 4.6.x and earlier:
  IOS_CFLAGS="-march=armv7s -mcpu=swift -mtune=swift -mfpu=neon"
  Xcode 5.0.x and later:
  IOS_CFLAGS="-arch armv7s"
-Follow the procedure under "Building libmozjpeg" above, adding
+Follow the procedure under "Building mozjpeg" above, adding
-  --host arm-apple-darwin10 --enable-static --disable-shared \
+  --host arm-apple-darwin10 \
    CC="$IOS_GCC" LD="$IOS_GCC" \
    CFLAGS="-mfloat-abi=softfp -isysroot $IOS_SYSROOT -O3 $IOS_CFLAGS" \
    LDFLAGS="-mfloat-abi=softfp -isysroot $IOS_SYSROOT $IOS_CFLAGS"
-to the configure command line.  If using Xcode 5.0.x or later, also add
+to the configure command line.
 ARM 32-bit Build (Xcode 5.0.x and later, Clang):
 Set the following shell variables for simplicity:
  IOS_PLATFORMDIR=/Applications/Xcode.app/Contents/Developer/Platforms/iPhoneOS.platform
  IOS_SYSROOT=$IOS_PLATFORMDIR/Developer/SDKs/iPhoneOS*.sdk
  IOS_GCC=/Applications/Xcode.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/clang
  ARMv7 (code will run on iPhone 3GS-4S/iPad 1st-3rd Generation and newer):
  IOS_CFLAGS="-arch armv7"
  ARMv7s (code will run on iPhone 5/iPad 4th Generation and newer):
  IOS_CFLAGS="-arch armv7s"
 Follow the procedure under "Building libjpeg-turbo" above, adding
  --host arm-apple-darwin10 \
    CC="$IOS_GCC" LD="$IOS_GCC" \
    CFLAGS="-mfloat-abi=softfp -isysroot $IOS_SYSROOT -O3 $IOS_CFLAGS" \
    LDFLAGS="-mfloat-abi=softfp -isysroot $IOS_SYSROOT $IOS_CFLAGS" \
    CCASFLAGS="-no-integrated-as $IOS_CFLAGS"
 to the configure command line.
 ARMv8 64-bit Build (Xcode 5.0.x and later, Clang):
 Code will run on iPhone 5S/iPad Mini 2 and newer.
 Set the following shell variables for simplicity:
  IOS_PLATFORMDIR=/Applications/Xcode.app/Contents/Developer/Platforms/iPhoneOS.platform
  IOS_SYSROOT=$IOS_PLATFORMDIR/Developer/SDKs/iPhoneOS*.sdk
  IOS_GCC=/Applications/Xcode.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/clang
  IOS_CFLAGS="-arch arm64"
 Follow the procedure under "Building libjpeg-turbo" above, adding
  --host aarch64-apple-darwin \
    CC="$IOS_GCC" LD="$IOS_GCC" \
    CFLAGS="-isysroot $IOS_SYSROOT -O3 $IOS_CFLAGS" \
    LDFLAGS="-isysroot $IOS_SYSROOT $IOS_CFLAGS"
 to the configure command line.
 NOTE:  You can also add -miphoneos-version-min={version} to $IOS_CFLAGS above
 in order to support older versions of iOS than the default version supported by
 the SDK.
-Once built, lipo can be used to combine the ARM v6, v7, and/or v7s variants
+Once built, lipo can be used to combine the ARMv6, v7, v7s, and/or v8 variants
 into a universal library.
 NOTE: If you are building libmozjpeg from the "official" project tarball,
 then it is highly likely that you will need to run 'autoreconf -fiv' in the
 source tree prior to building ARM v7 or v7s iOS binaries using the techniques
 described above.  Otherwise, you may get a libtool error such as "unable to
 infer tagged configuration."
 Building libjpeg-turbo for Android
 ----------------------------------
@@ -440,12 +469,12 @@ Build Requirements
   http://msdn.microsoft.com/en-us/windows/bb980924.aspx
   The Windows SDK includes both 32-bit and 64-bit Visual C++ compilers and
-   everything necessary to build libmozjpeg.
+   everything necessary to build mozjpeg.
   * You can also use Microsoft Visual Studio Express Edition, which is a free
     download.  (NOTE: versions prior to 2012 can only be used to build 32-bit
     code.)
-   * If you intend to build libmozjpeg from the command line, then add the
+   * If you intend to build mozjpeg from the command line, then add the
     appropriate compiler and SDK directories to the INCLUDE, LIB, and PATH
     environment variables.  This is generally accomplished by executing
     vcvars32.bat or vcvars64.bat and SetEnv.cmd.  vcvars32.bat and
@@ -467,7 +496,7 @@ Build Requirements
 -- NASM (http://www.nasm.us/) 0.98 or later (NASM 2.05 or later is required for
   a 64-bit build)
-- If building the MozJPEG Java wrapper, JDK 1.5 or later is required.  This
+-- If building the TurboJPEG Java wrapper, JDK 1.5 or later is required.  This
   can be downloaded from http://www.java.com.
@@ -477,17 +506,17 @@ Out-of-Tree Builds
 Binary objects, libraries, and executables are generated in the same directory
 from which cmake was executed (the "binary directory"), and this directory need
-not necessarily be the same as the libmozjpeg source directory.  You can
+not necessarily be the same as the mozjpeg source directory.  You can
 create multiple independent binary directories, in which different versions of
-libmozjpeg can be built from the same source tree using different compilers
+mozjpeg can be built from the same source tree using different compilers
 or settings.  In the sections below, {build_directory} refers to the binary
-directory, whereas {source_directory} refers to the libmozjpeg source
+directory, whereas {source_directory} refers to the mozjpeg source
 directory.  For in-tree builds, these directories are the same.
-======================
+================
-Building libmozjpeg
+Building mozjpeg
-======================
+================
 Visual C++ (Command Line)
@@ -497,7 +526,7 @@ Visual C++ (Command Line)
  cmake -G "NMake Makefiles" -DCMAKE_BUILD_TYPE=Release {source_directory}
  nmake
-This will build either a 32-bit or a 64-bit version of libmozjpeg, depending
+This will build either a 32-bit or a 64-bit version of mozjpeg, depending
 on which version of cl.exe is in the PATH.
 The following files will be generated under {build_directory}:
@@ -508,12 +537,12 @@ The following files will be generated under {build_directory}:
      DLL for the libjpeg API
  sharedlib/jpeg.lib
      Import library for the libjpeg API
-  mozjpeg-static.lib
+  turbojpeg-static.lib
-      Static link library for the MozJPEG API
+      Static link library for the TurboJPEG API
-  mozjpeg.dll
+  turbojpeg.dll
-      DLL for the MozJPEG API
+      DLL for the TurboJPEG API
-  mozjpeg.lib
+  turbojpeg.lib
-      Import library for the MozJPEG API
+      Import library for the TurboJPEG API
 {version} is 62, 7, or 8, depending on whether libjpeg v6b (default), v7, or
 v8 emulation is enabled.
@@ -536,7 +565,7 @@ and 64-bit builds.
 You can then open ALL_BUILD.vcproj in Visual Studio and build one of the
 configurations in that project ("Debug", "Release", etc.) to generate a full
-build of libmozjpeg.
+build of mozjpeg.
 This will generate the following files under {build_directory}:
@@ -546,12 +575,12 @@ This will generate the following files under {build_directory}:
      DLL for the libjpeg API
  sharedlib/{configuration}/jpeg.lib
      Import library for the libjpeg API
-  {configuration}/mozjpeg-static.lib
+  {configuration}/turbojpeg-static.lib
-      Static link library for the MozJPEG API
+      Static link library for the TurboJPEG API
-  {configuration}/mozjpeg.dll
+  {configuration}/turbojpeg.dll
-      DLL for the MozJPEG API
+      DLL for the TurboJPEG API
-  {configuration}/mozjpeg.lib
+  {configuration}/turbojpeg.lib
-      Import library for the MozJPEG API
+      Import library for the TurboJPEG API
 {configuration} is Debug, Release, RelWithDebInfo, or MinSizeRel, depending on
 the configuration you built in the IDE, and {version} is 62, 7, or 8,
@@ -576,12 +605,12 @@ This will generate the following files under {build_directory}
      DLL for the libjpeg API
  sharedlib/libjpeg.dll.a
      Import library for the libjpeg API
-  libmozjpeg.a
+  libturbojpeg.a
-      Static link library for the MozJPEG API
+      Static link library for the TurboJPEG API
-  libmozjpeg.dll
+  libturbojpeg.dll
-      DLL for the MozJPEG API
+      DLL for the TurboJPEG API
-  libmozjpeg.dll.a
+  libturbojpeg.dll.a
-      Import library for the MozJPEG API
+      Import library for the TurboJPEG API
 {version} is 62, 7, or 8, depending on whether libjpeg v6b (default), v7, or
 v8 emulation is enabled.
@@ -599,8 +628,8 @@ libjpeg v7 or v8 API/ABI Emulation
 -----------------------------------
 Add "-DWITH_JPEG7=1" to the cmake command line to build a version of
-libmozjpeg that is API/ABI-compatible with libjpeg v7.  Add "-DWITH_JPEG8=1"
+mozjpeg that is API/ABI-compatible with libjpeg v7.  Add "-DWITH_JPEG8=1"
-to the cmake command to build a version of libmozjpeg that is
+to the cmake command to build a version of mozjpeg that is
 API/ABI-compatible with libjpeg v8.  See README-turbo.txt for more information
 on libjpeg v7 and v8 emulation.
@@ -609,7 +638,7 @@ In-Memory Source/Destination Managers
 -------------------------------------
 When using libjpeg v6b or v7 API/ABI emulation, add -DWITH_MEM_SRCDST=0 to the
-CMake command line to build a version of libmozjpeg that lacks the
+CMake command line to build a version of mozjpeg that lacks the
 jpeg_mem_src() and jpeg_mem_dest() functions.  These functions were not part of
 the original libjpeg v6b and v7 APIs, so removing them ensures strict
 conformance with those APIs.  See README-turbo.txt for more information.
@@ -619,7 +648,7 @@ Arithmetic Coding Support
 -------------------------
 Since the patent on arithmetic coding has expired, this functionality has been
-included in this release of libmozjpeg.  libmozjpeg's implementation is
+included in this release of mozjpeg.  mozjpeg's implementation is
 based on the implementation in libjpeg v8, but it works when emulating libjpeg
 v7 or v6b as well.  The default is to enable both arithmetic encoding and
 decoding, but those who have philosophical objections to arithmetic coding can
@@ -627,11 +656,11 @@ add "-DWITH_ARITH_ENC=0" or "-DWITH_ARITH_DEC=0" to the cmake command line to
 disable encoding or decoding (respectively.)
-MozJPEG Java Wrapper
+TurboJPEG Java Wrapper
 ----------------------
 Add "-DWITH_JAVA=1" to the cmake command line to incorporate an optional Java
-Native Interface wrapper into the MozJPEG shared library and build the Java
+Native Interface wrapper into the TurboJPEG shared library and build the Java
-front-end classes to support it.  This allows the MozJPEG shared library to
+front-end classes to support it.  This allows the TurboJPEG shared library to
 be used directly from Java applications.  See java/README for more details.
 If you are using CMake 2.8, you can set the Java_JAVAC_EXECUTABLE,
@@ -642,11 +671,11 @@ instead.  You can also set the JAVACFLAGS CMake variable to specify arguments
 that should be passed to the Java compiler when building the front-end classes.
-========================
+==================
-Installing libmozjpeg
+Installing mozjpeg
-========================
+==================
-You can use the build system to install libmozjpeg into a directory of your
+You can use the build system to install mozjpeg into a directory of your
 choosing (as opposed to creating an installer.)  To do this, add:
  -DCMAKE_INSTALL_PREFIX={install_directory}
@@ -656,12 +685,12 @@ to the cmake command line.
 For example,
  cmake -G "NMake Makefiles" -DCMAKE_BUILD_TYPE=Release \
-    -DCMAKE_INSTALL_PREFIX=c:\libmozjpeg {source_directory}
+    -DCMAKE_INSTALL_PREFIX=c:\mozjpeg {source_directory}
  nmake install
-will install the header files in c:\libmozjpeg\include, the library files
+will install the header files in c:\mozjpeg\include, the library files
-in c:\libmozjpeg\lib, the DLL's in c:\libmozjpeg\bin, and the
+in c:\mozjpeg\lib, the DLL's in c:\mozjpeg\bin, and the
-documentation in c:\libmozjpeg\doc.
+documentation in c:\mozjpeg\doc.
 =============
@@ -679,7 +708,7 @@ Build Recipes
    {source_directory}
  make
-This produces a 64-bit build of libmozjpeg that does not depend on
+This produces a 64-bit build of mozjpeg that does not depend on
 cygwin1.dll or other Cygwin DLL's.  The mingw64-x86_64-gcc-core and
 mingw64-x86_64-gcc-g++ packages (and their dependencies) must be installed.
@@ -694,7 +723,7 @@ mingw64-x86_64-gcc-g++ packages (and their dependencies) must be installed.
    {source_directory}
  make
-This produces a 32-bit build of libmozjpeg that does not depend on
+This produces a 32-bit build of mozjpeg that does not depend on
 cygwin1.dll or other Cygwin DLL's.  The mingw64-i686-gcc-core and
 mingw64-i686-gcc-g++ packages (and their dependencies) must be installed.
@@ -746,33 +775,39 @@ make udmg [BUILDDIR32={32-bit build directory}]
  On 64-bit OS X systems, this creates a Macintosh package and disk image that
  contains universal i386/x86-64 binaries.  You should first configure a 32-bit
-  out-of-tree build of libmozjpeg, then configure a 64-bit out-of-tree
+  out-of-tree build of mozjpeg, then configure a 64-bit out-of-tree
  build, then run 'make udmg' from the 64-bit build directory.  The build
  system will look for the 32-bit build under {source_directory}/osxx86 by
  default, but you can override this by setting the BUILDDIR32 variable on the
  make command line as shown above.
 make iosdmg [BUILDDIR32={32-bit build directory}] \
-  [BUILDDIRARMV6={ARM v6 build directory}] \
+  [BUILDDIRARMV6={ARMv6 build directory}] \
-  [BUILDDIRARMV7={ARM v7 build directory}] \
+  [BUILDDIRARMV7={ARMv7 build directory}] \
-  [BUILDDIRARMV7S={ARM v7s build directory}]
+  [BUILDDIRARMV7S={ARMv7s build directory}] \
  [BUILDDIRARMV8={ARMv8 build directory}]
  On OS X systems, this creates a Macintosh package and disk image in which the
-  libmozjpeg static libraries contain ARM architectures necessary to build
+  mozjpeg static libraries contain ARM architectures necessary to build
  iOS applications.  If building on an x86-64 system, the binaries will also
  contain the i386 architecture, as with 'make udmg' above.  You should first
-  configure ARM v6, ARM v7, and/or ARM v7s out-of-tree builds of libmozjpeg
+  configure ARMv6, ARMv7, ARMv7s, and/or ARMv8 out-of-tree builds of
-  (see "Building libmozjpeg for iOS" above.)  If you are building an x86-64
+  mozjpeg (see "Building mozjpeg for iOS" above.)  If you are
-  version of libmozjpeg, you should configure a 32-bit out-of-tree build as
+  building an x86-64 version of mozjpeg, you should configure a 32-bit
-  well.  Next, build libmozjpeg as you would normally, using an out-of-tree
+  out-of-tree build as well.  Next, build mozjpeg as you would normally,
-  build.  When it is built, run 'make iosdmg' from the build directory.  The
+  using an out-of-tree build.  When it is built, run 'make iosdmg' from the
-  build system will look for the ARM v6 build under {source_directory}/iosarmv6
+  build directory.  The build system will look for the ARMv6 build under
-  by default, the ARM v7 build under {source_directory}/iosarmv7 by default,
+  {source_directory}/iosarmv6 by default, the ARMv7 build under
-  the ARM v7s build under {source_directory}/iosarmv7s by default, and (if
+  {source_directory}/iosarmv7 by default, the ARMv7s build under
-  applicable) the 32-bit build under {source_directory}/osxx86 by default, but
+  {source_directory}/iosarmv7s by default, the ARMv8 build under
-  you can override this by setting the BUILDDIR32, BUILDDIRARMV6,
+  {source_directory}/iosarmv8 by default, and (if applicable) the 32-bit build
-  BUILDDIRARMV7, and/or BUILDDIRARMV7S variables on the make command line as
+  under {source_directory}/osxx86 by default, but you can override this by
-  shown above.
+  setting the BUILDDIR32, BUILDDIRARMV6, BUILDDIRARMV7, BUILDDIRARMV7S, and/or
  BUILDDIRARMV8 variables on the make command line as shown above.
  NOTE: If including an ARMv8 build in the package, then you may need to use
  Xcode's version of lipo instead of the operating system's.  To do this, pass
  an argument of LIPO="xcrun lipo" on the make command line.
 make cygwinpkg
@@ -794,7 +829,7 @@ If using MinGW:
 If using the Visual Studio IDE, build the "installer" project.
-The installer package (libmozjpeg[-gcc][64].exe) will be located under
+The installer package (mozjpeg[-gcc][64].exe) will be located under
 {build_directory}.  If building using the Visual Studio IDE, then the installer
 package will be located in a subdirectory with the same name as the
 configuration you built (such as {build_directory}\Debug\ or
@@ -808,24 +843,24 @@ Building a Windows installer requires the Nullsoft Install System
 **     Regression testing
 *******************************************************************************
-The most common way to test libmozjpeg is by invoking 'make test' on
+The most common way to test mozjpeg is by invoking 'make test' on
 Unix/Linux platforms or 'ctest' on Windows platforms, once the build has
 completed.  This runs a series of tests to ensure that mathematical
-compatibility has been maintained between libmozjpeg and libjpeg v6b.  This
+compatibility has been maintained.  This also invokes the TurboJPEG unit tests,
-also invokes the MozJPEG unit tests, which ensure that the colorspace
+which ensure that the colorspace extensions, YUV encoding, decompression
-extensions, YUV encoding, decompression scaling, and other features of the
+scaling, and other features of the TurboJPEG C and Java APIs are working
-MozJPEG C and Java APIs are working properly (and, by extension, that the
+properly (and, by extension, that the equivalent features of the underlying
-equivalent features of the underlying libjpeg API are also working.)
+libjpeg API are also working.)
 Invoking 'make testclean' or 'nmake testclean' (if using NMake) or building
 the 'testclean' target (if using the Visual Studio IDE) will clean up the
 output images generated by 'make test'.
-On Unix/Linux platforms, more extensive tests of the MozJPEG C and Java
+On Unix/Linux platforms, more extensive tests of the TurboJPEG C and Java
-wrappers can be run by invoking 'make tjtest'.  These extended MozJPEG tests
+wrappers can be run by invoking 'make tjtest'.  These extended TurboJPEG tests
-essentially iterate through all of the available features of the MozJPEG APIs
+essentially iterate through all of the available features of the TurboJPEG APIs
-that are not covered by the MozJPEG unit tests (this includes the lossless
+that are not covered by the TurboJPEG unit tests (this includes the lossless
 transform options) and compare the images generated by each feature to images
 generated using the equivalent feature in the libjpeg API.  The extended
-MozJPEG tests are meant to test for regressions in the MozJPEG wrappers,
+TurboJPEG tests are meant to test for regressions in the TurboJPEG wrappers,
 not in the underlying libjpeg API library.
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -8,8 +8,8 @@ if(POLICY CMP0022)
  cmake_policy(SET CMP0022 OLD)
 endif()
-project(libmozjpeg C)
+project(mozjpeg C)
-set(VERSION 2.1)
+set(VERSION 3.1)
 if(CYGWIN OR NOT CMAKE_HOST_SYSTEM_NAME STREQUAL "Windows")
  execute_process(COMMAND "date" "+%Y%m%d" OUTPUT_VARIABLE BUILD)
@@ -47,8 +47,8 @@ message(STATUS "VERSION = ${VERSION}, BUILD = ${BUILD}")
 option(WITH_SIMD "Include SIMD extensions" TRUE)
 option(WITH_ARITH_ENC "Include arithmetic encoding support" TRUE)
 option(WITH_ARITH_DEC "Include arithmetic decoding support" TRUE)
-option(WITH_JPEG7 "Emulate libjpeg v7 API/ABI (this makes libmozjpeg backward incompatible with libjpeg v6b)" FALSE)
+option(WITH_JPEG7 "Emulate libjpeg v7 API/ABI (this makes mozjpeg backward incompatible with libjpeg v6b)" FALSE)
-option(WITH_JPEG8 "Emulate libjpeg v8 API/ABI (this makes libmozjpeg backward incompatible with libjpeg v6b)" FALSE)
+option(WITH_JPEG8 "Emulate libjpeg v8 API/ABI (this makes mozjpeg backward incompatible with libjpeg v6b)" FALSE)
 option(WITH_MEM_SRCDST "Include in-memory source/destination manager functions when emulating the libjpeg v6b or v7 API/ABI" TRUE)
 option(WITH_TURBOJPEG "Include the TurboJPEG wrapper library and associated test programs" TRUE)
 option(WITH_JAVA "Build Java wrapper for the TurboJPEG library" FALSE)
@@ -155,7 +155,7 @@ if(CMAKE_INSTALL_PREFIX_INITIALIZED_TO_DEFAULT)
    set(CMAKE_INSTALL_PREFIX_DEFAULT ${CMAKE_INSTALL_PREFIX_DEFAULT}64)
  endif()
  set(CMAKE_INSTALL_PREFIX "c:/${CMAKE_INSTALL_PREFIX_DEFAULT}" CACHE PATH
-    "Directory into which to install libmozjpeg (default: c:/${CMAKE_INSTALL_PREFIX_DEFAULT})"
+    "Directory into which to install mozjpeg (default: c:/${CMAKE_INSTALL_PREFIX_DEFAULT})"
    FORCE)
 endif()
@@ -179,8 +179,8 @@ endif()
 # Targets
 #
-set(JPEG_SOURCES jcapimin.c jcapistd.c jccoefct.c jccolor.c jcdctmgr.c jchuff.c
+set(JPEG_SOURCES jcapimin.c jcapistd.c jccoefct.c jccolor.c jcdctmgr.c jcext.c
-  jccompat.c jcinit.c jcmainct.c jcmarker.c jcmaster.c jcomapi.c jcparam.c
+  jchuff.c jcinit.c jcmainct.c jcmarker.c jcmaster.c jcomapi.c jcparam.c
  jcphuff.c jcprepct.c jcsample.c jctrans.c jdapimin.c jdapistd.c jdatadst.c
  jdatasrc.c jdcoefct.c jdcolor.c jddctmgr.c jdhuff.c jdinput.c jdmainct.c
  jdmarker.c jdmaster.c jdmerge.c jdphuff.c jdpostct.c jdsample.c jdtrans.c
@@ -438,7 +438,7 @@ foreach(libtype shared static)
  # CC: null  SAMP: fullsize  FDCT: islow  ENT: huff
  add_test(cjpeg${suffix}-rgb-islow
-    ${dir}cjpeg${suffix} -rgb -dct int -outfile testout_rgb_islow.jpg
+    ${dir}cjpeg${suffix} -revert -rgb -dct int -outfile testout_rgb_islow.jpg
      ${CMAKE_SOURCE_DIR}/testimages/testorig.ppm)
  add_test(cjpeg${suffix}-rgb-islow-cmp
    ${CMAKE_COMMAND} -DMD5=${MD5_JPEG_RGB_ISLOW} -DFILE=testout_rgb_islow.jpg
@@ -471,7 +471,7 @@ foreach(libtype shared static)
  # CC: RGB->YCC  SAMP: fullsize/h2v1  FDCT: ifast  ENT: 2-pass huff
  add_test(cjpeg${suffix}-422-ifast-opt
-    ${dir}cjpeg${suffix} -sample 2x1 -dct fast -opt
+    ${dir}cjpeg${suffix} -revert -sample 2x1 -dct fast -opt
      -outfile testout_422_ifast_opt.jpg
      ${CMAKE_SOURCE_DIR}/testimages/testorig.ppm)
  add_test(cjpeg${suffix}-422-ifast-opt-cmp
@@ -513,7 +513,7 @@ foreach(libtype shared static)
  # CC: RGB->YCC  SAMP: fullsize/h2v2  FDCT: ifast  ENT: prog huff
  add_test(cjpeg${suffix}-420-q100-ifast-prog
-    ${dir}cjpeg${suffix} -sample 2x2 -quality 100 -dct fast -prog
+    ${dir}cjpeg${suffix} -revert -sample 2x2 -quality 100 -dct fast -prog
      -outfile testout_420_q100_ifast_prog.jpg
      ${CMAKE_SOURCE_DIR}/testimages/testorig.ppm)
  add_test(cjpeg${suffix}-420-q100-ifast-prog-cmp
@@ -539,7 +539,7 @@ foreach(libtype shared static)
  # CC: RGB->Gray  SAMP: fullsize  FDCT: islow  ENT: huff
  add_test(cjpeg${suffix}-gray-islow
-    ${dir}cjpeg${suffix} -gray -dct int -outfile testout_gray_islow.jpg
+    ${dir}cjpeg${suffix} -revert -gray -dct int -outfile testout_gray_islow.jpg
      ${CMAKE_SOURCE_DIR}/testimages/testorig.ppm)
  add_test(cjpeg${suffix}-gray-islow-cmp
    ${CMAKE_COMMAND} -DMD5=${MD5_JPEG_GRAY_ISLOW}
@@ -557,7 +557,7 @@ foreach(libtype shared static)
  add_test(djpeg${suffix}-gray-islow-rgb
    ${dir}djpeg${suffix} -dct int -rgb -outfile testout_gray_islow_rgb.ppm
      testout_gray_islow.jpg)
-  add_test(cjpeg${suffix}-gray-islow-rgb-cmp
+  add_test(djpeg${suffix}-gray-islow-rgb-cmp
    ${CMAKE_COMMAND} -DMD5=${MD5_PPM_GRAY_ISLOW_RGB}
      -DFILE=testout_gray_islow_rgb.ppm
      -P ${CMAKE_SOURCE_DIR}/cmakescripts/md5cmp.cmake)
@@ -583,7 +583,7 @@ foreach(libtype shared static)
  # CC: RGB->YCC  SAMP: fullsize smooth/h2v2 smooth  FDCT: islow
  # ENT: 2-pass huff
  add_test(cjpeg${suffix}-420s-ifast-opt
-    ${dir}cjpeg${suffix} -sample 2x2 -smooth 1 -dct int -opt -outfile
+    ${dir}cjpeg${suffix} -revert -sample 2x2 -smooth 1 -dct int -opt -outfile
      testout_420s_ifast_opt.jpg ${CMAKE_SOURCE_DIR}/testimages/testorig.ppm)
  add_test(cjpeg${suffix}-420s-ifast-opt-cmp
    ${CMAKE_COMMAND} -DMD5=${MD5_JPEG_420S_IFAST_OPT}
@@ -592,7 +592,7 @@ foreach(libtype shared static)
  # CC: RGB->YCC  SAMP: fullsize/int  FDCT: float  ENT: prog huff
  add_test(cjpeg${suffix}-3x2-float-prog
-    ${dir}cjpeg${suffix} -sample 3x2 -dct float -prog
+    ${dir}cjpeg${suffix} -revert -sample 3x2 -dct float -prog
      -outfile testout_3x2_float_prog.jpg
      ${CMAKE_SOURCE_DIR}/testimages/testorig.ppm)
  add_test(cjpeg${suffix}-3x2-float-prog-cmp
@@ -610,7 +610,7 @@ foreach(libtype shared static)
  if(WITH_ARITH_ENC)
    # CC: YCC->RGB  SAMP: fullsize/h2v2  FDCT: islow  ENT: arith
    add_test(cjpeg${suffix}-420-islow-ari
-      ${dir}cjpeg${suffix} -dct int -arithmetic
+      ${dir}cjpeg${suffix} -revert -dct int -arithmetic
        -outfile testout_420_islow_ari.jpg
        ${CMAKE_SOURCE_DIR}/testimages/testorig.ppm)
    add_test(cjpeg${suffix}-420-islow-ari-cmp
@@ -618,7 +618,7 @@ foreach(libtype shared static)
        -DFILE=testout_420_islow_ari.jpg
        -P ${CMAKE_SOURCE_DIR}/cmakescripts/md5cmp.cmake)
    add_test(jpegtran${suffix}-420-islow-ari
-      ${dir}jpegtran${suffix} -arithmetic
+      ${dir}jpegtran${suffix} -revert -arithmetic
        -outfile testout_420_islow_ari.jpg
        ${CMAKE_SOURCE_DIR}/testimages/testimgint.jpg)
    add_test(jpegtran${suffix}-420-islow-ari-cmp
@@ -627,7 +627,7 @@ foreach(libtype shared static)
        -P ${CMAKE_SOURCE_DIR}/cmakescripts/md5cmp.cmake)
    # CC: YCC->RGB  SAMP: fullsize  FDCT: islow  ENT: prog arith
    add_test(cjpeg${suffix}-444-islow-progari
-      ${dir}cjpeg${suffix} -sample 1x1 -dct int -progressive -arithmetic
+      ${dir}cjpeg${suffix} -revert -sample 1x1 -dct int -progressive -arithmetic
        -outfile testout_444_islow_progari.jpg
        ${CMAKE_SOURCE_DIR}/testimages/testorig.ppm)
    add_test(cjpeg${suffix}-444-islow-progari-cmp
@@ -637,15 +637,15 @@ foreach(libtype shared static)
  endif()
  if(WITH_ARITH_DEC)
    # CC: RGB->YCC  SAMP: h2v2 merged  IDCT: ifast  ENT: arith
-    add_test(cjpeg${suffix}-420m-ifast-ari
+    add_test(djpeg${suffix}-420m-ifast-ari
      ${dir}djpeg${suffix} -fast -ppm -outfile testout_420m_ifast_ari.ppm
        ${CMAKE_SOURCE_DIR}/testimages/testimgari.jpg)
-    add_test(cjpeg${suffix}-420m-ifast-ari-cmp
+    add_test(djpeg${suffix}-420m-ifast-ari-cmp
      ${CMAKE_COMMAND} -DMD5=${MD5_PPM_420M_IFAST_ARI}
        -DFILE=testout_420m_ifast_ari.ppm
        -P ${CMAKE_SOURCE_DIR}/cmakescripts/md5cmp.cmake)
    add_test(jpegtran${suffix}-420-islow
-      ${dir}jpegtran${suffix} -outfile testout_420_islow.jpg
+      ${dir}jpegtran${suffix} -revert -outfile testout_420_islow.jpg
        ${CMAKE_SOURCE_DIR}/testimages/testimgari.jpg)
    add_test(jpegtran${suffix}-420-islow-cmp
      ${CMAKE_COMMAND} -DMD5=${MD5_JPEG_420_ISLOW}
@@ -732,7 +732,7 @@ foreach(libtype shared static)
        -P ${CMAKE_SOURCE_DIR}/cmakescripts/md5cmp.cmake)
  endif()
  add_test(jpegtran${suffix}-crop
-    ${dir}jpegtran${suffix} -crop 120x90+20+50 -transpose -perfect
+    ${dir}jpegtran${suffix} -revert -crop 120x90+20+50 -transpose -perfect
      -outfile testout_crop.jpg ${CMAKE_SOURCE_DIR}/testimages/${TESTORIG})
  add_test(jpegtran${suffix}-crop-cmp
    ${CMAKE_COMMAND} -DMD5=${MD5_JPEG_CROP} -DFILE=testout_crop.jpg
@@ -778,16 +778,16 @@ endif()
 STRING(REGEX REPLACE "/" "\\\\" INST_DIR "${CMAKE_INSTALL_PREFIX}")
-configure_file(release/libmozjpeg.nsi.in libmozjpeg.nsi @ONLY)
+configure_file(release/mozjpeg.nsi.in mozjpeg.nsi @ONLY)
 if(WITH_JAVA)
  set(JAVA_DEPEND java)
 endif()
 add_custom_target(installer
-  makensis -nocd ${INST_DEFS} libmozjpeg.nsi
+  makensis -nocd ${INST_DEFS} mozjpeg.nsi
  DEPENDS jpeg jpeg-static turbojpeg turbojpeg-static rdjpgcom wrjpgcom
    cjpeg djpeg jpegtran tjbench ${JAVA_DEPEND}
-  SOURCES libmozjpeg.nsi)
+  SOURCES mozjpeg.nsi)
 if(WITH_TURBOJPEG)
  set(TURBOJPEG_TARGETS turbojpeg turbojpeg-static tjbench)
--- a/ChangeLog-mozilla.txt
+++ b/ChangeLog-mozilla.txt
@@ -1,4 +0,0 @@
 Version 1.0 (March 4, 2014)
 ===========================
 * Add 'jpgcrush' functionality. Figures out which progressive coding configuration uses the fewest bits.
--- a/ChangeLog.txt
+++ b/ChangeLog.txt
@@ -13,6 +13,60 @@ instead of -1 if componentID was > 0 and subsamp was TJSAMP_GRAY.
 [3] Fixed an issue in tjBufSizeYUV2() wherby it would erroneously return 0
 instead of -1 if width was < 1.
 [5] The Huffman encoder now uses clz and bsr instructions for bit counting on
 ARM64 platforms (see 1.4 beta1 [5].)
 [6] The close() method in the TJCompressor and TJDecompressor Java classes is
 now idempotent.  Previously, that method would call the native tjDestroy()
 function even if the TurboJPEG instance had already been destroyed.  This
 caused an exception to be thrown during finalization, if the close() method had
 already been called.  The exception was caught, but it was still an expensive
 operation.
 [7] The TurboJPEG API previously generated an error ("Could not determine
 subsampling type for JPEG image") when attempting to decompress grayscale JPEG
 images that were compressed with a sampling factor other than 1 (for instance,
 with 'cjpeg -grayscale -sample 2x2').  Subsampling technically has no meaning
 with grayscale JPEGs, and thus the horizontal and vertical sampling factors
 for such images are ignored by the decompressor.  However, the TurboJPEG API
 was being too rigid and was expecting the sampling factors to be equal to 1
 before it treated the image as a grayscale JPEG.
 [8] cjpeg, djpeg, and jpegtran now accept an argument of -version, which will
 print the library version and exit.
 [9] Referring to 1.4 beta1 [15], another extremely rare circumstance was
 discovered under which the Huffman encoder's local buffer can be overrun
 when a buffered destination manager is being used and an
 extremely-high-frequency block (basically junk image data) is being encoded.
 Even though the Huffman local buffer was increased from 128 bytes to 136 bytes
 to address the previous issue, the new issue caused even the larger buffer to
 be overrun.  Further analysis reveals that, in the absolute worst case (such as
 setting alternating AC coefficients to 32767 and -32768 in the JPEG scanning
 order), the Huffman encoder can produce encoded blocks that approach double the
 size of the unencoded blocks.  Thus, the Huffman local buffer was increased to
 256 bytes, which should prevent any such issue from re-occurring in the future.
 [10] The new tjPlaneSizeYUV(), tjPlaneWidth(), and tjPlaneHeight() functions
 were not actually usable on any platform except OS X and Windows, because
 those functions were not included in the libturbojpeg mapfile.  This has been
 fixed.
 [11] Restored the JPP(), JMETHOD(), and FAR macros in the libjpeg-turbo header
 files.  The JPP() and JMETHOD() macros were originally implemented in libjpeg
 as a way of supporting non-ANSI compilers that lacked support for prototype
 parameters.  libjpeg-turbo has never supported such compilers, but some
 software packages still use the macros to define their own prototypes.
 Similarly, libjpeg-turbo has never supported MS-DOS and other platforms that
 have far symbols, but some software packages still use the FAR macro.  A pretty
 good argument can be made that this is a bad practice on the part of the
 software in question, but since this affects more than one package, it's just
 easier to fix it here.
 [12] Fixed issues that were preventing the ARM 64-bit SIMD code from compiling
 for iOS, and included an ARMv8 architecture in all of the binaries installed by
 the "official" libjpeg-turbo SDK for OS X.
 1.3.90 (1.4 beta1)
 ==================
@@ -132,14 +186,6 @@ compressed JPEG image.  If the size of the previous JPEG image was not as large
 as the current JPEG image, then TurboJPEG would unnecessarily reallocate the
 destination buffer.
 [7] Improved the accuracy and performance of the non-SIMD implementation of the
 floating point inverse DCT (using code borrowed from libjpeg v8a and later.)
 The accuracy of this implementation now matches the accuracy of the SSE/SSE2
 implementation.  Note, however, that the floating point DCT/IDCT algorithms are
 mainly a legacy feature.  They generally do not produce significantly better
 accuracy than the slow integer DCT/IDCT algorithms, and they are quite a bit
 slower.
 1.3.1
 =====
@@ -273,7 +319,7 @@ configure/CMake switch in order to retain strict API/ABI compatibility with the
 libjpeg v6b or v7 API/ABI (or with previous versions of libjpeg-turbo.)  See
 README-turbo.txt for more details.
-[13] Added ARM v7s architecture to libjpeg.a and libturbojpeg.a in the official
+[13] Added ARMv7s architecture to libjpeg.a and libturbojpeg.a in the official
 libjpeg-turbo binary package for OS X, so that those libraries can be used to
 build applications that leverage the faster CPUs in the iPhone 5 and iPad 4.
@@ -356,7 +402,7 @@ K component is assigned a component ID of 1 instead of 4.  Although these files
 are in violation of the spec, other JPEG implementations handle them
 correctly.
-[7] Added ARM v6 and ARM v7 architectures to libjpeg.a and libturbojpeg.a in
+[7] Added ARMv6 and ARMv7 architectures to libjpeg.a and libturbojpeg.a in
 the official libjpeg-turbo binary package for OS X, so that those libraries can
 be used to build both OS X and iOS applications.
--- a/LICENSE.txt
+++ b/LICENSE.txt
@@ -2,6 +2,6 @@ Redistribution and use in source and binary forms, with or without modification,
 * 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 the libmozjpeg Project nor the names of its contributors may be used to endorse or promote products derived from this software without specific prior written permission.
+* Neither the name of the mozjpeg Project 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 HOLDERS 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.
--- a/Makefile.am
+++ b/Makefile.am
@@ -11,12 +11,12 @@ endif
 nodist_include_HEADERS = jconfig.h
-HDRS = jchuff.h jdct.h jdhuff.h jerror.h jinclude.h jmemsys.h jmorecfg.h \
+HDRS = jchuff.h jcmaster.h jdct.h jdhuff.h jerror.h jinclude.h jmemsys.h \
-	jpegint.h jpeglib.h jversion.h jsimd.h jsimddct.h jpegcomp.h \
+	jmorecfg.h jpegint.h jpeglib.h jversion.h jsimd.h jsimddct.h jpegcomp.h \
 	jpeg_nbits_table.h
 libjpeg_la_SOURCES = $(HDRS) jcapimin.c jcapistd.c jccoefct.c jccolor.c \
-	jccompat.c jcdctmgr.c jchuff.c jcinit.c jcmainct.c jcmarker.c \
+	jcdctmgr.c jcext.c jchuff.c jcinit.c jcmainct.c jcmarker.c \
 	jcmaster.c jcomapi.c jcparam.c jcphuff.c jcprepct.c jcsample.c \
 	jctrans.c jdapimin.c jdapistd.c jdatadst.c jdatasrc.c jdcoefct.c \
 	jdcolor.c jddctmgr.c jdhuff.c jdinput.c jdmainct.c jdmarker.c \
@@ -182,7 +182,8 @@ dist_example_DATA = example.c
 EXTRA_DIST = win release $(DOCS) testimages CMakeLists.txt \
 	sharedlib/CMakeLists.txt cmakescripts libjpeg.map.in doc doxygen.config \
-	doxygen-extra.css jccolext.c jdcolext.c jdcol565.c jdmrgext.c jstdhuff.c
+	doxygen-extra.css jccolext.c jdcolext.c jdcol565.c jdmrgext.c jdmrg565.c \
 	jstdhuff.c
 dist-hook:
 	rm -rf `find $(distdir) -name .svn`
@@ -275,7 +276,10 @@ MD5_JPEG_CROP = b4197f377e621c4e9b1d20471432610d
 endif
-test: testclean all
+.PHONY: test
 test: tjquicktest bittest
 tjquicktest: testclean all
 if WITH_TURBOJPEG
 if WITH_JAVA
@@ -293,12 +297,14 @@ endif
 	./tjunittest -yuv -noyuvpad
 endif
 bittest: testclean all
 # These tests are carefully crafted to provide full coverage of as many of the
 # underlying algorithms as possible (including all of the SIMD-accelerated
 # ones.)
 # CC: null  SAMP: fullsize  FDCT: islow  ENT: huff
-	./cjpeg -rgb -dct int -outfile testout_rgb_islow.jpg $(srcdir)/testimages/testorig.ppm
+	./cjpeg -revert -rgb -dct int -outfile testout_rgb_islow.jpg $(srcdir)/testimages/testorig.ppm
 	md5/md5cmp $(MD5_JPEG_RGB_ISLOW) testout_rgb_islow.jpg
 # CC: null  SAMP: fullsize  IDCT: islow  ENT: huff
 	./djpeg -dct int -ppm -outfile testout_rgb_islow.ppm testout_rgb_islow.jpg
@@ -318,7 +324,7 @@ else
 endif
 # CC: RGB->YCC  SAMP: fullsize/h2v1  FDCT: ifast  ENT: 2-pass huff
-	./cjpeg -sample 2x1 -dct fast -opt -outfile testout_422_ifast_opt.jpg $(srcdir)/testimages/testorig.ppm
+	./cjpeg -revert -sample 2x1 -dct fast -opt -outfile testout_422_ifast_opt.jpg $(srcdir)/testimages/testorig.ppm
 	md5/md5cmp $(MD5_JPEG_422_IFAST_OPT) testout_422_ifast_opt.jpg
 # CC: YCC->RGB  SAMP: fullsize/h2v1 fancy  IDCT: ifast  ENT: huff
 	./djpeg -dct fast -outfile testout_422_ifast.ppm testout_422_ifast_opt.jpg
@@ -342,7 +348,7 @@ else
 endif
 # CC: RGB->YCC  SAMP: fullsize/h2v2  FDCT: ifast  ENT: prog huff
-	./cjpeg -sample 2x2 -quality 100 -dct fast -prog -outfile testout_420_q100_ifast_prog.jpg $(srcdir)/testimages/testorig.ppm
+	./cjpeg -revert -sample 2x2 -quality 100 -dct fast -prog -outfile testout_420_q100_ifast_prog.jpg $(srcdir)/testimages/testorig.ppm
 	md5/md5cmp $(MD5_JPEG_420_IFAST_Q100_PROG) testout_420_q100_ifast_prog.jpg
 # CC: YCC->RGB  SAMP: fullsize/h2v2 fancy  IDCT: ifast  ENT: prog huff
 	./djpeg -dct fast -outfile testout_420_q100_ifast.ppm testout_420_q100_ifast_prog.jpg
@@ -354,7 +360,7 @@ endif
 	rm testout_420m_q100_ifast.ppm testout_420_q100_ifast_prog.jpg
 # CC: RGB->Gray  SAMP: fullsize  FDCT: islow  ENT: huff
-	./cjpeg -gray -dct int -outfile testout_gray_islow.jpg $(srcdir)/testimages/testorig.ppm
+	./cjpeg -revert -gray -dct int -outfile testout_gray_islow.jpg $(srcdir)/testimages/testorig.ppm
 	md5/md5cmp $(MD5_JPEG_GRAY_ISLOW) testout_gray_islow.jpg
 # CC: Gray->Gray  SAMP: fullsize  IDCT: islow  ENT: huff
 	./djpeg -dct int -outfile testout_gray_islow.ppm testout_gray_islow.jpg
@@ -379,12 +385,12 @@ endif
 # CC: RGB->YCC  SAMP: fullsize smooth/h2v2 smooth  FDCT: islow
 # ENT: 2-pass huff
-	./cjpeg -sample 2x2 -smooth 1 -dct int -opt -outfile testout_420s_ifast_opt.jpg $(srcdir)/testimages/testorig.ppm
+	./cjpeg -revert -sample 2x2 -smooth 1 -dct int -opt -outfile testout_420s_ifast_opt.jpg $(srcdir)/testimages/testorig.ppm
 	md5/md5cmp $(MD5_JPEG_420S_IFAST_OPT) testout_420s_ifast_opt.jpg
 	rm testout_420s_ifast_opt.jpg
 # CC: RGB->YCC  SAMP: fullsize/int  FDCT: float  ENT: prog huff
-	./cjpeg -sample 3x2 -dct float -prog -outfile testout_3x2_float_prog.jpg $(srcdir)/testimages/testorig.ppm
+	./cjpeg -revert -sample 3x2 -dct float -prog -outfile testout_3x2_float_prog.jpg $(srcdir)/testimages/testorig.ppm
 	md5/md5cmp $(MD5_JPEG_3x2_FLOAT_PROG) testout_3x2_float_prog.jpg
 # CC: YCC->RGB  SAMP: fullsize/int  IDCT: float  ENT: prog huff
 	./djpeg -dct float -outfile testout_3x2_float.ppm testout_3x2_float_prog.jpg
@@ -393,14 +399,14 @@ endif
 if WITH_ARITH_ENC
 # CC: YCC->RGB  SAMP: fullsize/h2v2  FDCT: islow  ENT: arith
-	./cjpeg -dct int -arithmetic -outfile testout_420_islow_ari.jpg $(srcdir)/testimages/testorig.ppm
+	./cjpeg -revert -dct int -arithmetic -outfile testout_420_islow_ari.jpg $(srcdir)/testimages/testorig.ppm
 	md5/md5cmp $(MD5_JPEG_420_ISLOW_ARI) testout_420_islow_ari.jpg
 	rm testout_420_islow_ari.jpg
-	./jpegtran -arithmetic -outfile testout_420_islow_ari.jpg $(srcdir)/testimages/testimgint.jpg
+	./jpegtran -revert -arithmetic -outfile testout_420_islow_ari.jpg $(srcdir)/testimages/testimgint.jpg
 	md5/md5cmp $(MD5_JPEG_420_ISLOW_ARI) testout_420_islow_ari.jpg
 	rm testout_420_islow_ari.jpg
 # CC: YCC->RGB  SAMP: fullsize  FDCT: islow  ENT: prog arith
-	./cjpeg -sample 1x1 -dct int -progressive -arithmetic -outfile testout_444_islow_progari.jpg $(srcdir)/testimages/testorig.ppm
+	./cjpeg -revert -sample 1x1 -dct int -progressive -arithmetic -outfile testout_444_islow_progari.jpg $(srcdir)/testimages/testorig.ppm
 	md5/md5cmp $(MD5_JPEG_444_ISLOW_PROGARI) testout_444_islow_progari.jpg
 	rm testout_444_islow_progari.jpg
 endif
@@ -409,7 +415,7 @@ if WITH_ARITH_DEC
 	./djpeg -fast -ppm -outfile testout_420m_ifast_ari.ppm $(srcdir)/testimages/testimgari.jpg
 	md5/md5cmp $(MD5_PPM_420M_IFAST_ARI) testout_420m_ifast_ari.ppm
 	rm testout_420m_ifast_ari.ppm
-	./jpegtran -outfile testout_420_islow.jpg $(srcdir)/testimages/testimgari.jpg
+	./jpegtran -revert -outfile testout_420_islow.jpg $(srcdir)/testimages/testimgari.jpg
 	md5/md5cmp $(MD5_JPEG_420_ISLOW) testout_420_islow.jpg
 	rm testout_420_islow.jpg
 endif
@@ -486,7 +492,7 @@ else
 	rm testout_420m_islow_565D.bmp
 endif
-	./jpegtran -crop 120x90+20+50 -transpose -perfect -outfile testout_crop.jpg $(srcdir)/testimages/$(TESTORIG)
+	./jpegtran -revert -crop 120x90+20+50 -transpose -perfect -outfile testout_crop.jpg $(srcdir)/testimages/$(TESTORIG)
 	md5/md5cmp $(MD5_JPEG_CROP) testout_crop.jpg
 	rm testout_crop.jpg
@@ -536,26 +542,26 @@ if WITH_JAVA
 endif
-pkgscripts/libmozjpeg.spec: pkgscripts/libmozjpeg.spec.tmpl
+pkgscripts/mozjpeg.spec: pkgscripts/mozjpeg.spec.tmpl
-	cat pkgscripts/libmozjpeg.spec.tmpl | sed s@%{__prefix}@$(prefix)@g | \
+	cat pkgscripts/mozjpeg.spec.tmpl | sed s@%{__prefix}@$(prefix)@g | \
 		sed s@%{__bindir}@$(bindir)@g | sed s@%{__datadir}@$(datadir)@g | \
 		sed s@%{__docdir}@$(docdir)@g | sed s@%{__includedir}@$(includedir)@g | \
 		sed s@%{__libdir}@$(libdir)@g | sed s@%{__mandir}@$(mandir)@g \
-		> pkgscripts/libmozjpeg.spec
+		> pkgscripts/mozjpeg.spec
-rpm: all pkgscripts/libmozjpeg.spec
+rpm: all pkgscripts/mozjpeg.spec
 	TMPDIR=`mktemp -d /tmp/${PACKAGE_NAME}-build.XXXXXX`; \
 	mkdir -p $$TMPDIR/RPMS; \
 	ln -fs `pwd` $$TMPDIR/BUILD; \
 	rm -f ${PKGNAME}-${VERSION}.${RPMARCH}.rpm; \
 	rpmbuild -bb --define "_blddir $$TMPDIR/buildroot"  \
 		--define "_topdir $$TMPDIR" \
-		--target ${RPMARCH} pkgscripts/libmozjpeg.spec; \
+		--target ${RPMARCH} pkgscripts/mozjpeg.spec; \
 	cp $$TMPDIR/RPMS/${RPMARCH}/${PKGNAME}-${VERSION}-${BUILD}.${RPMARCH}.rpm \
 		${PKGNAME}-${VERSION}.${RPMARCH}.rpm; \
 	rm -rf $$TMPDIR
-srpm: dist-gzip pkgscripts/libmozjpeg.spec
+srpm: dist-gzip pkgscripts/mozjpeg.spec
 	TMPDIR=`mktemp -d /tmp/${PACKAGE_NAME}-build.XXXXXX`; \
 	mkdir -p $$TMPDIR/RPMS; \
 	mkdir -p $$TMPDIR/SRPMS; \
@@ -564,10 +570,10 @@ srpm: dist-gzip pkgscripts/libmozjpeg.spec
 	mkdir -p $$TMPDIR/SPECS; \
 	rm -f ${PKGNAME}-${VERSION}.src.rpm; \
 	cp ${PACKAGE_NAME}-${VERSION}.tar.gz $$TMPDIR/SOURCES; \
-	cat pkgscripts/libmozjpeg.spec | sed s/%{_blddir}/%{_tmppath}/g \
+	cat pkgscripts/mozjpeg.spec | sed s/%{_blddir}/%{_tmppath}/g \
 		| sed s/#--\>//g \
-		> $$TMPDIR/SPECS/libmozjpeg.spec; \
+		> $$TMPDIR/SPECS/mozjpeg.spec; \
-	rpmbuild -bs --define "_topdir $$TMPDIR" $$TMPDIR/SPECS/libmozjpeg.spec; \
+	rpmbuild -bs --define "_topdir $$TMPDIR" $$TMPDIR/SPECS/mozjpeg.spec; \
 	cp $$TMPDIR/SRPMS/${PKGNAME}-${VERSION}-${BUILD}.src.rpm \
 		${PKGNAME}-${VERSION}.src.rpm; \
 	rm -rf $$TMPDIR
@@ -597,12 +603,12 @@ udmg: all pkgscripts/makemacpkg pkgscripts/uninstall
 	sh pkgscripts/makemacpkg -build32 ${BUILDDIR32}
 iosdmg: all pkgscripts/makemacpkg pkgscripts/uninstall
-	sh pkgscripts/makemacpkg -build32 ${BUILDDIR32} -buildarmv6 ${BUILDDIRARMV6} -buildarmv7 ${BUILDDIRARMV7} -buildarmv7s ${BUILDDIRARMV7S}
+	sh pkgscripts/makemacpkg -build32 ${BUILDDIR32} -buildarmv6 ${BUILDDIRARMV6} -buildarmv7 ${BUILDDIRARMV7} -buildarmv7s ${BUILDDIRARMV7S} -buildarmv8 ${BUILDDIRARMV8} -lipo "${LIPO}"
 else
 iosdmg: all pkgscripts/makemacpkg pkgscripts/uninstall
-	sh pkgscripts/makemacpkg -buildarmv6 ${BUILDDIRARMV6} -buildarmv7 ${BUILDDIRARMV7} -buildarmv7s ${BUILDDIRARMV7S}
+	sh pkgscripts/makemacpkg -buildarmv6 ${BUILDDIRARMV6} -buildarmv7 ${BUILDDIRARMV7} -buildarmv7s ${BUILDDIRARMV7S} -buildarmv8 ${BUILDDIRARMV8} -lipo "${LIPO}"
 endif
--- a/7
+++ b/7
@@ -36,7 +36,6 @@ TO DO               Plans for future IJG releases.
 Other documentation files in the distribution are:
 User documentation:
  install.txt       How to configure and install the IJG software.
  usage.txt         Usage instructions for cjpeg, djpeg, jpegtran,
                    rdjpgcom, and wrjpgcom.
  *.1               Unix-style man pages for programs (same info as usage.txt).
@@ -48,9 +47,9 @@ Programmer and internal documentation:
  structure.txt     Overview of the JPEG library's internal structure.
  coderules.txt     Coding style rules --- please read if you contribute code.
-Please read at least the files install.txt and usage.txt.  Some information
+Please read at least usage.txt.  Some information can also be found in the JPEG
-can also be found in the JPEG FAQ (Frequently Asked Questions) article.  See
+FAQ (Frequently Asked Questions) article.  See ARCHIVE LOCATIONS below to find
-ARCHIVE LOCATIONS below to find out where to obtain the FAQ article.
+out where to obtain the FAQ article.
 If you want to understand how the JPEG code works, we suggest reading one or
 more of the REFERENCES, then looking at the documentation files (in roughly
--- a/README-mozilla.txt
+++ b/README-mozilla.txt
@@ -0,0 +1,194 @@
 Mozilla JPEG Encoder Project
 ============================
 mozjpeg is a fork of libjpeg-turbo that aims to speed up load times of web
 pages by reducing the size (and, by extension, the transmission time) of JPEG
 files.  It accomplishes this by enabling optimized Huffman trees and
 progressive entropy coding by default in the JPEG compressor, as well as
 splitting the spectrum of DCT coefficients into separate scans and using
 Trellis quantisation.
 Although it is based on libjpeg-turbo, mozjpeg is not intended to be a
 general-purpose or high-performance JPEG library.  Its performance is highly
 "asymmetric".  That is, the JPEG files it generates require much more time to
 compress than to decompress.  When the default settings are used, mozjpeg is
 considerably slower than libjpeg-turbo or even libjpeg at compressing images.
 Thus, it is not generally suitable for real-time compression.  It is best used
 as part of a web encoding workflow.
 libjpeg API Extensibility Framework
 ===================================
 mozjpeg's implementation of the libjpeg API includes an extensibility framework
 that allows new features to be added without modifying the transparent libjpeg
 compress/decompress structures (which would break backward ABI compatibility.)
 Extension parameters are placed into the opaque jpeg_comp_master structure, and
 a set of accessor functions and globally unique tokens allows for 
 getting/setting those parameters without directly accessing the structure.
 Currently, only the accessor functions necessary to support the mozjpeg
 extensions are implemented, but the framework can be easily extended in the
 future to accommodate additional simple parameter types, complex or
 multi-valued parameters, or decompressor extensions.
 The currently-implemented accessor functions are as follows:
 boolean jpeg_c_bool_param_supported (j_compress_ptr cinfo,
                                     J_BOOLEAN_PARAM param)
        Returns TRUE if the given boolean extension parameter is supported by
        this implementation of the libjpeg API, or FALSE otherwise.
 void jpeg_c_set_bool_param (j_compress_ptr cinfo,
                            J_BOOLEAN_PARAM param, boolean value);
        Set the given boolean extension parameter to the given value (TRUE or
        FALSE.)
 boolean jpeg_c_get_bool_param (j_compress_ptr cinfo, J_BOOLEAN_PARAM param)
        Get the value of the given boolean extension parameter (TRUE or FALSE.)
 boolean jpeg_c_float_param_supported (j_compress_ptr cinfo,
                                      J_FLOAT_PARAM param)
        Returns TRUE if the given floating point extension parameter is
        supported by this implementation of the libjpeg API, or FALSE
        otherwise.
 void jpeg_c_set_float_param (j_compress_ptr cinfo, J_FLOAT_PARAM param,
                             float value)
        Set the given floating point extension parameter to the given value.
 float jpeg_c_get_float_param (j_compress_ptr cinfo, J_FLOAT_PARAM param);
        Get the value of the given floating point extension parameter.
 boolean jpeg_c_int_param_supported (j_compress_ptr cinfo,
                                    J_INT_PARAM param)
        Returns TRUE if the given integer extension parameter is supported by
        this implementation of the libjpeg API, or FALSE otherwise.
 void jpeg_c_set_int_param (j_compress_ptr cinfo, J_INT_PARAM param,
                          int value)
        Set the given integer extension parameter to the given value.
 int jpeg_c_get_int_param (j_compress_ptr cinfo, J_INT_PARAM param)
        Get the value of the given integer extension parameter.
 Boolean Extension Parameters Supported by mozjpeg
 -------------------------------------------------
 * JBOOLEAN_OPTIMIZE_SCANS (default: TRUE)
  Specifies whether scan parameters should be optimized.  Parameter
  optimization is done as in jpgcrush. jpeg_simple_progression() should be called
  after setting JBOOLEAN_OPTIMIZE_SCANS.
  When disabling JBOOLEAN_OPTIMIZE_SCANS, cinfo.scan_info should additionally be
  set to NULL to disable use of the progressive coding mode, if so desired.
 * JBOOLEAN_TRELLIS_QUANT (default: TRUE)
  Specifies whether to apply trellis quantization.  For each 8x8 block, trellis
  quantization determines the best tradeoff between rate and distortion.
 * JBOOLEAN_TRELLIS_QUANT_DC (default: TRUE)
  Specifies whether to apply trellis quantization to DC coefficients.
 * JBOOLEAN_TRELLIS_EOB_OPT (default: FALSE)
  Specifies whether to optimize runs of zero blocks in trellis quantization.
  This is applicable only when JBOOLEAN_USE_SCANS_IN_TRELLIS is enabled.
 * JBOOLEAN_USE_LAMBDA_WEIGHT_TBL currently has no effect.
 * JBOOLEAN_USE_SCANS_IN_TRELLIS (default: FALSE)
  Specifies whether multiple scans should be considered during trellis
  quantization.
 * JBOOLEAN_TRELLIS_Q_OPT (default: FALSE)
  Specifies whether to optimize the quantization table after trellis
  quantization.  If enabled, then a revised quantization table is derived so
  as to minimize the reconstruction error of the quantized coefficients.
 * JBOOLEAN_OVERSHOOT_DERINGING (default: TRUE)
  Specifies whether overshooting is applied to samples with extreme values
  (for example, 0 and 255 for 8-bit samples).  Overshooting may reduce ringing
  artifacts from compression, in particular in areas where black text appears
  on a white background.
 Floating Point Extension Parameters Supported by mozjpeg
 --------------------------------------------------------
 * JFLOAT_LAMBDA_LOG_SCALE1 (default: 14.75)
  JFLOAT_LAMBDA_LOG_SCALE2 (default: 16.5)
  These parameters specify the lambda value used in trellis quantization.  The
  lambda value (Lagrange multiplier) in the
    R + lambda * D
  equation is derived from
    lambda = 2^s1 / ((2^s2 + n) * q^2),
  where s1 and s2 are the values of JFLOAT_LAMBDA_LOG_SCALE1 and
  JFLOAT_LAMBDA_LOG_SCALE2, n is the average of the squared unquantized AC
  coefficients within the current 8x8 block, and q is the quantization table
  entry associated with the current coefficient frequency.  If
  JFLOAT_LAMBDA_LOG_SCALE2 is 0, then an alternate form is used that does not
  rely on n:
    lambda = 2^(s1-12) / q^2.
 * JFLOAT_TRELLIS_DELTA_DC_WEIGHT (default: 0.0)
  This parameter controls how distortion is calculated in DC trellis quantization
  (enabled with JBOOLEAN_TRELLIS_QUANT_DC). It defines weighting between distortion
  of the DC coefficient and distortion of the vertical gradient of DC coefficients.
  The value of the parameter corresponds to the weight applied to the distortion
  of the vertical gradient.
 Integer Extension Parameters Supported by mozjpeg
 -------------------------------------------------
 * JINT_COMPRESS_PROFILE (default: JCP_MAX_COMPRESSION)
  Select a compression profile, which is a set of default parameters that will
  achieve a desired compression goal.  This parameter controls the behavior of
  the jpeg_set_defaults() function.  Thus, setting JINT_COMPRESS_PROFILE does
  not cause any other parameters to be modified until jpeg_set_defaults() is
  called.  The following compression profiles are supported:
  - JCP_MAX_COMPRESSION (default)
    Increase the compression ratio as much as possible, at the expense of
    increased encoding time.  This enables progressive entropy coding and all
    mozjpeg extensions.
  - JCP_FASTEST
    Use the libjpeg[-turbo] defaults (baseline entropy coding, no mozjpeg
    extensions enabled.)
 * JINT_TRELLIS_FREQ_SPLIT (default: 8)
  Specifies the position within the zigzag scan at which the split between
  scans is positioned in the context of trellis quantization.
  JBOOLEAN_USE_SCANS_IN_TRELLIS must be enabled for this parameter to have any
  effect.
 * JINT_TRELLIS_NUM_LOOPS (default: 1)
  Specifies the number of trellis quantization passes.  Huffman tables are
  updated between passes.
 * JINT_BASE_QUANT_TBL_IDX (default: 3)
  Specifies which quantization table set to use.  The following options are
  available:
  0 = Tables from JPEG Annex K
  1 = Flat table
  2 = Table tuned for MSSIM on Kodak image set
  3 = Table from http://www.imagemagick.org/discourse-server/viewtopic.php?f=22&t=20333&p=98008#p98008
  4 = Table tuned for PSNR-HVS-M on Kodak image set
  5 = Table from:  Relevance of Human Vision to JPEG-DCT Compression
      (1992) Klein, Silverstein and Carney
  6 = Table from:  DCTune Perceptual Optimization of Compressed Dental X-Rays
      (1997) Watson, Taylor, Borthwick
  7 = Table from:  A Visual Detection Model for DCT Coefficient Quantization
      (12/9/93) Ahumada, Watson, Peterson
  8 = Table from:  An Improved Detection Model for DCT Coefficient Quantization
      (1993) Peterson, Ahumada and Watson
 * JINT_DC_SCAN_OPT_MODE (default: 1)
  Specifies the DC scan optimization mode.  The following options are
  available:
  0 = One scan for all components
  1 = One scan per component
  2 = Optimize between one scan for all components and one scan for the first
      component plus one scan for the remaining components
--- a/README-turbo.txt
+++ b/README-turbo.txt
--- a/README.md
+++ b/README.md
@@ -9,6 +9,7 @@ The idea is to reduce transfer times for JPEGs on the Web, thus reducing page lo
 More information:
 * [Version 1.0 Announcement](https://blog.mozilla.org/research/2014/03/05/introducing-the-mozjpeg-project/)
 * [Version 2.0 Announcement](https://blog.mozilla.org/research/2014/07/15/mozilla-advances-jpeg-encoding-with-mozjpeg-2-0/)
 * [Mailing List](https://lists.mozilla.org/listinfo/dev-mozjpeg)</a>
 * [Version 3.0 Announcement](https://boomswaggerboom.wordpress.com/2014/12/30/mozjpeg-3-0-released/)
 * [Version 2.0 Announcement](https://blog.mozilla.org/research/2014/07/15/mozilla-advances-jpeg-encoding-with-mozjpeg-2-0/)
 * [Version 1.0 Announcement](https://blog.mozilla.org/research/2014/03/05/introducing-the-mozjpeg-project/)
--- a/cderror.h
+++ b/cderror.h
@@ -74,6 +74,7 @@ JMESSAGE(JWRN_GIF_NOMOREDATA, "Ran out of GIF bits")
 #ifdef PPM_SUPPORTED
 JMESSAGE(JERR_PPM_COLORSPACE, "PPM output must be grayscale or RGB")
 JMESSAGE(JERR_PPM_NONNUMERIC, "Nonnumeric data in PPM file")
 JMESSAGE(JERR_PPM_TOOLARGE, "Integer value too large in PPM file")
 JMESSAGE(JERR_PPM_NOT, "Not a PPM/PGM file")
 JMESSAGE(JTRC_PGM, "%ux%u PGM image")
 JMESSAGE(JTRC_PGM_TEXT, "%ux%u text PGM image")
--- a/cjpeg.1
+++ b/cjpeg.1
@@ -1,4 +1,4 @@
-.TH CJPEG 1 "11 May 2014"
+.TH CJPEG 1 "21 November 2014"
 .SH NAME
 cjpeg \- compress an image file to a JPEG file
 .SH SYNOPSIS
@@ -221,6 +221,9 @@ give more output.  Also, version information is printed at startup.
 .B \-debug
 Same as
 .BR \-verbose .
 .TP
 .B \-version
 Print version information and exit.
 .PP
 The
 .B \-restart
--- a/cjpeg.c
+++ b/cjpeg.c
@@ -5,7 +5,7 @@
 * Copyright (C) 1991-1998, Thomas G. Lane.
 * Modified 2003-2011 by Guido Vollbeding.
 * libjpeg-turbo Modifications:
- * Copyright (C) 2010, 2013, D. R. Commander.
+ * Copyright (C) 2010, 2013-2014, D. R. Commander.
 * mozjpeg Modifications:
 * Copyright (C) 2014, Mozilla Corporation.
 * For conditions of distribution and use, see the accompanying README file.
@@ -178,8 +178,11 @@ usage (void)
 #endif
  fprintf(stderr, "  -revert        Revert to standard defaults (instead of mozjpeg defaults)\n");
  fprintf(stderr, "  -fastcrush     Disable progressive scan optimization\n");
-  fprintf(stderr, "  -opt-dc-scan   Optimize DC scans (may be incompatible with some JPEG decoders)\n");
+  fprintf(stderr, "  -dc-scan-opt   DC scan optimization mode\n");
-  fprintf(stderr, "  -split-dc-scan Use one DC scan per component (may be incompatible with some JPEG decoders?)\n");
+  fprintf(stderr, "                 - 0 One scan for all components\n");
  fprintf(stderr, "                 - 1 One scan per component (default)\n");
  fprintf(stderr, "                 - 2 Optimize between one scan for all components and one scan for 1st component\n");
  fprintf(stderr, "                     plus one scan for remaining components\n");
  fprintf(stderr, "  -notrellis     Disable trellis optimization\n");
  fprintf(stderr, "  -trellis-dc    Enable trellis optimization of DC coefficients (default)\n");
  fprintf(stderr, "  -notrellis-dc  Disable trellis optimization of DC coefficients\n");
@@ -204,6 +207,14 @@ usage (void)
  fprintf(stderr, "  -dct float     Use floating-point DCT method%s\n",
          (JDCT_DEFAULT == JDCT_FLOAT ? " (default)" : ""));
 #endif
  fprintf(stderr, "  -quant-baseline Use 8-bit quantization table entries for baseline JPEG compatibility\n");
  fprintf(stderr, "  -quant-table N Use predefined quantization table N:\n");
  fprintf(stderr, "                 - 0 JPEG Annex K\n");
  fprintf(stderr, "                 - 1 Flat\n");
  fprintf(stderr, "                 - 2 Custom, tuned for MS-SSIM\n");
  fprintf(stderr, "                 - 3 ImageMagick table by N. Robidoux\n");
  fprintf(stderr, "                 - 4 Custom, tuned for PSNR-HVS\n");
  fprintf(stderr, "                 - 5 Table from paper by Klein, Silverstein and Carney\n");
  fprintf(stderr, "  -restart N     Set restart interval in rows, or in blocks with B\n");
 #ifdef INPUT_SMOOTHING_SUPPORTED
  fprintf(stderr, "  -smooth N      Smooth dithered input (N=1..100 is strength)\n");
@@ -214,6 +225,7 @@ usage (void)
  fprintf(stderr, "  -memdst        Compress to memory instead of file (useful for benchmarking)\n");
 #endif
  fprintf(stderr, "  -verbose  or  -debug   Emit debug output\n");
  fprintf(stderr, "  -version       Print version information and exit\n");
  fprintf(stderr, "Switches for wizards:\n");
  fprintf(stderr, "  -qtables file  Use quantization tables given in file\n");
  fprintf(stderr, "  -qslots N[,...]    Set component quantization tables\n");
@@ -278,6 +290,9 @@ parse_switches (j_compress_ptr cinfo, int argc, char **argv,
      /* Use arithmetic coding. */
 #ifdef C_ARITH_CODING_SUPPORTED
      cinfo->arith_code = TRUE;
      /* No table optimization required for AC */
      cinfo->optimize_coding = FALSE;
 #else
      fprintf(stderr, "%s: sorry, arithmetic coding not supported\n",
              progname);
@@ -324,13 +339,14 @@ parse_switches (j_compress_ptr cinfo, int argc, char **argv,
      }
      cinfo->err->trace_level++;
    } else if (keymatch(arg, "version", 4)) {
      fprintf(stderr, "%s version %s (build %s)\n",
              PACKAGE_NAME, VERSION, BUILD);
      exit(EXIT_SUCCESS);
    } else if (keymatch(arg, "fastcrush", 4)) {
      jpeg_c_set_bool_param(cinfo, JBOOLEAN_OPTIMIZE_SCANS, FALSE);
    } else if (keymatch(arg, "flat", 4)) {
      jpeg_c_set_bool_param(cinfo, JBOOLEAN_USE_FLAT_QUANT_TBL, TRUE);
      jpeg_set_quality(cinfo, 75, TRUE);
    } else if (keymatch(arg, "grayscale", 2) || keymatch(arg, "greyscale",2)) {
      /* Force a monochrome JPEG file to be generated. */
      jpeg_set_colorspace(cinfo, JCS_GRAYSCALE);
@@ -364,8 +380,12 @@ parse_switches (j_compress_ptr cinfo, int argc, char **argv,
        lval *= 1000L;
      cinfo->mem->max_memory_to_use = lval * 1000L;
-    } else if (keymatch(arg, "opt-dc-scan", 6)) {
+    } else if (keymatch(arg, "dc-scan-opt", 3)) {
-      jpeg_c_set_bool_param(cinfo, JBOOLEAN_ONE_DC_SCAN, FALSE);
+      if (++argn >= argc) {      /* advance to next argument */
        fprintf(stderr, "%s: missing argument for dc-scan-opt\n", progname);
        usage();
      }
      jpeg_c_set_int_param(cinfo, JINT_DC_SCAN_OPT_MODE, atoi(argv[argn]));
    } else if (keymatch(arg, "optimize", 1) || keymatch(arg, "optimise", 1)) {
      /* Enable entropy parm optimization. */
@@ -431,6 +451,22 @@ parse_switches (j_compress_ptr cinfo, int argc, char **argv,
      qtablefile = argv[argn];
      /* We postpone actually reading the file in case -quality comes later. */
    } else if (keymatch(arg, "quant-table", 7)) {
      int val;
      if (++argn >= argc)       /* advance to next argument */
        usage();
      val = atoi(argv[argn]);
      jpeg_c_set_int_param(cinfo, JINT_BASE_QUANT_TBL_IDX, val);
      if (jpeg_c_get_int_param(cinfo, JINT_BASE_QUANT_TBL_IDX) != val) {
        fprintf(stderr, "%s: %d is invalid argument for quant-table\n", progname, val);
        usage();
      }
      jpeg_set_quality(cinfo, 75, TRUE);
    } else if (keymatch(arg, "quant-baseline", 7)) {
      /* Force quantization table to meet baseline requirements */
      force_baseline = TRUE;
    } else if (keymatch(arg, "restart", 1)) {
      /* Restart interval in MCU rows (or in MCUs with 'b'). */
      long lval;
@@ -452,7 +488,7 @@ parse_switches (j_compress_ptr cinfo, int argc, char **argv,
    } else if (keymatch(arg, "revert", 3)) {
      /* revert to old JPEG default */
-      jpeg_c_set_bool_param(cinfo, JBOOLEAN_USE_MOZ_DEFAULTS, FALSE);
+      jpeg_c_set_int_param(cinfo, JINT_COMPRESS_PROFILE, JCP_FASTEST);
      jpeg_set_defaults(cinfo);
    } else if (keymatch(arg, "sample", 2)) {
@@ -490,10 +526,6 @@ parse_switches (j_compress_ptr cinfo, int argc, char **argv,
        usage();
      cinfo->smoothing_factor = val;
    } else if (keymatch(arg, "split-dc-scans", 3)) {
      jpeg_c_set_bool_param(cinfo, JBOOLEAN_ONE_DC_SCAN, FALSE);
      jpeg_c_set_bool_param(cinfo, JBOOLEAN_SEP_DC_SCAN, TRUE);
    } else if (keymatch(arg, "targa", 1)) {
      /* Input file is Targa format. */
      is_targa = TRUE;
@@ -506,35 +538,42 @@ parse_switches (j_compress_ptr cinfo, int argc, char **argv,
      /* disable trellis quantization */
      jpeg_c_set_bool_param(cinfo, JBOOLEAN_TRELLIS_QUANT, FALSE);
    } else if (keymatch(arg, "trellis-dc-ver-weight", 12)) {
      if (++argn >= argc) {      /* advance to next argument */
        fprintf(stderr, "%s: missing argument for trellis-dc-ver-weight\n", progname);
        usage();
      }
      jpeg_c_set_float_param(cinfo, JFLOAT_TRELLIS_DELTA_DC_WEIGHT, atof(argv[argn]));
    } else if (keymatch(arg, "trellis-dc", 9)) {
      /* enable DC trellis quantization */
      jpeg_c_set_bool_param(cinfo, JBOOLEAN_TRELLIS_QUANT_DC, TRUE);
    } else if (keymatch(arg, "tune-psnr", 6)) {
-      jpeg_c_set_bool_param(cinfo, JBOOLEAN_USE_FLAT_QUANT_TBL, TRUE);
+      jpeg_c_set_int_param(cinfo, JINT_BASE_QUANT_TBL_IDX, 1);
      jpeg_c_set_float_param(cinfo, JFLOAT_LAMBDA_LOG_SCALE1, 9.0);
      jpeg_c_set_float_param(cinfo, JFLOAT_LAMBDA_LOG_SCALE2, 0.0);
      jpeg_c_set_bool_param(cinfo, JBOOLEAN_USE_LAMBDA_WEIGHT_TBL, FALSE);
      jpeg_set_quality(cinfo, 75, TRUE);
    } else if (keymatch(arg, "tune-ssim", 6)) {
-      jpeg_c_set_bool_param(cinfo, JBOOLEAN_USE_FLAT_QUANT_TBL, TRUE);
+      jpeg_c_set_int_param(cinfo, JINT_BASE_QUANT_TBL_IDX, 1);
-      jpeg_c_set_float_param(cinfo, JFLOAT_LAMBDA_LOG_SCALE1, 12.0);
+      jpeg_c_set_float_param(cinfo, JFLOAT_LAMBDA_LOG_SCALE1, 11.5);
-      jpeg_c_set_float_param(cinfo, JFLOAT_LAMBDA_LOG_SCALE2, 13.5);
+      jpeg_c_set_float_param(cinfo, JFLOAT_LAMBDA_LOG_SCALE2, 12.75);
      jpeg_c_set_bool_param(cinfo, JBOOLEAN_USE_LAMBDA_WEIGHT_TBL, FALSE);
      jpeg_set_quality(cinfo, 75, TRUE);
    } else if (keymatch(arg, "tune-ms-ssim", 6)) {
-      jpeg_c_set_bool_param(cinfo, JBOOLEAN_USE_FLAT_QUANT_TBL, FALSE);
+      jpeg_c_set_int_param(cinfo, JINT_BASE_QUANT_TBL_IDX, 3);
-      jpeg_c_set_float_param(cinfo, JFLOAT_LAMBDA_LOG_SCALE1, 14.25);
+      jpeg_c_set_float_param(cinfo, JFLOAT_LAMBDA_LOG_SCALE1, 12.0);
-      jpeg_c_set_float_param(cinfo, JFLOAT_LAMBDA_LOG_SCALE2, 12.75);
+      jpeg_c_set_float_param(cinfo, JFLOAT_LAMBDA_LOG_SCALE2, 13.0);
      jpeg_c_set_bool_param(cinfo, JBOOLEAN_USE_LAMBDA_WEIGHT_TBL, TRUE);
      jpeg_set_quality(cinfo, 75, TRUE);
    } else if (keymatch(arg, "tune-hvs-psnr", 6)) {
-      jpeg_c_set_bool_param(cinfo, JBOOLEAN_USE_FLAT_QUANT_TBL, FALSE);
+      jpeg_c_set_int_param(cinfo, JINT_BASE_QUANT_TBL_IDX, 3);
-      jpeg_c_set_float_param(cinfo, JFLOAT_LAMBDA_LOG_SCALE1, 16.0);
+      jpeg_c_set_float_param(cinfo, JFLOAT_LAMBDA_LOG_SCALE1, 14.75);
-      jpeg_c_set_float_param(cinfo, JFLOAT_LAMBDA_LOG_SCALE2, 15.5);
+      jpeg_c_set_float_param(cinfo, JFLOAT_LAMBDA_LOG_SCALE2, 16.5);
      jpeg_c_set_bool_param(cinfo, JBOOLEAN_USE_LAMBDA_WEIGHT_TBL, TRUE);
      jpeg_set_quality(cinfo, 75, TRUE);
@@ -634,8 +673,6 @@ main (int argc, char **argv)
   */
  cinfo.in_color_space = JCS_RGB; /* arbitrary guess */
  if (jpeg_c_bool_param_supported(&cinfo, JBOOLEAN_USE_MOZ_DEFAULTS))
    jpeg_c_set_bool_param(&cinfo, JBOOLEAN_USE_MOZ_DEFAULTS, TRUE);
  jpeg_set_defaults(&cinfo);
  /* Scan command line to find file names.
@@ -753,7 +790,8 @@ main (int argc, char **argv)
          GETJOCTET(marker->data[3]) == 0x62 &&
          GETJOCTET(marker->data[4]) == 0x65)
        continue;                       /* reject duplicate Adobe */
-      jpeg_write_marker(&cinfo, marker->marker, marker->data, marker->data_length);
+      jpeg_write_marker(&cinfo, marker->marker, marker->data,
                        marker->data_length);
    }
  }
--- a/configure.ac
+++ b/configure.ac
@@ -2,12 +2,12 @@
 # Process this file with autoconf to produce a configure script.
 AC_PREREQ([2.56])
-AC_INIT([libmozjpeg], [2.1])
+AC_INIT([mozjpeg], [3.1])
 BUILD=`date +%Y%m%d`
 AM_INIT_AUTOMAKE([-Wall foreign dist-bzip2])
 m4_ifdef([AM_SILENT_RULES], [AM_SILENT_RULES(yes)])
-AC_PREFIX_DEFAULT(/opt/libmozjpeg)
+AC_PREFIX_DEFAULT(/opt/mozjpeg)
 # Checks for programs.
 SAVED_CFLAGS=${CFLAGS}
@@ -22,7 +22,7 @@ m4_ifdef([AM_PROG_AR], [AM_PROG_AR])
 AC_PROG_LIBTOOL
 AC_PROG_LN_S
-# When the prefix is /opt/libmozjpeg, we assume that an "official" binary is
+# When the prefix is /opt/mozjpeg, we assume that an "official" binary is
 # being created, and thus we install things into specific locations.
 old_prefix=${prefix}
@@ -31,12 +31,12 @@ if test "x$prefix" = "xNONE" -a "x$ac_default_prefix" != "x"; then
 fi
 DATADIR=`eval echo ${datadir}`
 DATADIR=`eval echo $DATADIR`
-if test "$DATADIR" = "/opt/libmozjpeg/share"; then
+if test "$DATADIR" = "/opt/mozjpeg/share"; then
  datadir='${prefix}'
 fi
 DATADIR=`eval echo ${datarootdir}`
 DATADIR=`eval echo $DATADIR`
-if test "$DATADIR" = "/opt/libmozjpeg/share"; then
+if test "$DATADIR" = "/opt/mozjpeg/share"; then
  datarootdir='${prefix}'
 fi
@@ -48,7 +48,7 @@ fi
 if test "x${libdir}" = 'x${exec_prefix}/lib' -o "x${libdir}" = 'x${prefix}/lib'; then
  LIBDIR=`eval echo ${libdir}`
  LIBDIR=`eval echo $LIBDIR`
-  if test "$LIBDIR" = "/opt/libmozjpeg/lib"; then
+  if test "$LIBDIR" = "/opt/mozjpeg/lib"; then
    case $host_os in
      darwin*)
        ;;
@@ -149,10 +149,10 @@ AC_ARG_VAR(JPEG_LIB_VERSION, [libjpeg API version (62, 70, or 80)])
 if test "x$JPEG_LIB_VERSION" = "x"; then
  AC_ARG_WITH([jpeg7],
    AC_HELP_STRING([--with-jpeg7],
-      [Emulate libjpeg v7 API/ABI (this makes libmozjpeg backward incompatible with libjpeg v6b.)]))
+      [Emulate libjpeg v7 API/ABI (this makes mozjpeg backward incompatible with libjpeg v6b.)]))
  AC_ARG_WITH([jpeg8],
    AC_HELP_STRING([--with-jpeg8],
-      [Emulate libjpeg v8 API/ABI (this makes libmozjpeg backward incompatible with libjpeg v6b.)]))
+      [Emulate libjpeg v8 API/ABI (this makes mozjpeg backward incompatible with libjpeg v6b.)]))
  if test "x${with_jpeg8}" = "xyes"; then
    JPEG_LIB_VERSION=80
  else
@@ -170,9 +170,9 @@ AC_DEFINE_UNQUOTED(JPEG_LIB_VERSION, [$JPEG_LIB_VERSION],
  [libjpeg API version])
 AC_ARG_VAR(SO_MAJOR_VERSION,
-  [Major version of the libmozjpeg shared library (default is determined by the API version)])
+  [Major version of the mozjpeg shared library (default is determined by the API version)])
 AC_ARG_VAR(SO_MINOR_VERSION,
-  [Minor version of the libmozjpeg shared library (default is determined by the API version)])
+  [Minor version of the mozjpeg shared library (default is determined by the API version)])
 if test "x$SO_MAJOR_VERSION" = "x"; then
  case "$JPEG_LIB_VERSION" in
    62)  SO_MAJOR_VERSION=$JPEG_LIB_VERSION ;;
@@ -217,12 +217,12 @@ AC_SUBST(SO_MINOR_VERSION)
 AC_SUBST(SO_AGE)
 AC_SUBST(MEM_SRCDST_FUNCTIONS)
-AC_DEFINE_UNQUOTED(LIBMOZJPEG_VERSION, [$VERSION], [libmozjpeg version])
+AC_DEFINE_UNQUOTED(MOZJPEG_VERSION, [$VERSION], [mozjpeg version])
 VERSION_SCRIPT=yes
 AC_ARG_ENABLE([ld-version-script],
  AS_HELP_STRING([--disable-ld-version-script],
-    [Disable linker version script for libmozjpeg (default is to use linker version script if the linker supports it)]),
+    [Disable linker version script for mozjpeg (default is to use linker version script if the linker supports it)]),
  [VERSION_SCRIPT=$enableval], [])
 AC_MSG_CHECKING([whether the linker supports version scripts])
@@ -250,7 +250,7 @@ if test "x$VERSION_SCRIPT_FLAG" = "x"; then
 fi
 LDFLAGS="$SAVED_LDFLAGS"
-AC_MSG_CHECKING([whether to use version script when building libmozjpeg])
+AC_MSG_CHECKING([whether to use version script when building mozjpeg])
 AC_MSG_RESULT($VERSION_SCRIPT)
 AM_CONDITIONAL(VERSION_SCRIPT, test "x$VERSION_SCRIPT" = "xyes")
@@ -443,7 +443,11 @@ if test "x${with_simd}" != "xno"; then
      AC_MSG_RESULT([yes (arm)])
      AC_MSG_CHECKING([if the assembler is GNU-compatible and can be used])
      AC_CHECK_COMPATIBLE_ARM_ASSEMBLER_IFELSE(
-        [AC_MSG_RESULT([yes])
+        [if test "x$ac_use_gas_preprocessor" = "xyes"; then
           AC_MSG_RESULT([yes (with gas-preprocessor)])
         else
           AC_MSG_RESULT([yes])
         fi
         simd_arch=arm],
        [AC_MSG_RESULT([no])
         with_simd=no])
@@ -459,7 +463,11 @@ if test "x${with_simd}" != "xno"; then
      AC_MSG_RESULT([yes (arm64)])
      AC_MSG_CHECKING([if the assembler is GNU-compatible and can be used])
      AC_CHECK_COMPATIBLE_ARM64_ASSEMBLER_IFELSE(
-        [AC_MSG_RESULT([yes])
+        [if test "x$ac_use_gas_preprocessor" = "xyes"; then
           AC_MSG_RESULT([yes (with gas-preprocessor)])
         else
           AC_MSG_RESULT([yes])
         fi
         simd_arch=aarch64],
        [AC_MSG_RESULT([no])
         with_simd=no])
@@ -520,7 +528,7 @@ AM_CONDITIONAL([SIMD_POWERPC], [test "x$simd_arch" = "xpowerpc"])
 AM_CONDITIONAL([X86_64], [test "x$host_cpu" = "xx86_64" -o "x$host_cpu" = "xamd64"])
 AM_CONDITIONAL([WITH_TURBOJPEG], [test "x$with_turbojpeg" != "xno"])
-AC_ARG_VAR(PKGNAME, [distribution package name (default: libmozjpeg)])
+AC_ARG_VAR(PKGNAME, [distribution package name (default: mozjpeg)])
 if test "x$PKGNAME" = "x"; then
  PKGNAME=$PACKAGE_NAME
 fi
@@ -551,7 +559,7 @@ AC_DEFINE_UNQUOTED([BUILD], "$BUILD", [libjpeg-turbo build number])
 AC_CONFIG_HEADERS([config.h])
 AC_CONFIG_HEADERS([jconfig.h])
 AC_CONFIG_HEADERS([jconfigint.h])
-AC_CONFIG_FILES([pkgscripts/libmozjpeg.spec.tmpl:release/libmozjpeg.spec.in])
+AC_CONFIG_FILES([pkgscripts/mozjpeg.spec.tmpl:release/mozjpeg.spec.in])
 AC_CONFIG_FILES([pkgscripts/makecygwinpkg.tmpl:release/makecygwinpkg.in])
 AC_CONFIG_FILES([pkgscripts/makedpkg.tmpl:release/makedpkg.in])
 AC_CONFIG_FILES([pkgscripts/makemacpkg.tmpl:release/makemacpkg.in])
--- a/djpeg.1
+++ b/djpeg.1
@@ -1,4 +1,4 @@
-.TH DJPEG 1 "11 May 2014"
+.TH DJPEG 1 "21 November 2014"
 .SH NAME
 djpeg \- decompress a JPEG file to an image file
 .SH SYNOPSIS
@@ -202,6 +202,9 @@ give more output.  Also, version information is printed at startup.
 .B \-debug
 Same as
 .BR \-verbose .
 .TP
 .B \-version
 Print version information and exit.
 .SH EXAMPLES
 .LP
 This example decompresses the JPEG file foo.jpg, quantizes it to
--- a/djpeg.c
+++ b/djpeg.c
@@ -165,6 +165,7 @@ usage (void)
 #endif
  fprintf(stderr, "  -verbose  or  -debug   Emit debug output\n");
  fprintf(stderr, "  -version       Print version information and exit\n");
  exit(EXIT_FAILURE);
 }
@@ -261,6 +262,11 @@ parse_switches (j_decompress_ptr cinfo, int argc, char **argv,
      }
      cinfo->err->trace_level++;
    } else if (keymatch(arg, "version", 4)) {
      fprintf(stderr, "%s version %s (build %s)\n",
              PACKAGE_NAME, VERSION, BUILD);
      exit(EXIT_SUCCESS);
    } else if (keymatch(arg, "fast", 1)) {
      /* Select recommended processing options for quick-and-dirty output. */
      cinfo->two_pass_quantize = FALSE;
--- a/example.c
+++ b/example.c
@@ -126,8 +126,6 @@ write_JPEG_file (char * filename, int quality)
  cinfo.image_height = image_height;
  cinfo.input_components = 3;           /* # of color components per pixel */
  cinfo.in_color_space = JCS_RGB;       /* colorspace of input image */
  jpeg_c_set_bool_param(&cinfo, JBOOLEAN_USE_MOZ_DEFAULTS, TRUE); /* use Mozilla defaults for improved compression */
  /* Now use the library's routine to set default compression parameters.
   * (You must set at least cinfo.in_color_space before calling this,
   * since the defaults depend on the source color space.)
--- a/java/org/libjpegturbo/turbojpeg/TJCompressor.java
+++ b/java/org/libjpegturbo/turbojpeg/TJCompressor.java
@@ -567,6 +567,7 @@ public class TJCompressor {
   * Free the native structures associated with this compressor instance.
   */
  public void close() throws Exception {
    if (handle != 0)
      destroy();
  }
--- a/java/org/libjpegturbo/turbojpeg/TJDecompressor.java
+++ b/java/org/libjpegturbo/turbojpeg/TJDecompressor.java
@@ -834,6 +834,7 @@ public class TJDecompressor {
   * Free the native structures associated with this decompressor instance.
   */
  public void close() throws Exception {
    if (handle != 0)
      destroy();
  }
--- a/jcapimin.c
+++ b/jcapimin.c
@@ -23,6 +23,7 @@
 #include "jinclude.h"
 #include "jpeglib.h"
 #include "jmemsys.h"
 #include "jcmaster.h"
 /*
@@ -97,8 +98,11 @@ jpeg_CreateCompress (j_compress_ptr cinfo, int version, size_t structsize)
   * here.  It is later reallocated by jinit_c_master_control().
   */
  cinfo->master = (struct jpeg_comp_master *)
-      jpeg_get_small ((j_common_ptr) cinfo, sizeof(struct jpeg_comp_master));
+      (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_PERMANENT,
-  MEMZERO(cinfo->master, sizeof(struct jpeg_comp_master));
+                                  sizeof(my_comp_master));
  MEMZERO(cinfo->master, sizeof(my_comp_master));
  cinfo->master->compress_profile = JCP_MAX_COMPRESSION;
 }
--- a/jcarith.c
+++ b/jcarith.c
@@ -18,7 +18,7 @@
 #define JPEG_INTERNALS
 #include "jinclude.h"
 #include "jpeglib.h"
-
+#include <math.h>
 /* Expanded entropy encoder object for arithmetic encoding. */
@@ -120,6 +120,10 @@ emit_byte (int val, j_compress_ptr cinfo)
 {
  struct jpeg_destination_mgr * dest = cinfo->dest;
  /* Do not emit bytes during trellis passes */
  if (cinfo->master->trellis_passes)
    return;
  *dest->next_output_byte++ = (JOCTET) val;
  if (--dest->free_in_buffer == 0)
    if (! (*dest->empty_output_buffer) (cinfo))
@@ -826,6 +830,7 @@ start_pass (j_compress_ptr cinfo, boolean gather_statistics)
  arith_entropy_ptr entropy = (arith_entropy_ptr) cinfo->entropy;
  int ci, tbl;
  jpeg_component_info * compptr;
  boolean progressive_mode;
  if (gather_statistics)
    /* Make sure to avoid that in the master control logic!
@@ -836,8 +841,12 @@ start_pass (j_compress_ptr cinfo, boolean gather_statistics)
  /* We assume jcmaster.c already validated the progressive scan parameters. */
  /* Trellis optimization does DC and AC in same pass and without refinement
   * so consider progressive mode to be off in such case */
  progressive_mode = (cinfo->master->trellis_passes) ? FALSE : cinfo->progressive_mode;
  /* Select execution routines */
-  if (cinfo->progressive_mode) {
+  if (progressive_mode) {
    if (cinfo->Ah == 0) {
      if (cinfo->Ss == 0)
        entropy->pub.encode_mcu = encode_mcu_DC_first;
@@ -856,7 +865,7 @@ start_pass (j_compress_ptr cinfo, boolean gather_statistics)
  for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
    compptr = cinfo->cur_comp_info[ci];
    /* DC needs no table for refinement scan */
-    if (cinfo->progressive_mode == 0 || (cinfo->Ss == 0 && cinfo->Ah == 0)) {
+    if (progressive_mode == 0 || (cinfo->Ss == 0 && cinfo->Ah == 0)) {
      tbl = compptr->dc_tbl_no;
      if (tbl < 0 || tbl >= NUM_ARITH_TBLS)
        ERREXIT1(cinfo, JERR_NO_ARITH_TABLE, tbl);
@@ -869,7 +878,7 @@ start_pass (j_compress_ptr cinfo, boolean gather_statistics)
      entropy->dc_context[ci] = 0;
    }
    /* AC needs no table when not present */
-    if (cinfo->progressive_mode == 0 || cinfo->Se) {
+    if (progressive_mode == 0 || cinfo->Se) {
      tbl = compptr->ac_tbl_no;
      if (tbl < 0 || tbl >= NUM_ARITH_TBLS)
        ERREXIT1(cinfo, JERR_NO_ARITH_TABLE, tbl);
@@ -878,7 +887,7 @@ start_pass (j_compress_ptr cinfo, boolean gather_statistics)
          ((j_common_ptr) cinfo, JPOOL_IMAGE, AC_STAT_BINS);
      MEMZERO(entropy->ac_stats[tbl], AC_STAT_BINS);
 #ifdef CALCULATE_SPECTRAL_CONDITIONING
-      if (cinfo->progressive_mode)
+      if (progressive_mode)
        /* Section G.1.3.2: Set appropriate arithmetic conditioning value Kx */
        cinfo->arith_ac_K[tbl] = cinfo->Ss + ((8 + cinfo->Se - cinfo->Ss) >> 4);
 #endif
@@ -925,3 +934,34 @@ jinit_arith_encoder (j_compress_ptr cinfo)
  /* Initialize index for fixed probability estimation */
  entropy->fixed_bin[0] = 113;
 }
 GLOBAL(void)
 jget_arith_rates (j_compress_ptr cinfo, int dc_tbl_no, int ac_tbl_no, arith_rates *r)
 {
  int i;
  arith_entropy_ptr entropy = (arith_entropy_ptr) cinfo->entropy;
  r->arith_dc_L = cinfo->arith_dc_L[dc_tbl_no];
  r->arith_dc_U = cinfo->arith_dc_U[dc_tbl_no];
  r->arith_ac_K = cinfo->arith_ac_K[ac_tbl_no];
  for (i = 0; i < DC_STAT_BINS; i++) {
    int state = entropy->dc_stats[dc_tbl_no][i];
    int mps_val = state >> 7;
    float prob_lps = (jpeg_aritab[state & 0x7f] >> 16) / 46340.95; /* 32768*sqrt(2) */
    float prob_0 = (mps_val) ? prob_lps : 1.0 - prob_lps;
    float prob_1 = 1.0 - prob_0;
    r->rate_dc[i][0] = -log(prob_0) / log(2.0);
    r->rate_dc[i][1] = -log(prob_1) / log(2.0);
  }
  for (i = 0; i < AC_STAT_BINS; i++) {
    int state = entropy->ac_stats[ac_tbl_no][i];
    int mps_val = state >> 7;
    float prob_lps = (jpeg_aritab[state & 0x7f] >> 16) / 46340.95;
    float prob_0 = (mps_val) ? prob_lps : 1.0 - prob_lps;
    float prob_1 = 1.0 - prob_0;
    r->rate_ac[i][0] = -log(prob_0) / log(2.0);
    r->rate_ac[i][1] = -log(prob_1) / log(2.0);
  }
 }
--- a/jccoefct.c
+++ b/jccoefct.c
@@ -190,7 +190,8 @@ compress_data (j_compress_ptr cinfo, JSAMPIMAGE input_buf)
            (*cinfo->fdct->forward_DCT) (cinfo, compptr,
                                         input_buf[compptr->component_index],
                                         coef->MCU_buffer[blkn],
-					 ypos, xpos, (JDIMENSION) blockcnt, NULL);
+                                         ypos, xpos, (JDIMENSION) blockcnt,
                                         NULL);
            if (blockcnt < compptr->MCU_width) {
              /* Create some dummy blocks at the right edge of the image. */
              jzero_far((void *) coef->MCU_buffer[blkn + blockcnt],
@@ -302,7 +303,8 @@ compress_first_pass (j_compress_ptr cinfo, JSAMPIMAGE input_buf)
      (*cinfo->fdct->forward_DCT) (cinfo, compptr,
                                   input_buf[ci], thisblockrow,
                                   (JDIMENSION) (block_row * DCTSIZE),
-				   (JDIMENSION) 0, blocks_across, buffer_dst[block_row]);
+                                   (JDIMENSION) 0, blocks_across,
                                   buffer_dst[block_row]);
      if (ndummy > 0) {
        /* Create dummy blocks at the right edge of the image. */
        thisblockrow += blocks_across; /* => first dummy block */
@@ -364,10 +366,23 @@ compress_trellis_pass (j_compress_ptr cinfo, JSAMPIMAGE input_buf)
    c_derived_tbl *dctbl = &dctbl_data;
    c_derived_tbl actbl_data;
    c_derived_tbl *actbl = &actbl_data;
 #ifdef C_ARITH_CODING_SUPPORTED
    arith_rates arith_r_data;
    arith_rates *arith_r = &arith_r_data;
 #endif
    compptr = cinfo->cur_comp_info[ci];
 #ifdef C_ARITH_CODING_SUPPORTED
    if (cinfo->arith_code)
      jget_arith_rates(cinfo, compptr->dc_tbl_no, compptr->ac_tbl_no, arith_r);
    else
 #endif
    {
      jpeg_make_c_derived_tbl(cinfo, TRUE, compptr->dc_tbl_no, &dctbl);
      jpeg_make_c_derived_tbl(cinfo, FALSE, compptr->ac_tbl_no, &actbl);
    }
    /* Align the virtual buffer for this component. */
    buffer = (*cinfo->mem->access_virt_barray)
@@ -402,12 +417,23 @@ compress_trellis_pass (j_compress_ptr cinfo, JSAMPIMAGE input_buf)
     */
    for (block_row = 0; block_row < block_rows; block_row++) {
      thisblockrow = buffer[block_row];
      lastblockrow = (block_row > 0) ? buffer[block_row-1] : NULL;
 #ifdef C_ARITH_CODING_SUPPORTED
      if (cinfo->arith_code)
        quantize_trellis_arith(cinfo, arith_r, thisblockrow,
                               buffer_dst[block_row], blocks_across,
                               cinfo->quant_tbl_ptrs[compptr->quant_tbl_no],
                               cinfo->master->norm_src[compptr->quant_tbl_no],
                               cinfo->master->norm_coef[compptr->quant_tbl_no],
                               &lastDC, lastblockrow, buffer_dst[block_row-1]);
      else
 #endif
        quantize_trellis(cinfo, dctbl, actbl, thisblockrow,
                         buffer_dst[block_row], blocks_across,
                         cinfo->quant_tbl_ptrs[compptr->quant_tbl_no],
                         cinfo->master->norm_src[compptr->quant_tbl_no],
                         cinfo->master->norm_coef[compptr->quant_tbl_no],
-                       &lastDC);
+                         &lastDC, lastblockrow, buffer_dst[block_row-1]);
      if (ndummy > 0) {
        /* Create dummy blocks at the right edge of the image. */
@@ -503,7 +529,6 @@ compress_output (j_compress_ptr cinfo, JSAMPIMAGE input_buf)
          }
        }
      }
      /* Try to write the MCU. */
      if (! (*cinfo->entropy->encode_mcu) (cinfo, coef->MCU_buffer)) {
        /* Suspension forced; update state counters and exit */
--- a/jcdctmgr.c
+++ b/jcdctmgr.c
@@ -22,9 +22,11 @@
 #include "jpeglib.h"
 #include "jdct.h"               /* Private declarations for DCT subsystem */
 #include "jsimddct.h"
 #include "jchuff.h"
 #include <assert.h>
 #include <math.h>
 /* Private subobject for this module */
 typedef void (*forward_DCT_method_ptr) (DCTELEM * data);
@@ -725,6 +727,17 @@ forward_DCT (j_compress_ptr cinfo, jpeg_component_info * compptr,
    /* Quantize/descale the coefficients, and store into coef_blocks[] */
    (*do_quantize) (coef_blocks[bi], divisors, workspace);
    if (do_preprocess) {
      int i;
      int maxval = (1 << MAX_COEF_BITS) - 1;
      for (i = 0; i < 64; i++) {
        if (coef_blocks[bi][i] < -maxval)
          coef_blocks[bi][i] = -maxval;
        if (coef_blocks[bi][i] > maxval)
          coef_blocks[bi][i] = maxval;
      }
    }
  }
 }
@@ -844,12 +857,22 @@ forward_DCT_float (j_compress_ptr cinfo, jpeg_component_info * compptr,
    /* Quantize/descale the coefficients, and store into coef_blocks[] */
    (*do_quantize) (coef_blocks[bi], divisors, workspace);
    if (do_preprocess) {
      int i;
      int maxval = (1 << MAX_COEF_BITS) - 1;
      for (i = 0; i < 64; i++) {
        if (coef_blocks[bi][i] < -maxval)
          coef_blocks[bi][i] = -maxval;
        if (coef_blocks[bi][i] > maxval)
          coef_blocks[bi][i] = maxval;
      }
    }
  }
 }
 #endif /* DCT_FLOAT_SUPPORTED */
 #include "jchuff.h"
 #include "jpeg_nbits_table.h"
 static const float jpeg_lambda_weights_flat[64] = {
@@ -874,11 +897,17 @@ static const float jpeg_lambda_weights_csf_luma[64] = {
  0.43454f, 0.42146f, 0.34609f, 0.24072f, 0.15975f, 0.10701f, 0.07558f, 0.05875f,
 };
-#define DC_TRELLIS_CANDIDATES 3
+#define DC_TRELLIS_MAX_CANDIDATES 9
 LOCAL(int) get_num_dc_trellis_candidates(int dc_quantval) {
  /* Higher qualities can tolerate higher DC distortion */
  return MIN(DC_TRELLIS_MAX_CANDIDATES, (2 + 60 / dc_quantval)|1);
 }
 GLOBAL(void)
 quantize_trellis(j_compress_ptr cinfo, c_derived_tbl *dctbl, c_derived_tbl *actbl, JBLOCKROW coef_blocks, JBLOCKROW src, JDIMENSION num_blocks,
-                 JQUANT_TBL * qtbl, double *norm_src, double *norm_coef, JCOEF *last_dc_val)
+                 JQUANT_TBL * qtbl, double *norm_src, double *norm_coef, JCOEF *last_dc_val,
                 JBLOCKROW coef_blocks_above, JBLOCKROW src_above)
 {
  int i, j, k, l;
  float accumulated_zero_dist[DCTSIZE2];
@@ -906,9 +935,12 @@ quantize_trellis(j_compress_ptr cinfo, c_derived_tbl *dctbl, c_derived_tbl *actb
  int zero_run;
  int run_bits;
  int rate;
-  float *accumulated_dc_cost[DC_TRELLIS_CANDIDATES];
+  float *accumulated_dc_cost[DC_TRELLIS_MAX_CANDIDATES];
-  int *dc_cost_backtrack[DC_TRELLIS_CANDIDATES];
+  int *dc_cost_backtrack[DC_TRELLIS_MAX_CANDIDATES];
-  JCOEF *dc_candidate[DC_TRELLIS_CANDIDATES];
+  JCOEF *dc_candidate[DC_TRELLIS_MAX_CANDIDATES];
  int mode = 1;
  float lambda_table[DCTSIZE2];
  const int dc_trellis_candidates = get_num_dc_trellis_candidates(qtbl->quantval[0]);
  Ss = cinfo->Ss;
  Se = cinfo->Se;
@@ -932,8 +964,9 @@ quantize_trellis(j_compress_ptr cinfo, c_derived_tbl *dctbl, c_derived_tbl *actb
    accumulated_block_cost[0] = 0;
    requires_eob[0] = 0;
  }
  if (cinfo->master->trellis_quant_dc) {
-    for (i = 0; i < DC_TRELLIS_CANDIDATES; i++) {
+    for (i = 0; i < dc_trellis_candidates; i++) {
      accumulated_dc_cost[i] = (float *)malloc(num_blocks * sizeof(float));
      dc_cost_backtrack[i] = (int *)malloc(num_blocks * sizeof(int));
      dc_candidate[i] = (JCOEF *)malloc(num_blocks * sizeof(JCOEF));
@@ -944,12 +977,19 @@ quantize_trellis(j_compress_ptr cinfo, c_derived_tbl *dctbl, c_derived_tbl *actb
      }
    }
  }
  norm = 0.0;
  for (i = 1; i < DCTSIZE2; i++) {
    norm += qtbl->quantval[i] * qtbl->quantval[i];
  }
  norm /= 63.0;
  if (mode == 1) {
    lambda_base = 1.0;
    lambda_tbl = lambda_table;
    for (i = 0; i < DCTSIZE2; i++)
      lambda_table[i] = 1.0 / (qtbl->quantval[i] * qtbl->quantval[i]);
  } else
    lambda_base = 1.0 / norm;
  for (bi = 0; bi < num_blocks; bi++) {
@@ -980,16 +1020,39 @@ quantize_trellis(j_compress_ptr cinfo, c_derived_tbl *dctbl, c_derived_tbl *actb
      float dc_candidate_dist;
      qval = (x + q/2) / q; /* quantized value (round nearest) */
-      for (k = 0; k < DC_TRELLIS_CANDIDATES; k++) {
+      for (k = 0; k < dc_trellis_candidates; k++) {
        int delta;
        int dc_delta;
        int bits;
-        dc_candidate[k][bi] = qval - DC_TRELLIS_CANDIDATES/2 + k;
+        dc_candidate[k][bi] = qval - dc_trellis_candidates/2 + k;
        if (dc_candidate[k][bi] >= (1<<MAX_COEF_BITS))
          dc_candidate[k][bi] = (1<<MAX_COEF_BITS)-1;
        if (dc_candidate[k][bi] <= -(1<<MAX_COEF_BITS))
          dc_candidate[k][bi] = -(1<<MAX_COEF_BITS)+1;
        delta = dc_candidate[k][bi] * q - x;
        dc_candidate_dist = delta * delta * lambda_dc;
        dc_candidate[k][bi] *= 1 + 2*sign;
        /* Take into account DC differences */
        if (coef_blocks_above && src_above && cinfo->master->trellis_delta_dc_weight > 0.0) {
          int dc_above_orig;
          int dc_above_recon;
          int dc_orig;
          int dc_recon;
          float vertical_dist;
          dc_above_orig = src_above[bi][0];
          dc_above_recon = coef_blocks_above[bi][0] * q;
          dc_orig = src[bi][0];
          dc_recon = dc_candidate[k][bi] * q;
          /* delta is difference of vertical gradients */
          delta = (dc_above_orig - dc_orig) - (dc_above_recon - dc_recon);
          vertical_dist = delta * delta * lambda_dc;
          dc_candidate_dist +=  cinfo->master->trellis_delta_dc_weight * (vertical_dist - dc_candidate_dist);
        }
        if (bi == 0) {
          dc_delta = dc_candidate[k][bi] - *last_dc_val;
@@ -1004,7 +1067,7 @@ quantize_trellis(j_compress_ptr cinfo, c_derived_tbl *dctbl, c_derived_tbl *actb
          accumulated_dc_cost[k][0] = cost;
          dc_cost_backtrack[k][0] = -1;
        } else {
-          for (l = 0; l < DC_TRELLIS_CANDIDATES; l++) {
+          for (l = 0; l < dc_trellis_candidates; l++) {
            dc_delta = dc_candidate[k][bi] - dc_candidate[l][bi-1];
            /* Derive number of suffix bits */
@@ -1047,6 +1110,9 @@ quantize_trellis(j_compress_ptr cinfo, c_derived_tbl *dctbl, c_derived_tbl *actb
        continue;
      }
      if (qval >= (1<<MAX_COEF_BITS))
        qval = (1<<MAX_COEF_BITS)-1;
      num_candidates = jpeg_nbits_table[qval];
      for (k = 0; k < num_candidates; k++) {
        int delta;
@@ -1211,7 +1277,7 @@ quantize_trellis(j_compress_ptr cinfo, c_derived_tbl *dctbl, c_derived_tbl *actb
  if (cinfo->master->trellis_quant_dc) {
    j = 0;
-    for (i = 1; i < DC_TRELLIS_CANDIDATES; i++) {
+    for (i = 1; i < dc_trellis_candidates; i++) {
      if (accumulated_dc_cost[i][num_blocks-1] < accumulated_dc_cost[j][num_blocks-1])
        j = i;
    }
@@ -1223,7 +1289,7 @@ quantize_trellis(j_compress_ptr cinfo, c_derived_tbl *dctbl, c_derived_tbl *actb
    /* Save DC predictor */
    *last_dc_val = coef_blocks[num_blocks-1][0];
-    for (i = 0; i < DC_TRELLIS_CANDIDATES; i++) {
+    for (i = 0; i < dc_trellis_candidates; i++) {
      free(accumulated_dc_cost[i]);
      free(dc_cost_backtrack[i]);
      free(dc_candidate[i]);
@@ -1232,6 +1298,344 @@ quantize_trellis(j_compress_ptr cinfo, c_derived_tbl *dctbl, c_derived_tbl *actb
 }
 #ifdef C_ARITH_CODING_SUPPORTED
 GLOBAL(void)
 quantize_trellis_arith(j_compress_ptr cinfo, arith_rates *r, JBLOCKROW coef_blocks, JBLOCKROW src, JDIMENSION num_blocks,
                 JQUANT_TBL * qtbl, double *norm_src, double *norm_coef, JCOEF *last_dc_val,
                 JBLOCKROW coef_blocks_above, JBLOCKROW src_above)
 {
  int i, j, k, l;
  float accumulated_zero_dist[DCTSIZE2];
  float accumulated_cost[DCTSIZE2];
  int run_start[DCTSIZE2];
  int bi;
  float best_cost;
  int last_coeff_idx; /* position of last nonzero coefficient */
  float norm = 0.0;
  float lambda_base;
  float lambda;
  float lambda_dc;
  const float *lambda_tbl = (cinfo->master->use_lambda_weight_tbl) ?
  jpeg_lambda_weights_csf_luma :
  jpeg_lambda_weights_flat;
  int Ss, Se;
  float cost;
  float run_bits;
  int rate;
  float *accumulated_dc_cost[DC_TRELLIS_MAX_CANDIDATES];
  int *dc_cost_backtrack[DC_TRELLIS_MAX_CANDIDATES];
  JCOEF *dc_candidate[DC_TRELLIS_MAX_CANDIDATES];
  int *dc_context[DC_TRELLIS_MAX_CANDIDATES];
  int mode = 1;
  float lambda_table[DCTSIZE2];
  const int dc_trellis_candidates = get_num_dc_trellis_candidates(qtbl->quantval[0]);
  Ss = cinfo->Ss;
  Se = cinfo->Se;
  if (Ss == 0)
    Ss = 1;
  if (Se < Ss)
    return;
  if (cinfo->master->trellis_quant_dc) {
    for (i = 0; i < dc_trellis_candidates; i++) {
      accumulated_dc_cost[i] = (float *)malloc(num_blocks * sizeof(float));
      dc_cost_backtrack[i] = (int *)malloc(num_blocks * sizeof(int));
      dc_candidate[i] = (JCOEF *)malloc(num_blocks * sizeof(JCOEF));
      dc_context[i] = (int *)malloc(num_blocks * sizeof(int));
      if (!accumulated_dc_cost[i] ||
          !dc_cost_backtrack[i] ||
          !dc_candidate[i] ||
          !dc_context[i]) {
        ERREXIT(cinfo, JERR_OUT_OF_MEMORY);
      }
    }
  }
  norm = 0.0;
  for (i = 1; i < DCTSIZE2; i++) {
    norm += qtbl->quantval[i] * qtbl->quantval[i];
  }
  norm /= 63.0;
  if (mode == 1) {
    lambda_base = 1.0;
    lambda_tbl = lambda_table;
    for (i = 0; i < DCTSIZE2; i++)
      lambda_table[i] = 1.0 / (qtbl->quantval[i] * qtbl->quantval[i]);
  } else
    lambda_base = 1.0 / norm;
  for (bi = 0; bi < num_blocks; bi++) {
    norm = 0.0;
    for (i = 1; i < DCTSIZE2; i++) {
      norm += src[bi][i] * src[bi][i];
    }
    norm /= 63.0;
    if (cinfo->master->lambda_log_scale2 > 0.0)
      lambda = pow(2.0, cinfo->master->lambda_log_scale1) * lambda_base /
      (pow(2.0, cinfo->master->lambda_log_scale2) + norm);
    else
      lambda = pow(2.0, cinfo->master->lambda_log_scale1 - 12.0) * lambda_base;
    lambda_dc = lambda * lambda_tbl[0];
    accumulated_zero_dist[Ss-1] = 0.0;
    accumulated_cost[Ss-1] = 0.0;
    /* Do DC coefficient */
    if (cinfo->master->trellis_quant_dc) {
      int sign = src[bi][0] >> 31;
      int x = abs(src[bi][0]);
      int q = 8 * qtbl->quantval[0];
      int qval;
      float dc_candidate_dist;
      qval = (x + q/2) / q; /* quantized value (round nearest) */
      /* loop over candidates in current block */
      for (k = 0; k < dc_trellis_candidates; k++) {
        int delta;
        int dc_delta;
        float bits;
        int m;
        int v2;
        dc_candidate[k][bi] = qval - dc_trellis_candidates/2 + k;
        delta = dc_candidate[k][bi] * q - x;
        dc_candidate_dist = delta * delta * lambda_dc;
        dc_candidate[k][bi] *= 1 + 2*sign;
        /* Take into account DC differences */
        if (coef_blocks_above && src_above && cinfo->master->trellis_delta_dc_weight > 0.0) {
          int dc_above_orig;
          int dc_above_recon;
          int dc_orig;
          int dc_recon;
          float vertical_dist;
          dc_above_orig = src_above[bi][0];
          dc_above_recon = coef_blocks_above[bi][0] * q;
          dc_orig = src[bi][0];
          dc_recon = dc_candidate[k][bi] * q;
          /* delta is difference of vertical gradients */
          delta = (dc_above_orig - dc_orig) - (dc_above_recon - dc_recon);
          vertical_dist = delta * delta * lambda_dc;
          dc_candidate_dist +=  cinfo->master->trellis_delta_dc_weight * (vertical_dist - dc_candidate_dist);
        }
        /* loop of candidates from previous block */
        for (l = 0; l < (bi == 0 ? 1 : dc_trellis_candidates); l++) {
          int dc_pred = (bi == 0 ? *last_dc_val : dc_candidate[l][bi-1]);
          int updated_dc_context = 0;
          int st = (bi == 0) ? 0 : dc_context[l][bi-1];
          dc_delta = dc_candidate[k][bi] - dc_pred;
          bits = r->rate_dc[st][dc_delta != 0];
          if (dc_delta != 0) {
            bits += r->rate_dc[st+1][dc_delta < 0];
            st += 2 + (dc_delta < 0);
            updated_dc_context = (dc_delta < 0) ? 8 : 4;
            dc_delta = abs(dc_delta);
            m = 0;
            if (dc_delta -= 1) {
              bits += r->rate_dc[st][1];
              st = 20;
              m = 1;
              v2 = dc_delta;
              while (v2 >>= 1) {
                bits += r->rate_dc[st][1];
                m <<= 1;
                st++;
              }
            }
            bits += r->rate_dc[st][0];
            if (m < (int) ((1L << r->arith_dc_L) >> 1))
              updated_dc_context = 0;    /* zero diff category */
            else if (m > (int) ((1L << r->arith_dc_U) >> 1))
              updated_dc_context += 8;   /* large diff category */
            st += 14;
            while (m >>= 1)
              bits += r->rate_dc[st][(m & dc_delta) ? 1 : 0];
          }
          cost = bits + dc_candidate_dist;
          if (bi != 0)
            cost += accumulated_dc_cost[l][bi-1];
          if (l == 0 || cost < accumulated_dc_cost[k][bi]) {
            accumulated_dc_cost[k][bi] = cost;
            dc_cost_backtrack[k][bi] = (bi == 0 ? -1 : l);
            dc_context[k][bi] = updated_dc_context;
          }
        }
      }
    }
    /* Do AC coefficients */
    for (i = Ss; i <= Se; i++) {
      int z = jpeg_natural_order[i];
      int sign = src[bi][z] >> 31;
      int x = abs(src[bi][z]);
      int q = 8 * qtbl->quantval[z];
      int candidate[16];
      float candidate_dist[16];
      int num_candidates;
      int qval;
      int delta;
      accumulated_zero_dist[i] = x * x * lambda * lambda_tbl[z] + accumulated_zero_dist[i-1];
      qval = (x + q/2) / q; /* quantized value (round nearest) */
      if (qval == 0) {
        coef_blocks[bi][z] = 0;
        accumulated_cost[i] = 1e38; /* Shouldn't be needed */
        continue;
      }
      k = 0;
      candidate[k] = qval;
      delta = candidate[k] * q - x;
      candidate_dist[k] = delta * delta * lambda * lambda_tbl[z];
      k++;
      if (qval > 1) {
        candidate[k] = qval - 1;
        delta = candidate[k] * q - x;
        candidate_dist[k] = delta * delta * lambda * lambda_tbl[z];
        k++;
      }
      num_candidates = k;
      accumulated_cost[i] = 1e38;
      for (j = Ss-1; j < i; j++) {
        int zz = jpeg_natural_order[j];
        if (j != Ss-1 && coef_blocks[bi][zz] == 0)
          continue;
        run_bits = r->rate_ac[3*j][0]; /* EOB */
        for (k = j+1; k < i; k++)
          run_bits += r->rate_ac[3*(k-1)+1][0];
        run_bits += r->rate_ac[3*(i-1)+1][1];
        for (k = 0; k < num_candidates; k++) {
          float coef_bits = 1.0; /* sign bit */
          int v = candidate[k];
          int v2;
          int m;
          int st;
          st = 3*(i-1)+2;
          m = 0;
          if (v -= 1) {
            coef_bits += r->rate_ac[st][1];
            m = 1;
            v2 = v;
            if (v2 >>= 1) {
              coef_bits += r->rate_ac[st][1];
              m <<= 1;
              st = (i <= r->arith_ac_K) ? 189 : 217;
              while (v2 >>= 1) {
                coef_bits += r->rate_ac[st][1];
                m <<= 1;
                st++;
              }
            }
          }
          coef_bits += r->rate_ac[st][0];
          st += 14;
          while (m >>= 1)
            coef_bits += r->rate_ac[st][(m & v) ? 1 : 0];
          rate = coef_bits + run_bits;
          cost = rate + candidate_dist[k];
          cost += accumulated_zero_dist[i-1] - accumulated_zero_dist[j] + accumulated_cost[j];
          if (cost < accumulated_cost[i]) {
            coef_blocks[bi][z] = (candidate[k] ^ sign) - sign;
            accumulated_cost[i] = cost;
            run_start[i] = j;
          }
        }
      }
    }
    last_coeff_idx = Ss-1;
    best_cost = accumulated_zero_dist[Se] + r->rate_ac[0][1];
    for (i = Ss; i <= Se; i++) {
      int z = jpeg_natural_order[i];
      if (coef_blocks[bi][z] != 0) {
        float cost = accumulated_cost[i] + accumulated_zero_dist[Se] - accumulated_zero_dist[i];
        if (i < Se)
          cost += r->rate_ac[3*(i-1)][1];
        if (cost < best_cost) {
          best_cost = cost;
          last_coeff_idx = i;
        }
      }
    }
    /* Zero out coefficients that are part of runs */
    i = Se;
    while (i >= Ss)
    {
      while (i > last_coeff_idx) {
        int z = jpeg_natural_order[i];
        coef_blocks[bi][z] = 0;
        i--;
      }
      last_coeff_idx = run_start[i];
      i--;
    }
  }
  if (cinfo->master->trellis_q_opt) {
    for (bi = 0; bi < num_blocks; bi++) {
      for (i = 1; i < DCTSIZE2; i++) {
        norm_src[i] += src[bi][i] * coef_blocks[bi][i];
        norm_coef[i] += 8 * coef_blocks[bi][i] * coef_blocks[bi][i];
      }
    }
  }
  if (cinfo->master->trellis_quant_dc) {
    j = 0;
    for (i = 1; i < dc_trellis_candidates; i++) {
      if (accumulated_dc_cost[i][num_blocks-1] < accumulated_dc_cost[j][num_blocks-1])
        j = i;
    }
    for (bi = num_blocks-1; bi >= 0; bi--) {
      coef_blocks[bi][0] = dc_candidate[j][bi];
      j = dc_cost_backtrack[j][bi];
    }
    /* Save DC predictor */
    *last_dc_val = coef_blocks[num_blocks-1][0];
    for (i = 0; i < dc_trellis_candidates; i++) {
      free(accumulated_dc_cost[i]);
      free(dc_cost_backtrack[i]);
      free(dc_candidate[i]);
      free(dc_context[i]);
    }
  }
 }
 #endif
 /*
 * Initialize FDCT manager.
 */
--- a/jccompat.c
+++ b/jccompat.c
@@ -1,5 +1,5 @@
 /*
- * jccompat.c
+ * jcext.c
 *
 * Copyright (C) 2014, D. R. Commander.
 * Copyright (C) 2014, Mozilla Corporation.
@@ -16,20 +16,15 @@
 GLOBAL(boolean)
-jpeg_c_bool_param_supported (j_compress_ptr cinfo, J_BOOLEAN_PARAM param)
+jpeg_c_bool_param_supported (const j_compress_ptr cinfo, J_BOOLEAN_PARAM param)
 {
  switch (param) {
  case JBOOLEAN_USE_MOZ_DEFAULTS:
  case JBOOLEAN_OPTIMIZE_SCANS:
  case JBOOLEAN_ONE_DC_SCAN:
  case JBOOLEAN_SEP_DC_SCAN:
  case JBOOLEAN_TRELLIS_QUANT:
  case JBOOLEAN_TRELLIS_QUANT_DC:
  case JBOOLEAN_TRELLIS_EOB_OPT:
  case JBOOLEAN_USE_FLAT_QUANT_TBL:
  case JBOOLEAN_USE_LAMBDA_WEIGHT_TBL:
  case JBOOLEAN_USE_SCANS_IN_TRELLIS:
  case JBOOLEAN_TRELLIS_PASSES:
  case JBOOLEAN_TRELLIS_Q_OPT:
  case JBOOLEAN_OVERSHOOT_DERINGING:
    return TRUE;
@@ -44,18 +39,9 @@ jpeg_c_set_bool_param (j_compress_ptr cinfo, J_BOOLEAN_PARAM param,
                       boolean value)
 {
  switch(param) {
  case JBOOLEAN_USE_MOZ_DEFAULTS:
    cinfo->master->use_moz_defaults = value;
    break;
  case JBOOLEAN_OPTIMIZE_SCANS:
    cinfo->master->optimize_scans = value;
    break;
  case JBOOLEAN_ONE_DC_SCAN:
    cinfo->master->one_dc_scan = value;
    break;
  case JBOOLEAN_SEP_DC_SCAN:
    cinfo->master->sep_dc_scan = value;
    break;
  case JBOOLEAN_TRELLIS_QUANT:
    cinfo->master->trellis_quant = value;
    break;
@@ -65,18 +51,12 @@ jpeg_c_set_bool_param (j_compress_ptr cinfo, J_BOOLEAN_PARAM param,
  case JBOOLEAN_TRELLIS_EOB_OPT:
    cinfo->master->trellis_eob_opt = value;
    break;
  case JBOOLEAN_USE_FLAT_QUANT_TBL:
    cinfo->master->use_flat_quant_tbl = value;
    break;
  case JBOOLEAN_USE_LAMBDA_WEIGHT_TBL:
    cinfo->master->use_lambda_weight_tbl = value;
    break;
  case JBOOLEAN_USE_SCANS_IN_TRELLIS:
    cinfo->master->use_scans_in_trellis = value;
    break;
  case JBOOLEAN_TRELLIS_PASSES:
    cinfo->master->trellis_passes = value;
    break;
  case JBOOLEAN_TRELLIS_Q_OPT:
    cinfo->master->trellis_q_opt = value;
    break;
@@ -90,31 +70,21 @@ jpeg_c_set_bool_param (j_compress_ptr cinfo, J_BOOLEAN_PARAM param,
 GLOBAL(boolean)
-jpeg_c_get_bool_param (j_compress_ptr cinfo, J_BOOLEAN_PARAM param)
+jpeg_c_get_bool_param (const j_compress_ptr cinfo, J_BOOLEAN_PARAM param)
 {
  switch(param) {
  case JBOOLEAN_USE_MOZ_DEFAULTS:
    return cinfo->master->use_moz_defaults;
  case JBOOLEAN_OPTIMIZE_SCANS:
    return cinfo->master->optimize_scans;
  case JBOOLEAN_ONE_DC_SCAN:
    return cinfo->master->one_dc_scan;
  case JBOOLEAN_SEP_DC_SCAN:
    return cinfo->master->sep_dc_scan;
  case JBOOLEAN_TRELLIS_QUANT:
    return cinfo->master->trellis_quant;
  case JBOOLEAN_TRELLIS_QUANT_DC:
    return cinfo->master->trellis_quant_dc;
  case JBOOLEAN_TRELLIS_EOB_OPT:
    return cinfo->master->trellis_eob_opt;
  case JBOOLEAN_USE_FLAT_QUANT_TBL:
    return cinfo->master->use_flat_quant_tbl;
  case JBOOLEAN_USE_LAMBDA_WEIGHT_TBL:
    return cinfo->master->use_lambda_weight_tbl;
  case JBOOLEAN_USE_SCANS_IN_TRELLIS:
    return cinfo->master->use_scans_in_trellis;
  case JBOOLEAN_TRELLIS_PASSES:
    return cinfo->master->trellis_passes;
  case JBOOLEAN_TRELLIS_Q_OPT:
    return cinfo->master->trellis_q_opt;
  case JBOOLEAN_OVERSHOOT_DERINGING:
@@ -128,11 +98,12 @@ jpeg_c_get_bool_param (j_compress_ptr cinfo, J_BOOLEAN_PARAM param)
 GLOBAL(boolean)
-jpeg_c_float_param_supported (j_compress_ptr cinfo, J_FLOAT_PARAM param)
+jpeg_c_float_param_supported (const j_compress_ptr cinfo, J_FLOAT_PARAM param)
 {
  switch (param) {
  case JFLOAT_LAMBDA_LOG_SCALE1:
  case JFLOAT_LAMBDA_LOG_SCALE2:
  case JFLOAT_TRELLIS_DELTA_DC_WEIGHT:
    return TRUE;
  }
@@ -150,6 +121,9 @@ jpeg_c_set_float_param (j_compress_ptr cinfo, J_FLOAT_PARAM param, float value)
  case JFLOAT_LAMBDA_LOG_SCALE2:
    cinfo->master->lambda_log_scale2 = value;
    break;
  case JFLOAT_TRELLIS_DELTA_DC_WEIGHT:
    cinfo->master->trellis_delta_dc_weight = value;
    break;
  default:
    ERREXIT(cinfo, JERR_BAD_PARAM);
  }
@@ -157,13 +131,15 @@ jpeg_c_set_float_param (j_compress_ptr cinfo, J_FLOAT_PARAM param, float value)
 GLOBAL(float)
-jpeg_c_get_float_param (j_compress_ptr cinfo, J_FLOAT_PARAM param)
+jpeg_c_get_float_param (const j_compress_ptr cinfo, J_FLOAT_PARAM param)
 {
  switch (param) {
  case JFLOAT_LAMBDA_LOG_SCALE1:
    return cinfo->master->lambda_log_scale1;
  case JFLOAT_LAMBDA_LOG_SCALE2:
    return cinfo->master->lambda_log_scale2;
  case JFLOAT_TRELLIS_DELTA_DC_WEIGHT:
    return cinfo->master->trellis_delta_dc_weight;
  default:
    ERREXIT(cinfo, JERR_BAD_PARAM);
  }
@@ -173,11 +149,14 @@ jpeg_c_get_float_param (j_compress_ptr cinfo, J_FLOAT_PARAM param)
 GLOBAL(boolean)
-jpeg_c_int_param_supported (j_compress_ptr cinfo, J_INT_PARAM param)
+jpeg_c_int_param_supported (const j_compress_ptr cinfo, J_INT_PARAM param)
 {
  switch (param) {
  case JINT_COMPRESS_PROFILE:
  case JINT_TRELLIS_FREQ_SPLIT:
  case JINT_TRELLIS_NUM_LOOPS:
  case JINT_BASE_QUANT_TBL_IDX:
  case JINT_DC_SCAN_OPT_MODE:
    return TRUE;
  }
@@ -189,12 +168,29 @@ GLOBAL(void)
 jpeg_c_set_int_param (j_compress_ptr cinfo, J_INT_PARAM param, int value)
 {
  switch (param) {
  case JINT_COMPRESS_PROFILE:
    switch (value) {
    case JCP_MAX_COMPRESSION:
    case JCP_FASTEST:
      cinfo->master->compress_profile = value;
      break;
    default:
      ERREXIT(cinfo, JERR_BAD_PARAM_VALUE);
    }
    break;
  case JINT_TRELLIS_FREQ_SPLIT:
    cinfo->master->trellis_freq_split = value;
    break;
  case JINT_TRELLIS_NUM_LOOPS:
    cinfo->master->trellis_num_loops = value;
    break;
  case JINT_BASE_QUANT_TBL_IDX:
    if (value >= 0 && value <= 8)
      cinfo->master->quant_tbl_master_idx = value;
    break;
  case JINT_DC_SCAN_OPT_MODE:
    cinfo->master->dc_scan_opt_mode = value;
    break;
  default:
    ERREXIT(cinfo, JERR_BAD_PARAM);
  }
@@ -202,13 +198,19 @@ jpeg_c_set_int_param (j_compress_ptr cinfo, J_INT_PARAM param, int value)
 GLOBAL(int)
-jpeg_c_get_int_param (j_compress_ptr cinfo, J_INT_PARAM param)
+jpeg_c_get_int_param (const j_compress_ptr cinfo, J_INT_PARAM param)
 {
  switch (param) {
  case JINT_COMPRESS_PROFILE:
    return cinfo->master->compress_profile;
  case JINT_TRELLIS_FREQ_SPLIT:
    return cinfo->master->trellis_freq_split;
  case JINT_TRELLIS_NUM_LOOPS:
    return cinfo->master->trellis_num_loops;
  case JINT_BASE_QUANT_TBL_IDX:
    return cinfo->master->quant_tbl_master_idx;
  case JINT_DC_SCAN_OPT_MODE:
    return cinfo->master->dc_scan_opt_mode;
  default:
    ERREXIT(cinfo, JERR_BAD_PARAM);
  }
--- a/jchuff.c
+++ b/jchuff.c
@@ -37,7 +37,7 @@
 */
 /* NOTE: Both GCC and Clang define __GNUC__ */
-#if defined __GNUC__ && defined __arm__
+#if defined __GNUC__ && (defined __arm__ || defined __aarch64__)
 #if !defined __thumb__ || defined __thumb2__
 #define USE_CLZ_INTRINSIC
 #endif
@@ -408,7 +408,16 @@ dump_buffer (working_state * state)
 #endif
-#define BUFSIZE (DCTSIZE2 * 2) + 8
+/* Although it is exceedingly rare, it is possible for a Huffman-encoded
 * coefficient block to be larger than the 128-byte unencoded block.  For each
 * of the 64 coefficients, PUT_BITS is invoked twice, and each invocation can
 * theoretically store 16 bits (for a maximum of 2048 bits or 256 bytes per
 * encoded block.)  If, for instance, one artificially sets the AC
 * coefficients to alternating values of 32767 and -32768 (using the JPEG
 * scanning order-- 1, 8, 16, etc.), then this will produce an encoded block
 * larger than 200 bytes.
 */
 #define BUFSIZE (DCTSIZE2 * 4)
 #define LOAD_BUFFER() { \
  if (state->free_in_buffer < BUFSIZE) { \
--- a/jchuff.h
+++ b/jchuff.h
@@ -45,4 +45,5 @@ EXTERN(void) jpeg_gen_optimal_table
 EXTERN(void) quantize_trellis
        (j_compress_ptr cinfo, c_derived_tbl *dctbl, c_derived_tbl *actbl, JBLOCKROW coef_blocks, JBLOCKROW src, JDIMENSION num_blocks,
-                 JQUANT_TBL * qtbl, double *norm_src, double *norm_coef, JCOEF *last_dc_val);
+                 JQUANT_TBL * qtbl, double *norm_src, double *norm_coef, JCOEF *last_dc_val,
         JBLOCKROW coef_blocks_above, JBLOCKROW src_above);
--- a/jcinit.c
+++ b/jcinit.c
@@ -63,7 +63,7 @@ jinit_compress_master (j_compress_ptr cinfo)
  /* Need a full-image coefficient buffer in any multi-pass mode. */
  jinit_c_coef_controller(cinfo,
    (boolean) (cinfo->num_scans > 1 || cinfo->optimize_coding ||
-               cinfo->master->optimize_scans));
+               cinfo->master->optimize_scans || cinfo->master->trellis_quant));
  jinit_c_main_controller(cinfo, FALSE /* never need full buffer here */);
  jinit_marker_writer(cinfo);
--- a/jcmainct.c
+++ b/jcmainct.c
@@ -17,14 +17,6 @@
 #include "jpeglib.h"
 /* Note: currently, there is no operating mode in which a full-image buffer
 * is needed at this step.  If there were, that mode could not be used with
 * "raw data" input, since this module is bypassed in that case.  However,
 * we've left the code here for possible use in special applications.
 */
 #undef FULL_MAIN_BUFFER_SUPPORTED
 /* Private buffer controller object */
 typedef struct {
@@ -40,13 +32,6 @@ typedef struct {
   * points to the currently accessible strips of the virtual arrays.
   */
  JSAMPARRAY buffer[MAX_COMPONENTS];
 #ifdef FULL_MAIN_BUFFER_SUPPORTED
  /* If using full-image storage, this array holds pointers to virtual-array
   * control blocks for each component.  Unused if not full-image storage.
   */
  jvirt_sarray_ptr whole_image[MAX_COMPONENTS];
 #endif
 } my_main_controller;
 typedef my_main_controller * my_main_ptr;
@@ -56,11 +41,6 @@ typedef my_main_controller * my_main_ptr;
 METHODDEF(void) process_data_simple_main
        (j_compress_ptr cinfo, JSAMPARRAY input_buf, JDIMENSION *in_row_ctr,
         JDIMENSION in_rows_avail);
 #ifdef FULL_MAIN_BUFFER_SUPPORTED
 METHODDEF(void) process_data_buffer_main
        (j_compress_ptr cinfo, JSAMPARRAY input_buf, JDIMENSION *in_row_ctr,
         JDIMENSION in_rows_avail);
 #endif
 /*
@@ -76,32 +56,14 @@ start_pass_main (j_compress_ptr cinfo, J_BUF_MODE pass_mode)
  if (cinfo->raw_data_in)
    return;
  if (pass_mode != JBUF_PASS_THRU)
    ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
  main_ptr->cur_iMCU_row = 0;   /* initialize counters */
  main_ptr->rowgroup_ctr = 0;
  main_ptr->suspended = FALSE;
  main_ptr->pass_mode = pass_mode;      /* save mode for use by process_data */
  switch (pass_mode) {
  case JBUF_PASS_THRU:
 #ifdef FULL_MAIN_BUFFER_SUPPORTED
    if (main_ptr->whole_image[0] != NULL)
      ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
 #endif
  main_ptr->pub.process_data = process_data_simple_main;
    break;
 #ifdef FULL_MAIN_BUFFER_SUPPORTED
  case JBUF_SAVE_SOURCE:
  case JBUF_CRANK_DEST:
  case JBUF_SAVE_AND_PASS:
    if (main_ptr->whole_image[0] == NULL)
      ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
    main_ptr->pub.process_data = process_data_buffer_main;
    break;
 #endif
  default:
    ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
    break;
  }
 }
@@ -160,85 +122,6 @@ process_data_simple_main (j_compress_ptr cinfo,
 }
 #ifdef FULL_MAIN_BUFFER_SUPPORTED
 /*
 * Process some data.
 * This routine handles all of the modes that use a full-size buffer.
 */
 METHODDEF(void)
 process_data_buffer_main (j_compress_ptr cinfo,
                          JSAMPARRAY input_buf, JDIMENSION *in_row_ctr,
                          JDIMENSION in_rows_avail)
 {
  my_main_ptr main_ptr = (my_main_ptr) cinfo->main;
  int ci;
  jpeg_component_info *compptr;
  boolean writing = (main_ptr->pass_mode != JBUF_CRANK_DEST);
  while (main_ptr->cur_iMCU_row < cinfo->total_iMCU_rows) {
    /* Realign the virtual buffers if at the start of an iMCU row. */
    if (main_ptr->rowgroup_ctr == 0) {
      for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
           ci++, compptr++) {
        main_ptr->buffer[ci] = (*cinfo->mem->access_virt_sarray)
          ((j_common_ptr) cinfo, main_ptr->whole_image[ci],
           main_ptr->cur_iMCU_row * (compptr->v_samp_factor * DCTSIZE),
           (JDIMENSION) (compptr->v_samp_factor * DCTSIZE), writing);
      }
      /* In a read pass, pretend we just read some source data. */
      if (! writing) {
        *in_row_ctr += cinfo->max_v_samp_factor * DCTSIZE;
        main_ptr->rowgroup_ctr = DCTSIZE;
      }
    }
    /* If a write pass, read input data until the current iMCU row is full. */
    /* Note: preprocessor will pad if necessary to fill the last iMCU row. */
    if (writing) {
      (*cinfo->prep->pre_process_data) (cinfo,
                                        input_buf, in_row_ctr, in_rows_avail,
                                        main_ptr->buffer, &main_ptr->rowgroup_ctr,
                                        (JDIMENSION) DCTSIZE);
      /* Return to application if we need more data to fill the iMCU row. */
      if (main_ptr->rowgroup_ctr < DCTSIZE)
        return;
    }
    /* Emit data, unless this is a sink-only pass. */
    if (main_ptr->pass_mode != JBUF_SAVE_SOURCE) {
      if (! (*cinfo->coef->compress_data) (cinfo, main_ptr->buffer)) {
        /* If compressor did not consume the whole row, then we must need to
         * suspend processing and return to the application.  In this situation
         * we pretend we didn't yet consume the last input row; otherwise, if
         * it happened to be the last row of the image, the application would
         * think we were done.
         */
        if (! main_ptr->suspended) {
          (*in_row_ctr)--;
          main_ptr->suspended = TRUE;
        }
        return;
      }
      /* We did finish the row.  Undo our little suspension hack if a previous
       * call suspended; then mark the main buffer empty.
       */
      if (main_ptr->suspended) {
        (*in_row_ctr)++;
        main_ptr->suspended = FALSE;
      }
    }
    /* If get here, we are done with this iMCU row.  Mark buffer empty. */
    main_ptr->rowgroup_ctr = 0;
    main_ptr->cur_iMCU_row++;
  }
 }
 #endif /* FULL_MAIN_BUFFER_SUPPORTED */
 /*
 * Initialize main buffer controller.
 */
@@ -264,25 +147,8 @@ jinit_c_main_controller (j_compress_ptr cinfo, boolean need_full_buffer)
   * may be of a different size.
   */
  if (need_full_buffer) {
 #ifdef FULL_MAIN_BUFFER_SUPPORTED
    /* Allocate a full-image virtual array for each component */
    /* Note we pad the bottom to a multiple of the iMCU height */
    for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
         ci++, compptr++) {
      main_ptr->whole_image[ci] = (*cinfo->mem->request_virt_sarray)
        ((j_common_ptr) cinfo, JPOOL_IMAGE, FALSE,
         compptr->width_in_blocks * DCTSIZE,
         (JDIMENSION) jround_up((long) compptr->height_in_blocks,
                                (long) compptr->v_samp_factor) * DCTSIZE,
         (JDIMENSION) (compptr->v_samp_factor * DCTSIZE));
    }
 #else
    ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
 #endif
  } else {
 #ifdef FULL_MAIN_BUFFER_SUPPORTED
    main_ptr->whole_image[0] = NULL; /* flag for no virtual arrays */
 #endif
    /* Allocate a strip buffer for each component */
    for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
         ci++, compptr++) {
--- a/jcmarker.c
+++ b/jcmarker.c
@@ -194,6 +194,9 @@ emit_multi_dqt (j_compress_ptr cinfo)
  int ci;
  int size = 0;
  if (cinfo->master->compress_profile == JCP_FASTEST)
    return -1;
  for (ci = 0; ci < cinfo->num_components; ci++) {
    int tbl_num = cinfo->comp_info[ci].quant_tbl_no;
    int i;
@@ -296,6 +299,9 @@ emit_multi_dht (j_compress_ptr cinfo)
  JHUFF_TBL *dcseen[NUM_HUFF_TBLS] = { NULL };
  JHUFF_TBL *acseen[NUM_HUFF_TBLS] = { NULL };
  if (cinfo->master->compress_profile == JCP_FASTEST)
    return 0;
  /* Calclate the total length. */
  for (i = 0; i < cinfo->comps_in_scan; i++) {
    jpeg_component_info *compptr = cinfo->cur_comp_info[i];
--- a/jcmaster.c
+++ b/jcmaster.c
@@ -21,42 +21,7 @@
 #include "jpeglib.h"
 #include "jpegcomp.h"
 #include "jmemsys.h"
-
+#include "jcmaster.h"
 /* Private state */
 typedef enum {
 	main_pass,		/* input data, also do first output step */
 	huff_opt_pass,		/* Huffman code optimization pass */
 	output_pass,		/* data output pass */
        trellis_pass            /* trellis quantization pass */
 } c_pass_type;
 typedef struct {
  struct jpeg_comp_master pub;	/* public fields */
  c_pass_type pass_type;	/* the type of the current pass */
  int pass_number;		/* # of passes completed */
  int total_passes;		/* total # of passes needed */
  int scan_number;		/* current index in scan_info[] */
  /* fields for scan optimisation */
  int pass_number_scan_opt_base; /* pass number where scan optimization begins */
  unsigned char * scan_buffer[64]; /* buffer for a given scan */
  unsigned long scan_size[64]; /* size for a given scan */
  int actual_Al[64]; /* actual value of Al used for a scan */
  unsigned long best_cost; /* bit count for best frequency split */
  int best_freq_split_idx_luma; /* index for best frequency split (luma) */
  int best_freq_split_idx_chroma; /* index for best frequency split (chroma) */
  int best_Al_luma; /* best value for Al found in scan search (luma) */
  int best_Al_chroma; /* best value for Al found in scan search (luma) */
  boolean interleave_chroma_dc; /* indicate whether to interleave chroma DC scans */
  struct jpeg_destination_mgr * saved_dest; /* saved value of cinfo->dest */
 } my_comp_master;
 typedef my_comp_master * my_master_ptr;
 /*
@@ -510,12 +475,12 @@ prepare_for_pass (j_compress_ptr cinfo)
      (*cinfo->prep->start_pass) (cinfo, JBUF_PASS_THRU);
    }
    (*cinfo->fdct->start_pass) (cinfo);
-    (*cinfo->entropy->start_pass) (cinfo, cinfo->optimize_coding);
+    (*cinfo->entropy->start_pass) (cinfo, (cinfo->optimize_coding || cinfo->master->trellis_quant) && !cinfo->arith_code);
    (*cinfo->coef->start_pass) (cinfo,
                                (master->total_passes > 1 ?
                                 JBUF_SAVE_AND_PASS : JBUF_PASS_THRU));
    (*cinfo->main->start_pass) (cinfo, JBUF_PASS_THRU);
-    if (cinfo->optimize_coding) {
+    if (cinfo->optimize_coding || cinfo->master->trellis_quant) {
      /* No immediate data output; postpone writing frame/scan headers */
      master->pub.call_pass_startup = FALSE;
    } else {
@@ -576,7 +541,7 @@ prepare_for_pass (j_compress_ptr cinfo)
        }
      }
    }
-    (*cinfo->entropy->start_pass) (cinfo, TRUE);
+    (*cinfo->entropy->start_pass) (cinfo, !cinfo->arith_code);
    (*cinfo->coef->start_pass) (cinfo, JBUF_REQUANT);
    master->pub.call_pass_startup = FALSE;
    break;
@@ -640,7 +605,9 @@ copy_buffer (j_compress_ptr cinfo, int scan_idx)
    size -= cinfo->dest->free_in_buffer;
    cinfo->dest->next_output_byte += cinfo->dest->free_in_buffer;
    cinfo->dest->free_in_buffer = 0;
-    (*cinfo->dest->empty_output_buffer)(cinfo);
+    
    if (!(*cinfo->dest->empty_output_buffer)(cinfo))
      ERREXIT(cinfo, JERR_UNSUPPORTED_SUSPEND);
  }
  MEMCOPY(cinfo->dest->next_output_byte, src, size);
@@ -659,6 +626,7 @@ select_scans (j_compress_ptr cinfo, int next_scan_number)
  int chroma_freq_split_scan_start = cinfo->master->num_scans_luma +
                                     cinfo->master->num_scans_chroma_dc +
                                     (6 * cinfo->master->Al_max_chroma + 4);
  int passes_per_scan = cinfo->optimize_coding ? 2 : 1;
  if (next_scan_number > 1 && next_scan_number <= luma_freq_split_scan_start) {
    if ((next_scan_number - 1) % 3 == 2) {
@@ -675,7 +643,7 @@ select_scans (j_compress_ptr cinfo, int next_scan_number)
        master->best_Al_luma = Al;
      } else {
        master->scan_number = luma_freq_split_scan_start - 1;
-        master->pass_number = 2 * master->scan_number + 1 + master->pass_number_scan_opt_base;
+        master->pass_number = passes_per_scan * (master->scan_number + 1) - 1 + master->pass_number_scan_opt_base;
      }
    }
@@ -701,7 +669,7 @@ select_scans (j_compress_ptr cinfo, int next_scan_number)
          (idx == 3 && master->best_freq_split_idx_luma != 2) ||
          (idx == 4 && master->best_freq_split_idx_luma != 4)) {
        master->scan_number = cinfo->master->num_scans_luma - 1;
-        master->pass_number = 2 * master->scan_number + 1 + master->pass_number_scan_opt_base;
+        master->pass_number = passes_per_scan * (master->scan_number + 1) - 1 + master->pass_number_scan_opt_base;
        master->pub.is_last_pass = (master->pass_number == master->total_passes - 1);
      }
    }
@@ -737,7 +705,7 @@ select_scans (j_compress_ptr cinfo, int next_scan_number)
          master->best_Al_chroma = Al;
        } else {
          master->scan_number = chroma_freq_split_scan_start - 1;
-          master->pass_number = 2 * master->scan_number + 1 + master->pass_number_scan_opt_base;
+          master->pass_number = passes_per_scan * (master->scan_number + 1) - 1 + master->pass_number_scan_opt_base;
        }
      }
@@ -765,7 +733,7 @@ select_scans (j_compress_ptr cinfo, int next_scan_number)
            (idx == 3 && master->best_freq_split_idx_chroma != 2) ||
            (idx == 4 && master->best_freq_split_idx_chroma != 4)) {
          master->scan_number = cinfo->num_scans - 1;
-          master->pass_number = 2 * master->scan_number + 1 + master->pass_number_scan_opt_base;
+          master->pass_number = passes_per_scan * (master->scan_number + 1) - 1 + master->pass_number_scan_opt_base;
          master->pub.is_last_pass = (master->pass_number == master->total_passes - 1);
        }
      }
@@ -779,10 +747,10 @@ select_scans (j_compress_ptr cinfo, int next_scan_number)
    copy_buffer(cinfo, 0);
    if (cinfo->num_scans > cinfo->master->num_scans_luma &&
-        !cinfo->master->one_dc_scan) {
+        cinfo->master->dc_scan_opt_mode != 0) {
      base_scan_idx = cinfo->master->num_scans_luma;
-      if (master->interleave_chroma_dc && !cinfo->master->sep_dc_scan)
+      if (master->interleave_chroma_dc && cinfo->master->dc_scan_opt_mode != 1)
        copy_buffer(cinfo, base_scan_idx);
      else {
        copy_buffer(cinfo, base_scan_idx+1);
@@ -883,7 +851,10 @@ finish_pass_master (j_compress_ptr cinfo)
    master->scan_number++;
    break;
  case trellis_pass:
-    master->pass_type = (cinfo->optimize_coding || master->pass_number < master->pass_number_scan_opt_base-1) ? huff_opt_pass : output_pass;
+    if (cinfo->optimize_coding)
      master->pass_type = huff_opt_pass;
    else
      master->pass_type = (master->pass_number < master->pass_number_scan_opt_base-1) ? trellis_pass : output_pass;
    if ((master->pass_number + 1) %
        (cinfo->num_components * (cinfo->master->use_scans_in_trellis ? 4 : 2)) == 0 &&
@@ -916,21 +887,13 @@ finish_pass_master (j_compress_ptr cinfo)
 GLOBAL(void)
 jinit_c_master_control (j_compress_ptr cinfo, boolean transcode_only)
 {
-  my_master_ptr master;
+  my_master_ptr master = (my_master_ptr) cinfo->master;
  master = (my_master_ptr)
      (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
                                  sizeof(my_comp_master));
  if (cinfo->master) {
    MEMCOPY(&master->pub, cinfo->master, sizeof(struct jpeg_comp_master));
    jpeg_free_small((j_common_ptr) cinfo, cinfo->master,
                    sizeof(struct jpeg_comp_master));
  }
  cinfo->master = (struct jpeg_comp_master *) master;
  master->pub.prepare_for_pass = prepare_for_pass;
  master->pub.pass_startup = pass_startup;
  master->pub.finish_pass = finish_pass_master;
  master->pub.is_last_pass = FALSE;
  master->pub.call_pass_startup = FALSE;
  /* Validate parameters, determine derived values */
  initial_setup(cinfo, transcode_only);
@@ -969,9 +932,14 @@ jinit_c_master_control (j_compress_ptr cinfo, boolean transcode_only)
  master->pass_number_scan_opt_base = 0;
  if (cinfo->master->trellis_quant) {
    if (cinfo->optimize_coding)
      master->pass_number_scan_opt_base =
        ((cinfo->master->use_scans_in_trellis) ? 4 : 2) * cinfo->num_components *
        cinfo->master->trellis_num_loops;
    else
      master->pass_number_scan_opt_base =
        ((cinfo->master->use_scans_in_trellis) ? 2 : 1) * cinfo->num_components *
        cinfo->master->trellis_num_loops + 1;
    master->total_passes += master->pass_number_scan_opt_base;
  }
--- a/jcmaster.h
+++ b/jcmaster.h
@@ -0,0 +1,47 @@
 /*
 * jcmaster.h
 *
 * This file was part of the Independent JPEG Group's software:
 * Copyright (C) 1991-1997, Thomas G. Lane.
 * mozjpeg Modifications:
 * Copyright (C) 2014, Mozilla Corporation.
 * For conditions of distribution and use, see the accompanying README file.
 *
 * This file contains the master control structures for the JPEG compressor.
 */
 /* Private state */
 typedef enum {
        main_pass,              /* input data, also do first output step */
        huff_opt_pass,          /* Huffman code optimization pass */
        output_pass,            /* data output pass */
        trellis_pass            /* trellis quantization pass */
 } c_pass_type;
 typedef struct {
  struct jpeg_comp_master pub;  /* public fields */
  c_pass_type pass_type;        /* the type of the current pass */
  int pass_number;              /* # of passes completed */
  int total_passes;             /* total # of passes needed */
  int scan_number;              /* current index in scan_info[] */
  /* fields for scan optimisation */
  int pass_number_scan_opt_base; /* pass number where scan optimization begins */
  unsigned char * scan_buffer[64]; /* buffer for a given scan */
  unsigned long scan_size[64]; /* size for a given scan */
  int actual_Al[64]; /* actual value of Al used for a scan */
  unsigned long best_cost; /* bit count for best frequency split */
  int best_freq_split_idx_luma; /* index for best frequency split (luma) */
  int best_freq_split_idx_chroma; /* index for best frequency split (chroma) */
  int best_Al_luma; /* best value for Al found in scan search (luma) */
  int best_Al_chroma; /* best value for Al found in scan search (luma) */
  boolean interleave_chroma_dc; /* indicate whether to interleave chroma DC scans */
  struct jpeg_destination_mgr * saved_dest; /* saved value of cinfo->dest */
 } my_comp_master;
 typedef my_comp_master * my_master_ptr;
--- a/jcparam.c
+++ b/jcparam.c
@@ -70,7 +70,10 @@ jpeg_add_quant_table (j_compress_ptr cinfo, int which_tbl,
 * The spec says that the values given produce "good" quality, and
 * when divided by 2, "very good" quality.
 */
-static const unsigned int std_luminance_quant_tbl[DCTSIZE2] = {
+static const unsigned int std_luminance_quant_tbl[9][DCTSIZE2] = {
  {
    /* JPEG Annex K
     */
    16,  11,  10,  16,  24,  40,  51,  61,
    12,  12,  14,  19,  26,  58,  60,  55,
    14,  13,  16,  24,  40,  57,  69,  56,
@@ -79,19 +82,10 @@ static const unsigned int std_luminance_quant_tbl[DCTSIZE2] = {
    24,  35,  55,  64,  81, 104, 113,  92,
    49,  64,  78,  87, 103, 121, 120, 101,
    72,  92,  95,  98, 112, 100, 103,  99
-};
+  },
-static const unsigned int std_chrominance_quant_tbl[DCTSIZE2] = {
+  {
-  17,  18,  24,  47,  99,  99,  99,  99,
+    /* flat
-  18,  21,  26,  66,  99,  99,  99,  99,
+     */
  24,  26,  56,  99,  99,  99,  99,  99,
  47,  66,  99,  99,  99,  99,  99,  99,
  99,  99,  99,  99,  99,  99,  99,  99,
  99,  99,  99,  99,  99,  99,  99,  99,
  99,  99,  99,  99,  99,  99,  99,  99,
  99,  99,  99,  99,  99,  99,  99,  99
 };
 static const unsigned int flat_quant_tbl[DCTSIZE2] = {
    16,  16,  16,  16,  16,  16,  16,  16,
    16,  16,  16,  16,  16,  16,  16,  16,
    16,  16,  16,  16,  16,  16,  16,  16,
@@ -100,6 +94,198 @@ static const unsigned int flat_quant_tbl[DCTSIZE2] = {
    16,  16,  16,  16,  16,  16,  16,  16,
    16,  16,  16,  16,  16,  16,  16,  16,
    16,  16,  16,  16,  16,  16,  16,  16
  },
  {
    12, 17, 20, 21, 30, 34, 56, 63,
    18, 20, 20, 26, 28, 51, 61, 55,
    19, 20, 21, 26, 33, 58, 69, 55,
    26, 26, 26, 30, 46, 87, 86, 66,
    31, 33, 36, 40, 46, 96, 100, 73,
    40, 35, 46, 62, 81, 100, 111, 91,
    46, 66, 76, 86, 102, 121, 120, 101,
    68, 90, 90, 96, 113, 102, 105, 103
  },
  {
    /* From http://www.imagemagick.org/discourse-server/viewtopic.php?f=22&t=20333&p=98008#p98008
     */
    16,  16,  16,  18,  25,  37,  56,  85,
    16,  17,  20,  27,  34,  40,  53,  75,
    16,  20,  24,  31,  43,  62,  91,  135,
    18,  27,  31,  40,  53,  74,  106, 156,
    25,  34,  43,  53,  69,  94,  131, 189,
    37,  40,  62,  74,  94,  124, 169, 238,
    56,  53,  91,  106, 131, 169, 226, 311,
    85,  75,  135, 156, 189, 238, 311, 418
  },
  {
    9, 10, 12, 14, 27, 32, 51, 62,
    11, 12, 14, 19, 27, 44, 59, 73,
    12, 14, 18, 25, 42, 59, 79, 78,
    17, 18, 25, 42, 61, 92, 87, 92,
    23, 28, 42, 75, 79, 112, 112, 99,
    40, 42, 59, 84, 88, 124, 132, 111,
    42, 64, 78, 95, 105, 126, 125, 99,
    70, 75, 100, 102, 116, 100, 107, 98
  },
  {
    /* Relevance of human vision to JPEG-DCT compression (1992) Klein, Silverstein and Carney.
     */
    10, 12, 14, 19, 26, 38, 57, 86,
    12, 18, 21, 28, 35, 41, 54, 76,
    14, 21, 25, 32, 44, 63, 92, 136,
    19, 28, 32, 41, 54, 75, 107, 157,
    26, 35, 44, 54, 70, 95, 132, 190,
    38, 41, 63, 75, 95, 125, 170, 239,
    57, 54, 92, 107, 132, 170, 227, 312,
    86, 76, 136, 157, 190, 239, 312, 419
  },
  {
    /* DCTune perceptual optimization of compressed dental X-Rays (1997) Watson, Taylor, Borthwick
     */
    7, 8, 10, 14, 23, 44, 95, 241,
    8, 8, 11, 15, 25, 47, 102, 255,
    10, 11, 13, 19, 31, 58, 127, 255,
    14, 15, 19, 27, 44, 83, 181, 255,
    23, 25, 31, 44, 72, 136, 255, 255,
    44, 47, 58, 83, 136, 255, 255, 255,
    95, 102, 127, 181, 255, 255, 255, 255,
    241, 255, 255, 255, 255, 255, 255, 255
  },
  {
    /* A visual detection model for DCT coefficient quantization (12/9/93) Ahumada, Watson, Peterson
     */
    15, 11, 11, 12, 15, 19, 25, 32,
    11, 13, 10, 10, 12, 15, 19, 24,
    11, 10, 14, 14, 16, 18, 22, 27,
    12, 10, 14, 18, 21, 24, 28, 33,
    15, 12, 16, 21, 26, 31, 36, 42,
    19, 15, 18, 24, 31, 38, 45, 53,
    25, 19, 22, 28, 36, 45, 55, 65,
    32, 24, 27, 33, 42, 53, 65, 77
  },
  {
    /* An improved detection model for DCT coefficient quantization (1993) Peterson, Ahumada and Watson
     */
    14, 10, 11, 14, 19, 25, 34, 45,
    10, 11, 11, 12, 15, 20, 26, 33,
    11, 11, 15, 18, 21, 25, 31, 38,
    14, 12, 18, 24, 28, 33, 39, 47,
    19, 15, 21, 28, 36, 43, 51, 59,
    25, 20, 25, 33, 43, 54, 64, 74,
    34, 26, 31, 39, 51, 64, 77, 91,
    45, 33, 38, 47, 59, 74, 91, 108
  }
 };
 static const unsigned int std_chrominance_quant_tbl[9][DCTSIZE2] = {
  {
    /* JPEG Annex K
     */
    17,  18,  24,  47,  99,  99,  99,  99,
    18,  21,  26,  66,  99,  99,  99,  99,
    24,  26,  56,  99,  99,  99,  99,  99,
    47,  66,  99,  99,  99,  99,  99,  99,
    99,  99,  99,  99,  99,  99,  99,  99,
    99,  99,  99,  99,  99,  99,  99,  99,
    99,  99,  99,  99,  99,  99,  99,  99,
    99,  99,  99,  99,  99,  99,  99,  99
  },
  {
    /* flat
     */
    16,  16,  16,  16,  16,  16,  16,  16,
    16,  16,  16,  16,  16,  16,  16,  16,
    16,  16,  16,  16,  16,  16,  16,  16,
    16,  16,  16,  16,  16,  16,  16,  16,
    16,  16,  16,  16,  16,  16,  16,  16,
    16,  16,  16,  16,  16,  16,  16,  16,
    16,  16,  16,  16,  16,  16,  16,  16,
    16,  16,  16,  16,  16,  16,  16,  16
  },
  {
    8, 12, 15, 15, 86, 96, 96, 98,
    13, 13, 15, 26, 90, 96, 99, 98,
    12, 15, 18, 96, 99, 99, 99, 99,
    17, 16, 90, 96, 99, 99, 99, 99,
    96, 96, 99, 99, 99, 99, 99, 99,
    99, 99, 99, 99, 99, 99, 99, 99,
    99, 99, 99, 99, 99, 99, 99, 99,
    99, 99, 99, 99, 99, 99, 99, 99
  },
  {
    /* From http://www.imagemagick.org/discourse-server/viewtopic.php?f=22&t=20333&p=98008#p98008
     */
    16,  16,  16,  18,  25,  37,  56,  85,
    16,  17,  20,  27,  34,  40,  53,  75,
    16,  20,  24,  31,  43,  62,  91,  135,
    18,  27,  31,  40,  53,  74,  106, 156,
    25,  34,  43,  53,  69,  94,  131, 189,
    37,  40,  62,  74,  94,  124, 169, 238,
    56,  53,  91,  106, 131, 169, 226, 311,
    85,  75,  135, 156, 189, 238, 311, 418
  },
  {
    9, 10, 17, 19, 62, 89, 91, 97,
    12, 13, 18, 29, 84, 91, 88, 98,
    14, 19, 29, 93, 95, 95, 98, 97,
    20, 26, 84, 88, 95, 95, 98, 94,
    26, 86, 91, 93, 97, 99, 98, 99,
    99, 100, 98, 99, 99, 99, 99, 99,
    99, 99, 99, 99, 99, 99, 99, 99,
    97, 97, 99, 99, 99, 99, 97, 99
  },
  {
    /* Relevance of human vision to JPEG-DCT compression (1992) Klein, Silverstein and Carney.
     * Copied from luma
     */
    10, 12, 14, 19, 26, 38, 57, 86,
    12, 18, 21, 28, 35, 41, 54, 76,
    14, 21, 25, 32, 44, 63, 92, 136,
    19, 28, 32, 41, 54, 75, 107, 157,
    26, 35, 44, 54, 70, 95, 132, 190,
    38, 41, 63, 75, 95, 125, 170, 239,
    57, 54, 92, 107, 132, 170, 227, 312,
    86, 76, 136, 157, 190, 239, 312, 419
  },
  {
    /* DCTune perceptual optimization of compressed dental X-Rays (1997) Watson, Taylor, Borthwick
     * Copied from luma
     */
    7, 8, 10, 14, 23, 44, 95, 241,
    8, 8, 11, 15, 25, 47, 102, 255,
    10, 11, 13, 19, 31, 58, 127, 255,
    14, 15, 19, 27, 44, 83, 181, 255,
    23, 25, 31, 44, 72, 136, 255, 255,
    44, 47, 58, 83, 136, 255, 255, 255,
    95, 102, 127, 181, 255, 255, 255, 255,
    241, 255, 255, 255, 255, 255, 255, 255
  },
  {
    /* A visual detection model for DCT coefficient quantization (12/9/93) Ahumada, Watson, Peterson
     * Copied from luma
     */
    15, 11, 11, 12, 15, 19, 25, 32,
    11, 13, 10, 10, 12, 15, 19, 24,
    11, 10, 14, 14, 16, 18, 22, 27,
    12, 10, 14, 18, 21, 24, 28, 33,
    15, 12, 16, 21, 26, 31, 36, 42,
    19, 15, 18, 24, 31, 38, 45, 53,
    25, 19, 22, 28, 36, 45, 55, 65,
    32, 24, 27, 33, 42, 53, 65, 77
  },
  {
    /* An improved detection model for DCT coefficient quantization (1993) Peterson, Ahumada and Watson
     * Copied from luma
     */
    14, 10, 11, 14, 19, 25, 34, 45,
    10, 11, 11, 12, 15, 20, 26, 33,
    11, 11, 15, 18, 21, 25, 31, 38,
    14, 12, 18, 24, 28, 33, 39, 47,
    19, 15, 21, 28, 36, 43, 51, 59,
    25, 20, 25, 33, 43, 54, 64, 74,
    34, 26, 31, 39, 51, 64, 77, 91,
    45, 33, 38, 47, 59, 74, 91, 108
  }
 };
 #if JPEG_LIB_VERSION >= 70
@@ -111,17 +297,10 @@ jpeg_default_qtables (j_compress_ptr cinfo, boolean force_baseline)
 */
 {
  /* Set up two quantization tables using the specified scaling */
-  if (cinfo->use_flat_quant_tbl) {
+  jpeg_add_quant_table(cinfo, 0, std_luminance_quant_tbl[cinfo->master->quant_tbl_master_idx],
    jpeg_add_quant_table(cinfo, 0, flat_quant_tbl,
                       cinfo->q_scale_factor[0], force_baseline);
-    jpeg_add_quant_table(cinfo, 1, flat_quant_tbl,
+  jpeg_add_quant_table(cinfo, 1, std_chrominance_quant_tbl[cinfo->master->quant_tbl_master_idx],
                       cinfo->q_scale_factor[1], force_baseline);
  } else {
  jpeg_add_quant_table(cinfo, 0, std_luminance_quant_tbl,
                       cinfo->q_scale_factor[0], force_baseline);
  jpeg_add_quant_table(cinfo, 1, std_chrominance_quant_tbl,
                       cinfo->q_scale_factor[1], force_baseline);
 }
 }
 #endif
@@ -136,17 +315,10 @@ jpeg_set_linear_quality (j_compress_ptr cinfo, int scale_factor,
 */
 {
  /* Set up two quantization tables using the specified scaling */
-  if (cinfo->master->use_flat_quant_tbl) {
+  jpeg_add_quant_table(cinfo, 0, std_luminance_quant_tbl[cinfo->master->quant_tbl_master_idx],
    jpeg_add_quant_table(cinfo, 0, flat_quant_tbl,
                       scale_factor, force_baseline);
-    jpeg_add_quant_table(cinfo, 1, flat_quant_tbl,
+  jpeg_add_quant_table(cinfo, 1, std_chrominance_quant_tbl[cinfo->master->quant_tbl_master_idx],
                       scale_factor, force_baseline);
  } else {
  jpeg_add_quant_table(cinfo, 0, std_luminance_quant_tbl,
                       scale_factor, force_baseline);
  jpeg_add_quant_table(cinfo, 1, std_chrominance_quant_tbl,
                       scale_factor, force_baseline);
 }
 }
@@ -245,19 +417,9 @@ jpeg_set_defaults (j_compress_ptr cinfo)
    cinfo->arith_ac_K[i] = 5;
  }
 #ifdef C_PROGRESSIVE_SUPPORTED
  cinfo->scan_info = NULL;
  cinfo->num_scans = 0;
  if (!cinfo->master->use_moz_defaults) {
  /* Default is no multiple-scan output */
  cinfo->scan_info = NULL;
  cinfo->num_scans = 0;
  }
 #else
  /* Default is no multiple-scan output */
  cinfo->scan_info = NULL;
  cinfo->num_scans = 0;
 #endif
  /* Expect normal source image, not raw downsampled data */
  cinfo->raw_data_in = FALSE;
@@ -266,7 +428,7 @@ jpeg_set_defaults (j_compress_ptr cinfo)
  cinfo->arith_code = FALSE;
 #ifdef ENTROPY_OPT_SUPPORTED
-  if (cinfo->master->use_moz_defaults)
+  if (cinfo->master->compress_profile == JCP_MAX_COMPRESSION)
    /* By default, do extra passes to optimize entropy coding */
    cinfo->optimize_coding = TRUE;
  else
@@ -293,7 +455,8 @@ jpeg_set_defaults (j_compress_ptr cinfo)
  cinfo->do_fancy_downsampling = TRUE;
 #endif
-  cinfo->master->overshoot_deringing = cinfo->master->use_moz_defaults;
+  cinfo->master->overshoot_deringing =
    cinfo->master->compress_profile == JCP_MAX_COMPRESSION;
  /* No input smoothing */
  cinfo->smoothing_factor = 0;
@@ -324,19 +487,22 @@ jpeg_set_defaults (j_compress_ptr cinfo)
  jpeg_default_colorspace(cinfo);
-  cinfo->master->one_dc_scan = TRUE;
+  cinfo->master->dc_scan_opt_mode = 1;
 #ifdef C_PROGRESSIVE_SUPPORTED
-  if (cinfo->master->use_moz_defaults) {
+  if (cinfo->master->compress_profile == JCP_MAX_COMPRESSION) {
    cinfo->master->optimize_scans = TRUE;
    jpeg_simple_progression(cinfo);
  } else
    cinfo->master->optimize_scans = FALSE;
 #endif
-  cinfo->master->trellis_quant = cinfo->master->use_moz_defaults;
+  cinfo->master->trellis_quant =
-  cinfo->master->lambda_log_scale1 = 16.0;
+    cinfo->master->compress_profile == JCP_MAX_COMPRESSION;
-  cinfo->master->lambda_log_scale2 = 15.5;
+  cinfo->master->lambda_log_scale1 = 14.75;
  cinfo->master->lambda_log_scale2 = 16.5;
  cinfo->master->quant_tbl_master_idx =
    cinfo->master->compress_profile == JCP_MAX_COMPRESSION ? 3 : 0;
  cinfo->master->use_lambda_weight_tbl = TRUE;
  cinfo->master->use_scans_in_trellis = FALSE;
@@ -344,6 +510,7 @@ jpeg_set_defaults (j_compress_ptr cinfo)
  cinfo->master->trellis_num_loops = 1;
  cinfo->master->trellis_q_opt = FALSE;
  cinfo->master->trellis_quant_dc = TRUE;
  cinfo->master->trellis_delta_dc_weight = 0.0;
 }
@@ -613,7 +780,7 @@ jpeg_search_progression (j_compress_ptr cinfo)
  /* last 4 done conditionally */
  /* luma DC by itself */
-  if (cinfo->master->one_dc_scan)
+  if (cinfo->master->dc_scan_opt_mode == 0)
    scanptr = fill_dc_scans(scanptr, ncomps, 0, 0);
  else
    scanptr = fill_dc_scans(scanptr, 1, 0, 0);
@@ -704,7 +871,7 @@ jpeg_simple_progression (j_compress_ptr cinfo)
    nscans = 10;
  } else {
    /* All-purpose script for other color spaces. */
-    if (cinfo->master->use_moz_defaults == TRUE) {
+    if (cinfo->master->compress_profile == JCP_MAX_COMPRESSION) {
      if (ncomps > MAX_COMPS_IN_SCAN)
        nscans = 5 * ncomps;      /* 2 DC + 4 AC scans per component */
      else
@@ -736,12 +903,12 @@ jpeg_simple_progression (j_compress_ptr cinfo)
  if (ncomps == 3 && cinfo->jpeg_color_space == JCS_YCbCr) {
    /* Custom script for YCbCr color images. */
-    if (cinfo->master->use_moz_defaults == TRUE) {
+    if (cinfo->master->compress_profile == JCP_MAX_COMPRESSION) {
      /* scan defined in jpeg_scan_rgb.txt in jpgcrush */
    /* Initial DC scan */
-      if (cinfo->master->one_dc_scan)
+      if (cinfo->master->dc_scan_opt_mode == 0)
        scanptr = fill_dc_scans(scanptr, ncomps, 0, 0);
-      else if (cinfo->master->sep_dc_scan) {
+      else if (cinfo->master->dc_scan_opt_mode == 1) {
        scanptr = fill_a_scan(scanptr, 0, 0, 0, 0, 0);
        scanptr = fill_a_scan(scanptr, 1, 0, 0, 0, 0);
        scanptr = fill_a_scan(scanptr, 2, 0, 0, 0, 0);
@@ -784,7 +951,7 @@ jpeg_simple_progression (j_compress_ptr cinfo)
    }
  } else {
    /* All-purpose script for other color spaces. */
-    if (cinfo->master->use_moz_defaults == TRUE) {
+    if (cinfo->master->compress_profile == JCP_MAX_COMPRESSION) {
      /* scan defined in jpeg_scan_bw.txt in jpgcrush */
      /* DC component, no successive approximation */
      scanptr = fill_dc_scans(scanptr, ncomps, 0, 0);
@@ -806,8 +973,7 @@ jpeg_simple_progression (j_compress_ptr cinfo)
      scanptr = fill_dc_scans(scanptr, ncomps, 1, 0);
      scanptr = fill_scans(scanptr, ncomps, 1, 63, 1, 0);
    }
  }
 }
 }
 #endif /* C_PROGRESSIVE_SUPPORTED */
--- a/jcphuff.c
+++ b/jcphuff.c
@@ -177,7 +177,7 @@ start_pass_phuff (j_compress_ptr cinfo, boolean gather_statistics)
        int i, j;
        for (i = 0; i < 16; i++)
          for (j = 0; j < 12; j++)
-          entropy->count_ptrs[tbl][16*i+j] = 1;
+            entropy->count_ptrs[tbl][16 * i + j] = 1;
      }
    } else {
      /* Compute derived values for Huffman table */
--- a/jcsample.c
+++ b/jcsample.c
@@ -514,8 +514,9 @@ jinit_downsampler (j_compress_ptr cinfo)
 #endif
          downsample->methods[ci] = h2v2_smooth_downsample;
        downsample->pub.need_context_rows = TRUE;
-      } else {
+      } else
 #endif
      {
        if (jsimd_can_h2v2_downsample())
          downsample->methods[ci] = jsimd_h2v2_downsample;
        else
--- a/jcstest.c
+++ b/jcstest.c
@@ -1,7 +1,5 @@
 /*
 * Copyright (C)2011 D. R. Commander.  All Rights Reserved.
 * mozjpeg Modifications:
 * Copyright (C) 2014, Mozilla Corporation.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions are met:
@@ -87,8 +85,6 @@ int main(void)
  jpeg_create_compress(&cinfo);
  cinfo.input_components = 3;
  if (jpeg_c_bool_param_supported(&cinfo, JBOOLEAN_USE_MOZ_DEFAULTS))
    jpeg_c_set_bool_param(&cinfo, JBOOLEAN_USE_MOZ_DEFAULTS, TRUE);
  jpeg_set_defaults(&cinfo);
  cinfo.in_color_space = JCS_EXT_RGB;
  jpeg_default_colorspace(&cinfo);
--- a/jctrans.c
+++ b/jctrans.c
@@ -69,7 +69,7 @@ jpeg_write_coefficients (j_compress_ptr cinfo, jvirt_barray_ptr * coef_arrays)
 */
 GLOBAL(void)
-jpeg_copy_critical_parameters (j_decompress_ptr srcinfo,
+jpeg_copy_critical_parameters (const j_decompress_ptr srcinfo,
                               j_compress_ptr dstinfo)
 {
  JQUANT_TBL ** qtblptr;
--- a/jdapimin.c
+++ b/jdapimin.c
@@ -333,7 +333,7 @@ jpeg_consume_input (j_decompress_ptr cinfo)
 */
 GLOBAL(boolean)
-jpeg_input_complete (j_decompress_ptr cinfo)
+jpeg_input_complete (const j_decompress_ptr cinfo)
 {
  /* Check for valid jpeg object */
  if (cinfo->global_state < DSTATE_START ||
@@ -348,7 +348,7 @@ jpeg_input_complete (j_decompress_ptr cinfo)
 */
 GLOBAL(boolean)
-jpeg_has_multiple_scans (j_decompress_ptr cinfo)
+jpeg_has_multiple_scans (const j_decompress_ptr cinfo)
 {
  /* Only valid after jpeg_read_header completes */
  if (cinfo->global_state < DSTATE_READY ||
--- a/jdatasrc.c
+++ b/jdatasrc.c
@@ -254,7 +254,7 @@ jpeg_stdio_src (j_decompress_ptr cinfo, FILE * infile)
 GLOBAL(void)
 jpeg_mem_src (j_decompress_ptr cinfo,
-              unsigned char * inbuffer, unsigned long insize)
+              const unsigned char * inbuffer, unsigned long insize)
 {
  struct jpeg_source_mgr * src;
@@ -278,6 +278,6 @@ jpeg_mem_src (j_decompress_ptr cinfo,
  src->resync_to_restart = jpeg_resync_to_restart; /* use default method */
  src->term_source = term_source;
  src->bytes_in_buffer = (size_t) insize;
-  src->next_input_byte = (JOCTET *) inbuffer;
+  src->next_input_byte = (const JOCTET *) inbuffer;
 }
 #endif
--- a/jerror.h
+++ b/jerror.h
@@ -208,6 +208,9 @@ JMESSAGE(JWRN_ARITH_BAD_CODE, "Corrupt JPEG data: bad arithmetic code")
 #endif
 #endif
 JMESSAGE(JERR_BAD_PARAM, "Bogus parameter")
 JMESSAGE(JERR_BAD_PARAM_VALUE, "Bogus parameter value")
 JMESSAGE(JERR_UNSUPPORTED_SUSPEND, "I/O suspension not supported in scan optimization")  
 #ifdef JMAKE_ENUM_LIST
--- a/jmorecfg.h
+++ b/jmorecfg.h
@@ -180,6 +180,24 @@ typedef unsigned int JDIMENSION;
 #define EXTERN(type)            extern type
 /* Originally, this macro was used as a way of defining function prototypes
 * for both modern compilers as well as older compilers that did not support
 * prototype parameters.  libjpeg-turbo has never supported these older,
 * non-ANSI compilers, but the macro is still included because there is some
 * software out there that uses it.
 */
 #define JMETHOD(type,methodname,arglist)  type (*methodname) arglist
 /* libjpeg-turbo no longer supports platforms that have far symbols (MS-DOS),
 * but again, some software relies on this macro.
 */
 #undef FAR
 #define FAR
 /*
 * On a few systems, type boolean and/or its values FALSE, TRUE may appear
 * in standard header files.  Or you may have conflicts with application-
--- a/jpegint.h
+++ b/jpegint.h
@@ -25,7 +25,6 @@ typedef enum {            /* Operating modes for buffer controllers */
  JBUF_CRANK_DEST,        /* Run dest subobject only, using saved data */
  JBUF_SAVE_AND_PASS,     /* Run both subobjects, save output */
  JBUF_REQUANT            /* Requantize */
 } J_BUF_MODE;
 /* Values of global_state field (jdapi.c has some dependencies on ordering!) */
@@ -59,23 +58,22 @@ struct jpeg_comp_master {
  boolean is_last_pass;         /* True during last pass */
  /* Extension parameters */
  boolean use_moz_defaults; /* TRUE=use Mozilla defaults */
  boolean optimize_scans; /* TRUE=optimize progressive coding scans */
  boolean one_dc_scan; /* TRUE=use a single DC scan interleaving all components */
  boolean sep_dc_scan; /* TRUE=each DC scan is separate */
  boolean trellis_quant; /* TRUE=use trellis quantization */
  boolean trellis_quant_dc; /* TRUE=use trellis quant for DC coefficient */
  boolean trellis_eob_opt; /* TRUE=optimize for sequences of EOB */
  boolean use_flat_quant_tbl; /* TRUE=use flat quantization table */
  boolean use_lambda_weight_tbl; /* TRUE=use lambda weighting table */
  boolean use_scans_in_trellis; /* TRUE=use scans in trellis optimization */
-  boolean trellis_passes; /* TRUE=currently doing trellis-related passes */
+  boolean trellis_passes; /* TRUE=currently doing trellis-related passes [not exposed] */
  boolean trellis_q_opt; /* TRUE=optimize quant table in trellis loop */
  boolean overshoot_deringing; /* TRUE=preprocess input to reduce ringing of edges on white background */
  double norm_src[NUM_QUANT_TBLS][DCTSIZE2];
  double norm_coef[NUM_QUANT_TBLS][DCTSIZE2];
  int compress_profile; /* compression profile */
  int dc_scan_opt_mode; /* DC scan optimization mode */
  int quant_tbl_master_idx; /* Quantization table master index */
  int trellis_freq_split; /* splitting point for frequency in trellis quantization */
  int trellis_num_loops; /* number of trellis loops */
@@ -89,8 +87,36 @@ struct jpeg_comp_master {
  float lambda_log_scale1;
  float lambda_log_scale2;
  float trellis_delta_dc_weight;
 };
 #ifdef C_ARITH_CODING_SUPPORTED
 /* The following two definitions specify the allocation chunk size
 * for the statistics area.
 * According to sections F.1.4.4.1.3 and F.1.4.4.2, we need at least
 * 49 statistics bins for DC, and 245 statistics bins for AC coding.
 *
 * We use a compact representation with 1 byte per statistics bin,
 * thus the numbers directly represent byte sizes.
 * This 1 byte per statistics bin contains the meaning of the MPS
 * (more probable symbol) in the highest bit (mask 0x80), and the
 * index into the probability estimation state machine table
 * in the lower bits (mask 0x7F).
 */
 #define DC_STAT_BINS 64
 #define AC_STAT_BINS 256
 typedef struct {
  float rate_dc[DC_STAT_BINS][2];
  float rate_ac[AC_STAT_BINS][2];
  int arith_dc_L;
  int arith_dc_U;
  int arith_ac_K;
 } arith_rates;
 #endif
 /* Main buffer control (downsampled-data buffer) */
 struct jpeg_c_main_controller {
  void (*start_pass) (j_compress_ptr cinfo, J_BUF_MODE pass_mode);
@@ -364,6 +390,16 @@ EXTERN(void) jcopy_sample_rows (JSAMPARRAY input_array, int source_row,
 EXTERN(void) jcopy_block_row (JBLOCKROW input_row, JBLOCKROW output_row,
                              JDIMENSION num_blocks);
 EXTERN(void) jzero_far (void * target, size_t bytestozero);
 #ifdef C_ARITH_CODING_SUPPORTED
 EXTERN(void) jget_arith_rates (j_compress_ptr cinfo, int dc_tbl_no, int ac_tbl_no, arith_rates *r);
 EXTERN(void) quantize_trellis_arith
 (j_compress_ptr cinfo, arith_rates *r, JBLOCKROW coef_blocks, JBLOCKROW src, JDIMENSION num_blocks,
 JQUANT_TBL * qtbl, double *norm_src, double *norm_coef, JCOEF *last_dc_val,
 JBLOCKROW coef_blocks_above, JBLOCKROW src_above);
 #endif
 /* Constant tables in jutils.c */
 #if 0                           /* This table is not actually needed in v6a */
 extern const int jpeg_zigzag_order[]; /* natural coef order to zigzag order */
--- a/jpeglib.h
+++ b/jpeglib.h
@@ -274,17 +274,12 @@ typedef enum {
 /* Boolean extension parameters */
 typedef enum {
  JBOOLEAN_USE_MOZ_DEFAULTS = 0xAE2F5D7F, /* TRUE=use Mozilla defaults */
  JBOOLEAN_OPTIMIZE_SCANS = 0x680C061E, /* TRUE=optimize progressive coding scans */
  JBOOLEAN_ONE_DC_SCAN = 0x3DA6A269, /* TRUE=use a single DC scan interleaving all components */
  JBOOLEAN_SEP_DC_SCAN = 0xE20DFA9F, /* TRUE=each DC scan is separate */
  JBOOLEAN_TRELLIS_QUANT = 0xC5122033, /* TRUE=use trellis quantization */
  JBOOLEAN_TRELLIS_QUANT_DC = 0x339D4C0C, /* TRUE=use trellis quant for DC coefficient */
  JBOOLEAN_TRELLIS_EOB_OPT = 0xD7F73780, /* TRUE=optimize for sequences of EOB */
  JBOOLEAN_USE_FLAT_QUANT_TBL = 0xE807EC6C, /* TRUE=use flat quantization table */
  JBOOLEAN_USE_LAMBDA_WEIGHT_TBL = 0x339DB65F, /* TRUE=use lambda weighting table */
  JBOOLEAN_USE_SCANS_IN_TRELLIS = 0xFD841435, /* TRUE=use scans in trellis optimization */
  JBOOLEAN_TRELLIS_PASSES = 0x3FF8A439, /* TRUE=currently doing trellis-related passes */
  JBOOLEAN_TRELLIS_Q_OPT = 0xE12AE269, /* TRUE=optimize quant table in trellis loop */
  JBOOLEAN_OVERSHOOT_DERINGING = 0x3F4BBBF9 /* TRUE=preprocess input to reduce ringing of edges on white background */
 } J_BOOLEAN_PARAM;
@@ -293,17 +288,29 @@ typedef enum {
 typedef enum {
  JFLOAT_LAMBDA_LOG_SCALE1 = 0x5B61A599,
-  JFLOAT_LAMBDA_LOG_SCALE2 = 0xB9BBAE03
+  JFLOAT_LAMBDA_LOG_SCALE2 = 0xB9BBAE03,
  JFLOAT_TRELLIS_DELTA_DC_WEIGHT = 0x13775453
 } J_FLOAT_PARAM;
 /* Integer parameters */
 typedef enum {
  JINT_COMPRESS_PROFILE = 0xE9918625, /* compression profile */
  JINT_TRELLIS_FREQ_SPLIT = 0x6FAFF127, /* splitting point for frequency in trellis quantization */
-  JINT_TRELLIS_NUM_LOOPS = 0xB63EBF39 /* number of trellis loops */
+  JINT_TRELLIS_NUM_LOOPS = 0xB63EBF39, /* number of trellis loops */
  JINT_BASE_QUANT_TBL_IDX = 0x44492AB1, /* base quantization table index */
  JINT_DC_SCAN_OPT_MODE = 0x0BE7AD3C /* DC scan optimization mode */
 } J_INT_PARAM;
 /* Values for the JINT_COMPRESS_PROFILE parameter (32-bit GUIDs) */
 enum {
  JCP_MAX_COMPRESSION = 0x5D083AAD, /* best compression ratio (progressive, all mozjpeg extensions) */
  JCP_FASTEST = 0x2AEA5CB4 /* libjpeg[-turbo] defaults (baseline, no mozjpeg extensions) */
 };
 /* Common fields between JPEG compression and decompression master structs. */
 #define jpeg_common_fields \
@@ -916,6 +923,16 @@ struct jpeg_memory_mgr {
 typedef boolean (*jpeg_marker_parser_method) (j_decompress_ptr cinfo);
 /* Originally, this macro was used as a way of defining function prototypes
 * for both modern compilers as well as older compilers that did not support
 * prototype parameters.  libjpeg-turbo has never supported these older,
 * non-ANSI compilers, but the macro is still included because there is some
 * software out there that uses it.
 */
 #define JPP(arglist)    arglist
 /* Default error-management setup */
 EXTERN(struct jpeg_error_mgr *) jpeg_std_error (struct jpeg_error_mgr * err);
@@ -949,7 +966,7 @@ EXTERN(void) jpeg_stdio_src (j_decompress_ptr cinfo, FILE * infile);
 /* Data source and destination managers: memory buffers. */
 EXTERN(void) jpeg_mem_dest (j_compress_ptr cinfo, unsigned char ** outbuffer,
                            unsigned long * outsize);
-EXTERN(void) jpeg_mem_src (j_decompress_ptr cinfo, unsigned char * inbuffer,
+EXTERN(void) jpeg_mem_src (j_decompress_ptr cinfo, const unsigned char * inbuffer,
                           unsigned long insize);
 #endif
@@ -1029,10 +1046,10 @@ EXTERN(JDIMENSION) jpeg_read_raw_data (j_decompress_ptr cinfo, JSAMPIMAGE data,
                                       JDIMENSION max_lines);
 /* Additional entry points for buffered-image mode. */
-EXTERN(boolean) jpeg_has_multiple_scans (j_decompress_ptr cinfo);
+EXTERN(boolean) jpeg_has_multiple_scans (const j_decompress_ptr cinfo);
 EXTERN(boolean) jpeg_start_output (j_decompress_ptr cinfo, int scan_number);
 EXTERN(boolean) jpeg_finish_output (j_decompress_ptr cinfo);
-EXTERN(boolean) jpeg_input_complete (j_decompress_ptr cinfo);
+EXTERN(boolean) jpeg_input_complete (const j_decompress_ptr cinfo);
 EXTERN(void) jpeg_new_colormap (j_decompress_ptr cinfo);
 EXTERN(int) jpeg_consume_input (j_decompress_ptr cinfo);
 /* Return value is one of: */
@@ -1061,7 +1078,7 @@ EXTERN(void) jpeg_set_marker_processor (j_decompress_ptr cinfo,
 EXTERN(jvirt_barray_ptr *) jpeg_read_coefficients (j_decompress_ptr cinfo);
 EXTERN(void) jpeg_write_coefficients (j_compress_ptr cinfo,
                                      jvirt_barray_ptr * coef_arrays);
-EXTERN(void) jpeg_copy_critical_parameters (j_decompress_ptr srcinfo,
+EXTERN(void) jpeg_copy_critical_parameters (const j_decompress_ptr srcinfo,
                                            j_compress_ptr dstinfo);
 /* If you choose to abort compression or decompression before completing
@@ -1083,25 +1100,25 @@ EXTERN(void) jpeg_destroy (j_common_ptr cinfo);
 EXTERN(boolean) jpeg_resync_to_restart (j_decompress_ptr cinfo, int desired);
 /* Accessor functions for extension parameters */
-EXTERN(boolean) jpeg_c_bool_param_supported (j_compress_ptr cinfo,
+EXTERN(boolean) jpeg_c_bool_param_supported (const j_compress_ptr cinfo,
                                             J_BOOLEAN_PARAM param);
 EXTERN(void) jpeg_c_set_bool_param (j_compress_ptr cinfo,
                                    J_BOOLEAN_PARAM param, boolean value);
-EXTERN(boolean) jpeg_c_get_bool_param (j_compress_ptr cinfo,
+EXTERN(boolean) jpeg_c_get_bool_param (const j_compress_ptr cinfo,
                                       J_BOOLEAN_PARAM param);
-EXTERN(boolean) jpeg_c_float_param_supported (j_compress_ptr cinfo,
+EXTERN(boolean) jpeg_c_float_param_supported (const j_compress_ptr cinfo,
                                              J_FLOAT_PARAM param);
 EXTERN(void) jpeg_c_set_float_param (j_compress_ptr cinfo, J_FLOAT_PARAM param,
                                     float value);
-EXTERN(float) jpeg_c_get_float_param (j_compress_ptr cinfo,
+EXTERN(float) jpeg_c_get_float_param (const j_compress_ptr cinfo,
                                      J_FLOAT_PARAM param);
-EXTERN(boolean) jpeg_c_int_param_supported (j_compress_ptr cinfo,
+EXTERN(boolean) jpeg_c_int_param_supported (const j_compress_ptr cinfo,
                                            J_INT_PARAM param);
 EXTERN(void) jpeg_c_set_int_param (j_compress_ptr cinfo, J_INT_PARAM param,
                                   int value);
-EXTERN(int) jpeg_c_get_int_param (j_compress_ptr cinfo, J_INT_PARAM param);
+EXTERN(int) jpeg_c_get_int_param (const j_compress_ptr cinfo, J_INT_PARAM param);
 /* These marker codes are exported since applications and data source modules
--- a/jpegtran.1
+++ b/jpegtran.1
@@ -1,4 +1,4 @@
-.TH JPEGTRAN 1 "1 January 2013"
+.TH JPEGTRAN 1 "21 November 2014"
 .SH NAME
 jpegtran \- lossless transformation of JPEG files
 .SH SYNOPSIS
@@ -214,6 +214,9 @@ give more output.  Also, version information is printed at startup.
 .B \-debug
 Same as
 .BR \-verbose .
 .TP
 .B \-version
 Print version information and exit.
 .SH EXAMPLES
 .LP
 This example converts a baseline JPEG file to progressive form:
--- a/jpegtran.c
+++ b/jpegtran.c
@@ -4,7 +4,7 @@
 * This file was part of the Independent JPEG Group's software:
 * Copyright (C) 1995-2010, Thomas G. Lane, Guido Vollbeding.
 * libjpeg-turbo Modifications:
- * Copyright (C) 2010, D. R. Commander.
+ * Copyright (C) 2010, 2014, D. R. Commander.
 * mozjpeg Modifications:
 * Copyright (C) 2014, Mozilla Corporation.
 * For conditions of distribution and use, see the accompanying README file.
@@ -44,7 +44,7 @@ static const char * progname;   /* program name for error messages */
 static char * outfilename;      /* for -outfile switch */
 static JCOPY_OPTION copyoption; /* -copy switch */
 static jpeg_transform_info transformoption; /* image transformation options */
-boolean memsrc;  /* for -memsrc switch */
+boolean memsrc = FALSE;  /* for -memsrc switch */
 #define INPUT_BUF_SIZE  4096
@@ -92,6 +92,7 @@ usage (void)
  fprintf(stderr, "  -maxmemory N   Maximum memory to use (in kbytes)\n");
  fprintf(stderr, "  -outfile name  Specify name for output file\n");
  fprintf(stderr, "  -verbose  or  -debug   Emit debug output\n");
  fprintf(stderr, "  -version       Print version information and exit\n");
  fprintf(stderr, "Switches for wizards:\n");
 #ifdef C_MULTISCAN_FILES_SUPPORTED
  fprintf(stderr, "  -scans file    Create multi-scan JPEG per script file\n");
@@ -174,6 +175,9 @@ parse_switches (j_compress_ptr cinfo, int argc, char **argv,
      /* Use arithmetic coding. */
 #ifdef C_ARITH_CODING_SUPPORTED
      cinfo->arith_code = TRUE;
      /* No table optimization required for AC */
      cinfo->optimize_coding = FALSE;
 #else
      fprintf(stderr, "%s: sorry, arithmetic coding not supported\n",
              progname);
@@ -222,6 +226,11 @@ parse_switches (j_compress_ptr cinfo, int argc, char **argv,
      }
      cinfo->err->trace_level++;
    } else if (keymatch(arg, "version", 4)) {
      fprintf(stderr, "%s version %s (build %s)\n",
              PACKAGE_NAME, VERSION, BUILD);
      exit(EXIT_SUCCESS);
    } else if (keymatch(arg, "flip", 1)) {
      /* Mirror left-right or top-bottom. */
      if (++argn >= argc)       /* advance to next argument */
@@ -310,7 +319,7 @@ parse_switches (j_compress_ptr cinfo, int argc, char **argv,
    } else if (keymatch(arg, "revert", 3)) {
      /* revert to old JPEG default */
-      jpeg_c_set_bool_param(cinfo, JBOOLEAN_USE_MOZ_DEFAULTS, FALSE);
+      jpeg_c_set_int_param(cinfo, JINT_COMPRESS_PROFILE, JCP_FASTEST);
    } else if (keymatch(arg, "rotate", 2)) {
      /* Rotate 90, 180, or 270 degrees (measured clockwise). */
@@ -415,8 +424,6 @@ main (int argc, char **argv)
  /* Initialize the JPEG compression object with default error handling. */
  dstinfo.err = jpeg_std_error(&jdsterr);
  jpeg_create_compress(&dstinfo);
  if (jpeg_c_bool_param_supported(&dstinfo, JBOOLEAN_USE_MOZ_DEFAULTS))
    jpeg_c_set_bool_param(&dstinfo, JBOOLEAN_USE_MOZ_DEFAULTS, TRUE);
  /* Scan command line to find file names.
   * It is convenient to use just one switch-parsing routine, but the switch
@@ -470,8 +477,10 @@ main (int argc, char **argv)
 #endif
  /* Specify data source for decompression */
-  if (jpeg_c_bool_param_supported(&dstinfo, JBOOLEAN_USE_MOZ_DEFAULTS))
+  if (jpeg_c_int_param_supported(&dstinfo, JINT_COMPRESS_PROFILE) &&
-    memsrc = jpeg_c_get_bool_param(&dstinfo, JBOOLEAN_USE_MOZ_DEFAULTS); /* needed to revert to original */
+      jpeg_c_get_int_param(&dstinfo, JINT_COMPRESS_PROFILE)
        == JCP_MAX_COMPRESSION)
    memsrc = TRUE; /* needed to revert to original */
 #if JPEG_LIB_VERSION >= 80 || defined(MEM_SRCDST_SUPPORTED)
  if (memsrc) {
    size_t nbytes;
@@ -557,8 +566,9 @@ main (int argc, char **argv)
  /* Specify data destination for compression */
 #if JPEG_LIB_VERSION >= 80 || defined(MEM_SRCDST_SUPPORTED)
-  if (jpeg_c_bool_param_supported(&dstinfo, JBOOLEAN_USE_MOZ_DEFAULTS) &&
+  if (jpeg_c_int_param_supported(&dstinfo, JINT_COMPRESS_PROFILE) &&
-      jpeg_c_get_bool_param(&dstinfo, JBOOLEAN_USE_MOZ_DEFAULTS))
+      jpeg_c_get_int_param(&dstinfo, JINT_COMPRESS_PROFILE)
        == JCP_MAX_COMPRESSION)
    jpeg_mem_dest(&dstinfo, &outbuffer, &outsize);
  else
 #endif
@@ -580,8 +590,9 @@ main (int argc, char **argv)
  /* Finish compression and release memory */
  jpeg_finish_compress(&dstinfo);
-  if (jpeg_c_bool_param_supported(&dstinfo, JBOOLEAN_USE_MOZ_DEFAULTS) &&
+  if (jpeg_c_int_param_supported(&dstinfo, JINT_COMPRESS_PROFILE) &&
-      jpeg_c_get_bool_param(&dstinfo, JBOOLEAN_USE_MOZ_DEFAULTS)) {
+      jpeg_c_get_int_param(&dstinfo, JINT_COMPRESS_PROFILE)
        == JCP_MAX_COMPRESSION) {
    size_t nbytes;
    unsigned char *buffer = outbuffer;
@@ -613,6 +624,9 @@ main (int argc, char **argv)
  end_progress_monitor((j_common_ptr) &dstinfo);
 #endif
  free(inbuffer);
  free(outbuffer);
  /* All done. */
  exit(jsrcerr.num_warnings + jdsterr.num_warnings ?EXIT_WARNING:EXIT_SUCCESS);
  return 0;                     /* suppress no-return-value warnings */
--- a/jpegyuv.c
+++ b/jpegyuv.c
@@ -53,7 +53,6 @@ int main(int argc, char *argv[]) {
  int chroma_width;
  int chroma_height;
  int frame_width;
  int frame_height;
  int yuv_size;
  JSAMPLE *image_buffer;
  JSAMPROW yrow_pointer[16];
@@ -110,10 +109,10 @@ int main(int argc, char *argv[]) {
  }
  frame_width = (cinfo.output_width + (16 - 1)) & ~(16 - 1);
  frame_height = (cinfo.output_height + (16 - 1)) & ~(16 - 1);
  image_buffer = malloc(frame_width*16 + 2*(frame_width/2)*8);
  if (!image_buffer) {
    free(yuv_buffer);
    fprintf(stderr, "Memory allocation failure!\n");
    return 1;
  }
@@ -165,6 +164,7 @@ int main(int argc, char *argv[]) {
  yuv_fd = fopen(yuv_path, "wb");
  if (!yuv_fd) {
    fprintf(stderr, "Invalid path to YUV file!");
    free(yuv_buffer);
    return 1;
  }
  if (fwrite(yuv_buffer, yuv_size, 1, yuv_fd) != 1) {
--- a/rdbmp.c
+++ b/rdbmp.c
@@ -381,7 +381,7 @@ start_input_bmp (j_compress_ptr cinfo, cjpeg_source_ptr sinfo)
    return;
  }
-  if (biWidth <= 0 || biHeight <= 0)
+  if (biWidth <= 0 || biHeight <= 0 || biWidth > 0x7fffffffL || biHeight > 0x7fffffffL)
    ERREXIT(cinfo, JERR_BMP_EMPTY);
  if (biPlanes != 1)
    ERREXIT(cinfo, JERR_BMP_BADPLANES);
--- a/rdpng.c
+++ b/rdpng.c
@@ -76,6 +76,10 @@ start_input_png (j_compress_ptr cinfo, cjpeg_source_ptr sinfo)
        cinfo->in_color_space = JCS_RGB;
        cinfo->input_components = 3;
    }
    if (bit_depth == 16)
        png_set_strip_16(source->png_ptr);
    cinfo->data_precision = 8;
    cinfo->image_width = width;
    cinfo->image_height = height;
--- a/rdppm.c
+++ b/rdppm.c
@@ -68,6 +68,7 @@ typedef struct {
  JSAMPROW pixrow;              /* compressor input buffer */
  size_t buffer_width;          /* width of I/O buffer */
  JSAMPLE *rescale;             /* => maxval-remapping array, or NULL */
  int maxval;
 } ppm_source_struct;
 typedef ppm_source_struct * ppm_source_ptr;
@@ -91,7 +92,7 @@ pbm_getc (FILE * infile)
 LOCAL(unsigned int)
-read_pbm_integer (j_compress_ptr cinfo, FILE * infile)
+read_pbm_integer (j_compress_ptr cinfo, FILE * infile, int maxval)
 /* Read an unsigned decimal integer from the PPM file */
 /* Swallows one trailing character after the integer */
 /* Note that on a 16-bit-int machine, only values up to 64k can be read. */
@@ -115,6 +116,10 @@ read_pbm_integer (j_compress_ptr cinfo, FILE * infile)
    val *= 10;
    val += ch - '0';
  }
  if (val > maxval)
    ERREXIT(cinfo, JERR_PPM_TOOLARGE);
  return val;
 }
@@ -139,10 +144,11 @@ get_text_gray_row (j_compress_ptr cinfo, cjpeg_source_ptr sinfo)
  register JSAMPROW ptr;
  register JSAMPLE *rescale = source->rescale;
  JDIMENSION col;
  int maxval = source->maxval;
  ptr = source->pub.buffer[0];
  for (col = cinfo->image_width; col > 0; col--) {
-    *ptr++ = rescale[read_pbm_integer(cinfo, infile)];
+    *ptr++ = rescale[read_pbm_integer(cinfo, infile, maxval)];
  }
  return 1;
 }
@@ -157,12 +163,13 @@ get_text_rgb_row (j_compress_ptr cinfo, cjpeg_source_ptr sinfo)
  register JSAMPROW ptr;
  register JSAMPLE *rescale = source->rescale;
  JDIMENSION col;
  int maxval = source->maxval;
  ptr = source->pub.buffer[0];
  for (col = cinfo->image_width; col > 0; col--) {
-    *ptr++ = rescale[read_pbm_integer(cinfo, infile)];
+    *ptr++ = rescale[read_pbm_integer(cinfo, infile, maxval)];
-    *ptr++ = rescale[read_pbm_integer(cinfo, infile)];
+    *ptr++ = rescale[read_pbm_integer(cinfo, infile, maxval)];
-    *ptr++ = rescale[read_pbm_integer(cinfo, infile)];
+    *ptr++ = rescale[read_pbm_integer(cinfo, infile, maxval)];
  }
  return 1;
 }
@@ -311,9 +318,9 @@ start_input_ppm (j_compress_ptr cinfo, cjpeg_source_ptr sinfo)
  }
  /* fetch the remaining header info */
-  w = read_pbm_integer(cinfo, source->pub.input_file);
+  w = read_pbm_integer(cinfo, source->pub.input_file, 65535);
-  h = read_pbm_integer(cinfo, source->pub.input_file);
+  h = read_pbm_integer(cinfo, source->pub.input_file, 65535);
-  maxval = read_pbm_integer(cinfo, source->pub.input_file);
+  maxval = read_pbm_integer(cinfo, source->pub.input_file, 65535);
  if (w <= 0 || h <= 0 || maxval <= 0) /* error check */
    ERREXIT(cinfo, JERR_PPM_NOT);
@@ -321,6 +328,7 @@ start_input_ppm (j_compress_ptr cinfo, cjpeg_source_ptr sinfo)
  cinfo->data_precision = BITS_IN_JSAMPLE; /* we always rescale data to this */
  cinfo->image_width = (JDIMENSION) w;
  cinfo->image_height = (JDIMENSION) h;
  source->maxval = maxval;
  /* initialize flags to most common settings */
  need_iobuffer = TRUE;         /* do we need an I/O buffer? */
--- a/rdswitch.c
+++ b/rdswitch.c
@@ -265,6 +265,9 @@ bogus:
    MEMCOPY(scanptr, scans, scanno * sizeof(jpeg_scan_info));
    cinfo->scan_info = scanptr;
    cinfo->num_scans = scanno;
    /* Disable scan optimization if using custom scan */
    jpeg_c_set_bool_param(cinfo, JBOOLEAN_OPTIMIZE_SCANS, FALSE);
  }
  fclose(fp);
@@ -279,7 +282,10 @@ bogus:
 * The spec says that the values given produce "good" quality, and
 * when divided by 2, "very good" quality.
 */
-static const unsigned int std_luminance_quant_tbl[DCTSIZE2] = {
+static const unsigned int std_luminance_quant_tbl[9][DCTSIZE2] = {
  {
    /* JPEG Annex K
     */
    16,  11,  10,  16,  24,  40,  51,  61,
    12,  12,  14,  19,  26,  58,  60,  55,
    14,  13,  16,  24,  40,  57,  69,  56,
@@ -288,19 +294,10 @@ static const unsigned int std_luminance_quant_tbl[DCTSIZE2] = {
    24,  35,  55,  64,  81, 104, 113,  92,
    49,  64,  78,  87, 103, 121, 120, 101,
    72,  92,  95,  98, 112, 100, 103,  99
-};
+  },
-static const unsigned int std_chrominance_quant_tbl[DCTSIZE2] = {
+  {
-  17,  18,  24,  47,  99,  99,  99,  99,
+    /* flat
-  18,  21,  26,  66,  99,  99,  99,  99,
+     */
  24,  26,  56,  99,  99,  99,  99,  99,
  47,  66,  99,  99,  99,  99,  99,  99,
  99,  99,  99,  99,  99,  99,  99,  99,
  99,  99,  99,  99,  99,  99,  99,  99,
  99,  99,  99,  99,  99,  99,  99,  99,
  99,  99,  99,  99,  99,  99,  99,  99
 };
 static const unsigned int flat_quant_tbl[DCTSIZE2] = {
    16,  16,  16,  16,  16,  16,  16,  16,
    16,  16,  16,  16,  16,  16,  16,  16,
    16,  16,  16,  16,  16,  16,  16,  16,
@@ -309,23 +306,211 @@ static const unsigned int flat_quant_tbl[DCTSIZE2] = {
    16,  16,  16,  16,  16,  16,  16,  16,
    16,  16,  16,  16,  16,  16,  16,  16,
    16,  16,  16,  16,  16,  16,  16,  16
  },
  {
    12, 17, 20, 21, 30, 34, 56, 63,
    18, 20, 20, 26, 28, 51, 61, 55,
    19, 20, 21, 26, 33, 58, 69, 55,
    26, 26, 26, 30, 46, 87, 86, 66,
    31, 33, 36, 40, 46, 96, 100, 73,
    40, 35, 46, 62, 81, 100, 111, 91,
    46, 66, 76, 86, 102, 121, 120, 101,
    68, 90, 90, 96, 113, 102, 105, 103
  },
  {
    /* From http://www.imagemagick.org/discourse-server/viewtopic.php?f=22&t=20333&p=98008#p98008
     */
    16,  16,  16,  18,  25,  37,  56,  85,
    16,  17,  20,  27,  34,  40,  53,  75,
    16,  20,  24,  31,  43,  62,  91,  135,
    18,  27,  31,  40,  53,  74,  106, 156,
    25,  34,  43,  53,  69,  94,  131, 189,
    37,  40,  62,  74,  94,  124, 169, 238,
    56,  53,  91,  106, 131, 169, 226, 311,
    85,  75,  135, 156, 189, 238, 311, 418
  },
  {
    9, 10, 12, 14, 27, 32, 51, 62,
    11, 12, 14, 19, 27, 44, 59, 73,
    12, 14, 18, 25, 42, 59, 79, 78,
    17, 18, 25, 42, 61, 92, 87, 92,
    23, 28, 42, 75, 79, 112, 112, 99,
    40, 42, 59, 84, 88, 124, 132, 111,
    42, 64, 78, 95, 105, 126, 125, 99,
    70, 75, 100, 102, 116, 100, 107, 98
  },
  {
    /* Relevance of human vision to JPEG-DCT compression (1992) Klein, Silverstein and Carney.
     */
    10, 12, 14, 19, 26, 38, 57, 86,
    12, 18, 21, 28, 35, 41, 54, 76,
    14, 21, 25, 32, 44, 63, 92, 136,
    19, 28, 32, 41, 54, 75, 107, 157,
    26, 35, 44, 54, 70, 95, 132, 190,
    38, 41, 63, 75, 95, 125, 170, 239,
    57, 54, 92, 107, 132, 170, 227, 312,
    86, 76, 136, 157, 190, 239, 312, 419
  },
  {
    /* DCTune perceptual optimization of compressed dental X-Rays (1997) Watson, Taylor, Borthwick
     */
    7, 8, 10, 14, 23, 44, 95, 241,
    8, 8, 11, 15, 25, 47, 102, 255,
    10, 11, 13, 19, 31, 58, 127, 255,
    14, 15, 19, 27, 44, 83, 181, 255,
    23, 25, 31, 44, 72, 136, 255, 255,
    44, 47, 58, 83, 136, 255, 255, 255,
    95, 102, 127, 181, 255, 255, 255, 255,
    241, 255, 255, 255, 255, 255, 255, 255
  },
  {
    /* A visual detection model for DCT coefficient quantization (12/9/93) Ahumada, Watson, Peterson
     */
    15, 11, 11, 12, 15, 19, 25, 32,
    11, 13, 10, 10, 12, 15, 19, 24,
    11, 10, 14, 14, 16, 18, 22, 27,
    12, 10, 14, 18, 21, 24, 28, 33,
    15, 12, 16, 21, 26, 31, 36, 42,
    19, 15, 18, 24, 31, 38, 45, 53,
    25, 19, 22, 28, 36, 45, 55, 65,
    32, 24, 27, 33, 42, 53, 65, 77
  },
  {
    /* An improved detection model for DCT coefficient quantization (1993) Peterson, Ahumada and Watson
     */
    14, 10, 11, 14, 19, 25, 34, 45,
    10, 11, 11, 12, 15, 20, 26, 33,
    11, 11, 15, 18, 21, 25, 31, 38,
    14, 12, 18, 24, 28, 33, 39, 47,
    19, 15, 21, 28, 36, 43, 51, 59,
    25, 20, 25, 33, 43, 54, 64, 74,
    34, 26, 31, 39, 51, 64, 77, 91,
    45, 33, 38, 47, 59, 74, 91, 108
  }
 };
 static const unsigned int std_chrominance_quant_tbl[9][DCTSIZE2] = {
  {
    /* JPEG Annex K
     */
    17,  18,  24,  47,  99,  99,  99,  99,
    18,  21,  26,  66,  99,  99,  99,  99,
    24,  26,  56,  99,  99,  99,  99,  99,
    47,  66,  99,  99,  99,  99,  99,  99,
    99,  99,  99,  99,  99,  99,  99,  99,
    99,  99,  99,  99,  99,  99,  99,  99,
    99,  99,  99,  99,  99,  99,  99,  99,
    99,  99,  99,  99,  99,  99,  99,  99
  },
  {
    /* flat
     */
    16,  16,  16,  16,  16,  16,  16,  16,
    16,  16,  16,  16,  16,  16,  16,  16,
    16,  16,  16,  16,  16,  16,  16,  16,
    16,  16,  16,  16,  16,  16,  16,  16,
    16,  16,  16,  16,  16,  16,  16,  16,
    16,  16,  16,  16,  16,  16,  16,  16,
    16,  16,  16,  16,  16,  16,  16,  16,
    16,  16,  16,  16,  16,  16,  16,  16
  },
  {
    8, 12, 15, 15, 86, 96, 96, 98,
    13, 13, 15, 26, 90, 96, 99, 98,
    12, 15, 18, 96, 99, 99, 99, 99,
    17, 16, 90, 96, 99, 99, 99, 99,
    96, 96, 99, 99, 99, 99, 99, 99,
    99, 99, 99, 99, 99, 99, 99, 99,
    99, 99, 99, 99, 99, 99, 99, 99,
    99, 99, 99, 99, 99, 99, 99, 99
  },
  {
    /* From http://www.imagemagick.org/discourse-server/viewtopic.php?f=22&t=20333&p=98008#p98008
     */
    16,  16,  16,  18,  25,  37,  56,  85,
    16,  17,  20,  27,  34,  40,  53,  75,
    16,  20,  24,  31,  43,  62,  91,  135,
    18,  27,  31,  40,  53,  74,  106, 156,
    25,  34,  43,  53,  69,  94,  131, 189,
    37,  40,  62,  74,  94,  124, 169, 238,
    56,  53,  91,  106, 131, 169, 226, 311,
    85,  75,  135, 156, 189, 238, 311, 418
  },
  {
    9, 10, 17, 19, 62, 89, 91, 97,
    12, 13, 18, 29, 84, 91, 88, 98,
    14, 19, 29, 93, 95, 95, 98, 97,
    20, 26, 84, 88, 95, 95, 98, 94,
    26, 86, 91, 93, 97, 99, 98, 99,
    99, 100, 98, 99, 99, 99, 99, 99,
    99, 99, 99, 99, 99, 99, 99, 99,
    97, 97, 99, 99, 99, 99, 97, 99
  },
  {
    /* Relevance of human vision to JPEG-DCT compression (1992) Klein, Silverstein and Carney.
     * Copied from luma
     */
    10, 12, 14, 19, 26, 38, 57, 86,
    12, 18, 21, 28, 35, 41, 54, 76,
    14, 21, 25, 32, 44, 63, 92, 136,
    19, 28, 32, 41, 54, 75, 107, 157,
    26, 35, 44, 54, 70, 95, 132, 190,
    38, 41, 63, 75, 95, 125, 170, 239,
    57, 54, 92, 107, 132, 170, 227, 312,
    86, 76, 136, 157, 190, 239, 312, 419
  },
  {
    /* DCTune perceptual optimization of compressed dental X-Rays (1997) Watson, Taylor, Borthwick
     * Copied from luma
     */
    7, 8, 10, 14, 23, 44, 95, 241,
    8, 8, 11, 15, 25, 47, 102, 255,
    10, 11, 13, 19, 31, 58, 127, 255,
    14, 15, 19, 27, 44, 83, 181, 255,
    23, 25, 31, 44, 72, 136, 255, 255,
    44, 47, 58, 83, 136, 255, 255, 255,
    95, 102, 127, 181, 255, 255, 255, 255,
    241, 255, 255, 255, 255, 255, 255, 255
  },
  {
    /* A visual detection model for DCT coefficient quantization (12/9/93) Ahumada, Watson, Peterson
     * Copied from luma
     */
    15, 11, 11, 12, 15, 19, 25, 32,
    11, 13, 10, 10, 12, 15, 19, 24,
    11, 10, 14, 14, 16, 18, 22, 27,
    12, 10, 14, 18, 21, 24, 28, 33,
    15, 12, 16, 21, 26, 31, 36, 42,
    19, 15, 18, 24, 31, 38, 45, 53,
    25, 19, 22, 28, 36, 45, 55, 65,
    32, 24, 27, 33, 42, 53, 65, 77
  },
  {
    /* An improved detection model for DCT coefficient quantization (1993) Peterson, Ahumada and Watson
     * Copied from luma
     */
    14, 10, 11, 14, 19, 25, 34, 45,
    10, 11, 11, 12, 15, 20, 26, 33,
    11, 11, 15, 18, 21, 25, 31, 38,
    14, 12, 18, 24, 28, 33, 39, 47,
    19, 15, 21, 28, 36, 43, 51, 59,
    25, 20, 25, 33, 43, 54, 64, 74,
    34, 26, 31, 39, 51, 64, 77, 91,
    45, 33, 38, 47, 59, 74, 91, 108
  }
 };
 LOCAL(void)
 jpeg_default_qtables (j_compress_ptr cinfo, boolean force_baseline)
 {
-  if (jpeg_c_bool_param_supported(cinfo, JBOOLEAN_USE_FLAT_QUANT_TBL) &&
+  int quant_tbl_master_idx = 0;
-      jpeg_c_get_bool_param(cinfo, JBOOLEAN_USE_FLAT_QUANT_TBL)) {
+  if (jpeg_c_int_param_supported(cinfo, JINT_BASE_QUANT_TBL_IDX))
-    jpeg_add_quant_table(cinfo, 0, flat_quant_tbl,
+    quant_tbl_master_idx = jpeg_c_get_int_param(cinfo, JINT_BASE_QUANT_TBL_IDX);
  jpeg_add_quant_table(cinfo, 0, std_luminance_quant_tbl[quant_tbl_master_idx],
                       q_scale_factor[0], force_baseline);
-    jpeg_add_quant_table(cinfo, 1, flat_quant_tbl,
+  jpeg_add_quant_table(cinfo, 1, std_chrominance_quant_tbl[quant_tbl_master_idx],
                       q_scale_factor[1], force_baseline);
  } else {
    jpeg_add_quant_table(cinfo, 0, std_luminance_quant_tbl,
                         q_scale_factor[0], force_baseline);
    jpeg_add_quant_table(cinfo, 1, std_chrominance_quant_tbl,
                         q_scale_factor[1], force_baseline);
  }
 }
 #endif
--- a/rdtarga.c
+++ b/rdtarga.c
@@ -364,7 +364,8 @@ start_input_tga (j_compress_ptr cinfo, cjpeg_source_ptr sinfo)
  if (cmaptype > 1 ||           /* cmaptype must be 0 or 1 */
      source->pixel_size < 1 || source->pixel_size > 4 ||
      (UCH(targaheader[16]) & 7) != 0 || /* bits/pixel must be multiple of 8 */
-      interlace_type != 0)      /* currently don't allow interlaced image */
+      interlace_type != 0 ||      /* currently don't allow interlaced image */
      width == 0 || height == 0)  /* image width/height must be nonzero */
    ERREXIT(cinfo, JERR_TGA_BADPARMS);
  if (subtype > 8) {
--- a/release/Distribution.xml
+++ b/release/Distribution.xml
@@ -1,6 +1,6 @@
 <?xml version="1.0" encoding="utf-8"?>
 <installer-gui-script minSpecVersion="1">
-	<title>libjpeg-turbo</title>
+	<title>mozjpeg</title>
 	<welcome file="Welcome.rtf" />
 	<readme file="ReadMe.txt" />
 	<license file="License.rtf" />
@@ -12,13 +12,13 @@
 	<options customize="never" />
 	<choices-outline>
 		<line choice="default">
-			<line choice="com.libjpeg-turbo.libjpeg-turbo"/>
+			<line choice="com.mozilla.mozjpeg"/>
 		</line>
 	</choices-outline>
 	<choice id="default"/>
-	<choice id="com.libjpeg-turbo.libjpeg-turbo" visible="false">
+	<choice id="com.mozilla.mozjpeg" visible="false">
-		<pkg-ref id="com.libjpeg-turbo.libjpeg-turbo"/>
+		<pkg-ref id="com.mozilla.mozjpeg"/>
 	</choice>
 	<pkg-ref auth="root"
-		id="com.libjpeg-turbo.libjpeg-turbo">libjpeg-turbo.pkg</pkg-ref>
+		id="com.mozilla.mozjpeg">mozjpeg.pkg</pkg-ref>
 </installer-gui-script>
--- a/release/License.rtf
+++ b/release/License.rtf
@@ -12,7 +12,7 @@
 \pard\tx220\tx720\pardeftab720\li720\fi-720
 \ls1\ilvl0\cf0 {\listtext	\'95	}Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer.\
 {\listtext	\'95	}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.\
-{\listtext	\'95	}Neither the name of the libmozjpeg Project nor the names of its contributors may be used to endorse or promote products derived from this software without specific prior written permission.\
+{\listtext	\'95	}Neither the name of the mozjpeg Project nor the names of its contributors may be used to endorse or promote products derived from this software without specific prior written permission.\
 \pard\pardeftab720\qc
 \cf0 \
 \pard\pardeftab720
--- a/release/ReadMe.txt
+++ b/release/ReadMe.txt
@@ -1,5 +1,3 @@
-libjpeg-turbo is a JPEG image codec that uses SIMD instructions (MMX, SSE2, NEON) to accelerate baseline JPEG compression and decompression on x86, x86-64, and ARM systems.  On such systems, libjpeg-turbo is generally 2-4x as fast as libjpeg, all else being equal.  On other types of systems, libjpeg-turbo can still outperform libjpeg by a significant amount, by virtue of its highly-optimized Huffman coding routines.  In many cases, the performance of libjpeg-turbo rivals that of proprietary high-speed JPEG codecs.
+mozjpeg is a fork of libjpeg-turbo that aims to speed up load times of web pages by reducing the size (and, by extension, the transmission time) of JPEG files.  It accomplishes this by enabling optimized Huffman trees and progressive entropy coding by default in the JPEG compressor, as well as splitting the spectrum of DCT coefficients into separate scans and using Trellis quantisation.
-libjpeg-turbo implements both the traditional libjpeg API as well as the less powerful but more straightforward TurboJPEG API.  libjpeg-turbo also features colorspace extensions that allow it to compress from/decompress to 32-bit and big-endian pixel buffers (RGBX, XBGR, etc.), as well as a full-featured Java interface.
+Although it is based on libjpeg-turbo, mozjpeg is not intended to be a general-purpose or high-performance JPEG library.  Its performance is highly "asymmetric".  That is, the JPEG files it generates require much more time to compress than to decompress.  When the default settings are used, mozjpeg is considerably slower than libjpeg-turbo or even libjpeg at compressing images.  Thus, it is not generally suitable for real-time compression.  It is best used as part of a web encoding workflow.
 libjpeg-turbo was originally based on libjpeg/SIMD, an MMX-accelerated derivative of libjpeg v6b developed by Miyasaka Masaru.  The TigerVNC and VirtualGL projects made numerous enhancements to the codec in 2009, and in early 2010, libjpeg-turbo spun off into an independent project, with the goal of making high-speed JPEG compression/decompression technology available to a broader range of users and developers.
--- a/release/Welcome.rtf
+++ b/release/Welcome.rtf
@@ -5,11 +5,11 @@
 \deftab720
 \pard\pardeftab720
-\f0\fs24 \cf0 This installer will install the libmozjpeg SDK and run-time libraries onto your computer so that you can use libmozjpeg to build new applications or accelerate existing ones.  To remove the libmozjpeg package, run\
+\f0\fs24 \cf0 This installer will install the mozjpeg SDK and run-time libraries onto your computer so that you can use mozjpeg to build new applications.  To remove the mozjpeg package, run\
 \
 \pard\pardeftab720
-\f1 \cf0   /opt/libmozjpeg/bin/uninstall\
+\f1 \cf0   /opt/mozjpeg/bin/uninstall\
 \pard\pardeftab720
 \f0 \cf0 \
--- a/release/deb-control.tmpl
+++ b/release/deb-control.tmpl
@@ -4,28 +4,21 @@ Section: misc
 Priority: optional
 Architecture: {__ARCH}
 Essential: no
-Maintainer: The libmozjpeg Project <joshmoz@gmail.com>
+Maintainer: Mozilla Research <joshmoz@gmail.com>
 Homepage: https://github.com/mozilla/mozjpeg
 Installed-Size: {__SIZE}
-Description: A SIMD-accelerated JPEG codec that provides both the libjpeg and TurboJPEG APIs
+Description: A JPEG codec that provides increased compression for JPEG images (at the expense of compression performance)
- libmozjpeg is a JPEG image codec that uses SIMD instructions (MMX, SSE2,
+ mozjpeg is a fork of libjpeg-turbo that aims to speed up load times of web
- NEON) to accelerate baseline JPEG compression and decompression on x86,
+ pages by reducing the size (and, by extension, the transmission time) of JPEG
- x86-64, and ARM systems.  On such systems, libmozjpeg is generally 2-4x as
+ files.  It accomplishes this by enabling optimized Huffman trees and
- fast as libjpeg, all else being equal.  On other types of systems,
+ progressive entropy coding by default in the JPEG compressor, as well as
- libmozjpeg can still outperform libjpeg by a significant amount, by virtue
+ splitting the spectrum of DCT coefficients into separate scans and using
- of its highly-optimized Huffman coding routines.  In many cases, the
+ Trellis quantisation.
 performance of libmozjpeg rivals that of proprietary high-speed JPEG
 codecs.
 .
- libmozjpeg implements both the traditional libjpeg API as well as the less
+ Although it is based on libjpeg-turbo, mozjpeg is not intended to be a
- powerful but more straightforward TurboJPEG API.  libmozjpeg also features
+ general-purpose or high-performance JPEG library.  Its performance is highly
- colorspace extensions that allow it to compress from/decompress to 32-bit and
+ "asymmetric".  That is, the JPEG files it generates require much more time to
- big-endian pixel buffers (RGBX, XBGR, etc.), as well as a full-featured Java
+ compress than to decompress.  When the default settings are used, mozjpeg is
- interface.
+ considerably slower than libjpeg-turbo or even libjpeg at compressing images.
- .
+ Thus, it is not generally suitable for real-time compression.  It is best used
- libmozjpeg was forked from libjpeg-turbo. libjpeg-turbo was originally based on
+ as part of a web encoding workflow.
 libjpeg/SIMD, an MMX-accelerated derivative of libjpeg v6b developed by Miyasaka
 Masaru.  The TigerVNC and VirtualGL projects made numerous enhancements to the
 codec in 2009, and in early 2010, libjpeg-turbo spun off into an independent
 project, with the goal of making high-speed JPEG compression/decompression technology
 available to a broader range of users and developers.
--- a/release/makecygwinpkg.in
+++ b/release/makecygwinpkg.in
@@ -31,7 +31,7 @@ __PWD=`pwd`
 make install DESTDIR=$TMPDIR/pkg docdir=/usr/share/doc/$PACKAGE_NAME-$VERSION \
 	exampledir=/usr/share/doc/$PACKAGE_NAME-$VERSION
 rm $TMPDIR/pkg$LIBDIR/*.la
-if [ "$PREFIX" = "/opt/libmozjpeg" -a "$DOCDIR" = "/opt/libmozjpeg/doc" ]; then
+if [ "$PREFIX" = "/opt/mozjpeg" -a "$DOCDIR" = "/opt/mozjpeg/doc" ]; then
 	ln -fs /usr/share/doc/$PACKAGE_NAME-$VERSION $TMPDIR/pkg$DOCDIR
 fi
 cd $TMPDIR/pkg
--- a/release/makedpkg.in
+++ b/release/makedpkg.in
@@ -45,7 +45,7 @@ makedeb()
 		make install DESTDIR=$TMPDIR docdir=/usr/share/doc/$DIRNAME-$VERSION \
 			exampledir=/usr/share/doc/$DIRNAME-$VERSION
 		rm -f $TMPDIR$LIBDIR/*.la
-		if [ "$PREFIX" = "/opt/libmozjpeg" -a "$DOCDIR" = "/opt/libmozjpeg/doc" ]; then
+		if [ "$PREFIX" = "/opt/mozjpeg" -a "$DOCDIR" = "/opt/mozjpeg/doc" ]; then
 			ln -fs /usr/share/doc/$DIRNAME-$VERSION $TMPDIR$DOCDIR
 		fi
 	fi
--- a/release/makemacpkg.in
+++ b/release/makemacpkg.in
@@ -17,7 +17,7 @@ onexit()
 usage()
 {
-	echo "$0 [-build32 [32-bit build dir]] [-buildarmv6 [ARM v6 build dir]] [-buildarmv7 [ARM v7 build dir]] [-buildarmv7s [ARM v7s build dir]]"
+	echo "$0 [-build32 [32-bit build dir]] [-buildarmv6 [ARMv6 build dir]] [-buildarmv7 [ARMv7 build dir]] [-buildarmv7s [ARMv7s build dir] [-buildarmv8 [ARMv8 build dir]] [-lipo [path to lipo]]"
 	exit 1
 }
@@ -33,7 +33,10 @@ BUILDDIRARMV7=@abs_top_srcdir@/iosarmv7
 BUILDARMV7=0
 BUILDDIRARMV7S=@abs_top_srcdir@/iosarmv7s
 BUILDARMV7S=0
 BUILDDIRARMV8=@abs_top_srcdir@/iosarmv8
 BUILDARMV8=0
 WITH_JAVA=@WITH_JAVA@
 LIPO=lipo
 PREFIX=%{__prefix}
 BINDIR=%{__bindir}
@@ -75,6 +78,21 @@ while [ $# -gt 0 ]; do
 			fi
 		fi
 		;;
 	-buildarmv8)
 		BUILDARMV8=1
 		if [ $# -gt 1 ]; then
 			if [[ ! "$2" =~ -.* ]]; then
 				BUILDDIRARMV8=$2;  shift
 			fi
 		fi
 		;;
 	-lipo)
 		if [ $# -gt 1 ]; then
 			if [[ ! "$2" =~ -.* ]]; then
 				LIPO=$2;  shift
 			fi
 		fi
 		;;
 	esac
 	shift
 done
@@ -91,7 +109,7 @@ make install DESTDIR=$PKGROOT docdir=/Library/Documentation/$PACKAGE_NAME \
 	exampledir=/Library/Documentation/$PACKAGE_NAME
 rm -f $PKGROOT$LIBDIR/*.la
-if [ "$PREFIX" = "/opt/libmozjpeg" -a "$DOCDIR" = "/opt/libmozjpeg/doc" ]; then
+if [ "$PREFIX" = "/opt/mozjpeg" -a "$DOCDIR" = "/opt/mozjpeg/doc" ]; then
 	ln -fs /Library/Documentation/$PACKAGE_NAME $PKGROOT$DOCDIR
 fi
@@ -110,50 +128,50 @@ if [ $BUILD32 = 1 ]; then
 	popd
 	if [ ! -h $TMPDIR/dist.x86/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib -a \
 		! -h $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib ]; then
-		lipo -create \
+		$LIPO -create \
 			-arch i386 $TMPDIR/dist.x86/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib \
 			-arch x86_64 $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib \
 			-output $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib
 	elif [ ! -h $TMPDIR/dist.x86/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib -a \
 		! -h $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib ]; then
-		lipo -create \
+		$LIPO -create \
 			-arch i386 $TMPDIR/dist.x86/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib \
 			-arch x86_64 $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib \
 			-output $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib
 	fi
-	lipo -create \
+	$LIPO -create \
 		-arch i386 $TMPDIR/dist.x86/$LIBDIR/libjpeg.a \
 		-arch x86_64 $PKGROOT/$LIBDIR/libjpeg.a \
 		-output $PKGROOT/$LIBDIR/libjpeg.a
-	lipo -create \
+	$LIPO -create \
 		-arch i386 $TMPDIR/dist.x86/$LIBDIR/libturbojpeg.0.dylib \
 		-arch x86_64 $PKGROOT/$LIBDIR/libturbojpeg.0.dylib \
 		-output $PKGROOT/$LIBDIR/libturbojpeg.0.dylib
-	lipo -create \
+	$LIPO -create \
 		-arch i386 $TMPDIR/dist.x86/$LIBDIR/libturbojpeg.a \
 		-arch x86_64 $PKGROOT/$LIBDIR/libturbojpeg.a \
 		-output $PKGROOT/$LIBDIR/libturbojpeg.a
-	lipo -create \
+	$LIPO -create \
 		-arch i386 $TMPDIR/dist.x86/$BINDIR/cjpeg \
 		-arch x86_64 $PKGROOT/$BINDIR/cjpeg \
 		-output $PKGROOT/$BINDIR/cjpeg
-	lipo -create \
+	$LIPO -create \
 		-arch i386 $TMPDIR/dist.x86/$BINDIR/djpeg \
 		-arch x86_64 $PKGROOT/$BINDIR/djpeg \
 		-output $PKGROOT/$BINDIR/djpeg
-	lipo -create \
+	$LIPO -create \
 		-arch i386 $TMPDIR/dist.x86/$BINDIR/jpegtran \
 		-arch x86_64 $PKGROOT/$BINDIR/jpegtran \
 		-output $PKGROOT/$BINDIR/jpegtran
-	lipo -create \
+	$LIPO -create \
 		-arch i386 $TMPDIR/dist.x86/$BINDIR/tjbench \
 		-arch x86_64 $PKGROOT/$BINDIR/tjbench \
 		-output $PKGROOT/$BINDIR/tjbench
-	lipo -create \
+	$LIPO -create \
 		-arch i386 $TMPDIR/dist.x86/$BINDIR/rdjpgcom \
 		-arch x86_64 $PKGROOT/$BINDIR/rdjpgcom \
 		-output $PKGROOT/$BINDIR/rdjpgcom
-	lipo -create \
+	$LIPO -create \
 		-arch i386 $TMPDIR/dist.x86/$BINDIR/wrjpgcom \
 		-arch x86_64 $PKGROOT/$BINDIR/wrjpgcom \
 		-output $PKGROOT/$BINDIR/wrjpgcom
@@ -162,71 +180,258 @@ fi
 if [ $BUILDARMV6 = 1 ]; then
 	if [ ! -d $BUILDDIRARMV6 ]; then
-		echo ERROR: ARM v6 build directory $BUILDDIRARMV6 does not exist
+		echo ERROR: ARMv6 build directory $BUILDDIRARMV6 does not exist
 		exit 1
 	fi
 	if [ ! -f $BUILDDIRARMV6/Makefile ]; then
-		echo ERROR: ARM v6 build directory $BUILDDIRARMV6 is not configured
+		echo ERROR: ARMv6 build directory $BUILDDIRARMV6 is not configured
 		exit 1
 	fi
 	mkdir -p $TMPDIR/dist.armv6
 	pushd $BUILDDIRARMV6
 	make install DESTDIR=$TMPDIR/dist.armv6
 	popd
-	lipo -create \
+	if [ ! -h $TMPDIR/dist.armv6/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib -a \
 		! -h $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib ]; then
 		$LIPO -create \
 			$PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib \
 			-arch arm $TMPDIR/dist.armv6/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib \
 			-output $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib
 	elif [ ! -h $TMPDIR/dist.armv6/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib -a \
 		! -h $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib ]; then
 		$LIPO -create \
 			$PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib \
 			-arch arm $TMPDIR/dist.armv6/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib \
 			-output $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib
 	fi
 	$LIPO -create \
 		$PKGROOT/$LIBDIR/libjpeg.a \
 		-arch arm $TMPDIR/dist.armv6/$LIBDIR/libjpeg.a \
 		-output $PKGROOT/$LIBDIR/libjpeg.a
-	lipo -create \
+	$LIPO -create \
 		$PKGROOT/$LIBDIR/libturbojpeg.0.dylib \
 		-arch arm $TMPDIR/dist.armv6/$LIBDIR/libturbojpeg.0.dylib \
 		-output $PKGROOT/$LIBDIR/libturbojpeg.0.dylib
 	$LIPO -create \
 		$PKGROOT/$LIBDIR/libturbojpeg.a \
 		-arch arm $TMPDIR/dist.armv6/$LIBDIR/libturbojpeg.a \
 		-output $PKGROOT/$LIBDIR/libturbojpeg.a
 	$LIPO -create \
 		$PKGROOT/$BINDIR/cjpeg \
 		-arch arm $TMPDIR/dist.armv6/$BINDIR/cjpeg \
 		-output $PKGROOT/$BINDIR/cjpeg
 	$LIPO -create \
 		$PKGROOT/$BINDIR/djpeg \
 		-arch arm $TMPDIR/dist.armv6/$BINDIR/djpeg \
 		-output $PKGROOT/$BINDIR/djpeg
 	$LIPO -create \
 		$PKGROOT/$BINDIR/jpegtran \
 		-arch arm $TMPDIR/dist.armv6/$BINDIR/jpegtran \
 		-output $PKGROOT/$BINDIR/jpegtran
 	$LIPO -create \
 		$PKGROOT/$BINDIR/tjbench \
 		-arch arm $TMPDIR/dist.armv6/$BINDIR/tjbench \
 		-output $PKGROOT/$BINDIR/tjbench
 	$LIPO -create \
 		$PKGROOT/$BINDIR/rdjpgcom \
 		-arch arm $TMPDIR/dist.armv6/$BINDIR/rdjpgcom \
 		-output $PKGROOT/$BINDIR/rdjpgcom
 	$LIPO -create \
 		$PKGROOT/$BINDIR/wrjpgcom \
 		-arch arm $TMPDIR/dist.armv6/$BINDIR/wrjpgcom \
 		-output $PKGROOT/$BINDIR/wrjpgcom
 fi
 if [ $BUILDARMV7 = 1 ]; then
 	if [ ! -d $BUILDDIRARMV7 ]; then
-		echo ERROR: ARM v7 build directory $BUILDDIRARMV7 does not exist
+		echo ERROR: ARMv7 build directory $BUILDDIRARMV7 does not exist
 		exit 1
 	fi
 	if [ ! -f $BUILDDIRARMV7/Makefile ]; then
-		echo ERROR: ARM v7 build directory $BUILDDIRARMV7 is not configured
+		echo ERROR: ARMv7 build directory $BUILDDIRARMV7 is not configured
 		exit 1
 	fi
 	mkdir -p $TMPDIR/dist.armv7
 	pushd $BUILDDIRARMV7
 	make install DESTDIR=$TMPDIR/dist.armv7
 	popd
-	lipo -create \
+	if [ ! -h $TMPDIR/dist.armv7/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib -a \
 		! -h $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib ]; then
 		$LIPO -create \
 			$PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib \
 			-arch arm $TMPDIR/dist.armv7/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib \
 			-output $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib
 	elif [ ! -h $TMPDIR/dist.armv7/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib -a \
 		! -h $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib ]; then
 		$LIPO -create \
 			$PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib \
 			-arch arm $TMPDIR/dist.armv7/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib \
 			-output $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib
 	fi
 	$LIPO -create \
 		$PKGROOT/$LIBDIR/libjpeg.a \
 		-arch arm $TMPDIR/dist.armv7/$LIBDIR/libjpeg.a \
 		-output $PKGROOT/$LIBDIR/libjpeg.a
-	lipo -create \
+	$LIPO -create \
 		$PKGROOT/$LIBDIR/libturbojpeg.0.dylib \
 		-arch arm $TMPDIR/dist.armv7/$LIBDIR/libturbojpeg.0.dylib \
 		-output $PKGROOT/$LIBDIR/libturbojpeg.0.dylib
 	$LIPO -create \
 		$PKGROOT/$LIBDIR/libturbojpeg.a \
 		-arch arm $TMPDIR/dist.armv7/$LIBDIR/libturbojpeg.a \
 		-output $PKGROOT/$LIBDIR/libturbojpeg.a
 	$LIPO -create \
 		$PKGROOT/$BINDIR/cjpeg \
 		-arch arm $TMPDIR/dist.armv7/$BINDIR/cjpeg \
 		-output $PKGROOT/$BINDIR/cjpeg
 	$LIPO -create \
 		$PKGROOT/$BINDIR/djpeg \
 		-arch arm $TMPDIR/dist.armv7/$BINDIR/djpeg \
 		-output $PKGROOT/$BINDIR/djpeg
 	$LIPO -create \
 		$PKGROOT/$BINDIR/jpegtran \
 		-arch arm $TMPDIR/dist.armv7/$BINDIR/jpegtran \
 		-output $PKGROOT/$BINDIR/jpegtran
 	$LIPO -create \
 		$PKGROOT/$BINDIR/tjbench \
 		-arch arm $TMPDIR/dist.armv7/$BINDIR/tjbench \
 		-output $PKGROOT/$BINDIR/tjbench
 	$LIPO -create \
 		$PKGROOT/$BINDIR/rdjpgcom \
 		-arch arm $TMPDIR/dist.armv7/$BINDIR/rdjpgcom \
 		-output $PKGROOT/$BINDIR/rdjpgcom
 	$LIPO -create \
 		$PKGROOT/$BINDIR/wrjpgcom \
 		-arch arm $TMPDIR/dist.armv7/$BINDIR/wrjpgcom \
 		-output $PKGROOT/$BINDIR/wrjpgcom
 fi
 if [ $BUILDARMV7S = 1 ]; then
 	if [ ! -d $BUILDDIRARMV7S ]; then
-		echo ERROR: ARM v7s build directory $BUILDDIRARMV7S does not exist
+		echo ERROR: ARMv7s build directory $BUILDDIRARMV7S does not exist
 		exit 1
 	fi
 	if [ ! -f $BUILDDIRARMV7S/Makefile ]; then
-		echo ERROR: ARM v7s build directory $BUILDDIRARMV7S is not configured
+		echo ERROR: ARMv7s build directory $BUILDDIRARMV7S is not configured
 		exit 1
 	fi
 	mkdir -p $TMPDIR/dist.armv7s
 	pushd $BUILDDIRARMV7S
 	make install DESTDIR=$TMPDIR/dist.armv7s
 	popd
-	lipo -create \
+	if [ ! -h $TMPDIR/dist.armv7s/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib -a \
 		! -h $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib ]; then
 		$LIPO -create \
 			$PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib \
 			-arch arm $TMPDIR/dist.armv7s/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib \
 			-output $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib
 	elif [ ! -h $TMPDIR/dist.armv7s/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib -a \
 		! -h $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib ]; then
 		$LIPO -create \
 			$PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib \
 			-arch arm $TMPDIR/dist.armv7s/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib \
 			-output $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib
 	fi
 	$LIPO -create \
 		$PKGROOT/$LIBDIR/libjpeg.a \
 		-arch arm $TMPDIR/dist.armv7s/$LIBDIR/libjpeg.a \
 		-output $PKGROOT/$LIBDIR/libjpeg.a
-	lipo -create \
+	$LIPO -create \
 		$PKGROOT/$LIBDIR/libturbojpeg.0.dylib \
 		-arch arm $TMPDIR/dist.armv7s/$LIBDIR/libturbojpeg.0.dylib \
 		-output $PKGROOT/$LIBDIR/libturbojpeg.0.dylib
 	$LIPO -create \
 		$PKGROOT/$LIBDIR/libturbojpeg.a \
 		-arch arm $TMPDIR/dist.armv7s/$LIBDIR/libturbojpeg.a \
 		-output $PKGROOT/$LIBDIR/libturbojpeg.a
 	$LIPO -create \
 		$PKGROOT/$BINDIR/cjpeg \
 		-arch arm $TMPDIR/dist.armv7s/$BINDIR/cjpeg \
 		-output $PKGROOT/$BINDIR/cjpeg
 	$LIPO -create \
 		$PKGROOT/$BINDIR/djpeg \
 		-arch arm $TMPDIR/dist.armv7s/$BINDIR/djpeg \
 		-output $PKGROOT/$BINDIR/djpeg
 	$LIPO -create \
 		$PKGROOT/$BINDIR/jpegtran \
 		-arch arm $TMPDIR/dist.armv7s/$BINDIR/jpegtran \
 		-output $PKGROOT/$BINDIR/jpegtran
 	$LIPO -create \
 		$PKGROOT/$BINDIR/tjbench \
 		-arch arm $TMPDIR/dist.armv7s/$BINDIR/tjbench \
 		-output $PKGROOT/$BINDIR/tjbench
 	$LIPO -create \
 		$PKGROOT/$BINDIR/rdjpgcom \
 		-arch arm $TMPDIR/dist.armv7s/$BINDIR/rdjpgcom \
 		-output $PKGROOT/$BINDIR/rdjpgcom
 	$LIPO -create \
 		$PKGROOT/$BINDIR/wrjpgcom \
 		-arch arm $TMPDIR/dist.armv7s/$BINDIR/wrjpgcom \
 		-output $PKGROOT/$BINDIR/wrjpgcom
 fi
 if [ $BUILDARMV8 = 1 ]; then
 	if [ ! -d $BUILDDIRARMV8 ]; then
 		echo ERROR: ARMv8 build directory $BUILDDIRARMV8 does not exist
 		exit 1
 	fi
 	if [ ! -f $BUILDDIRARMV8/Makefile ]; then
 		echo ERROR: ARMv8 build directory $BUILDDIRARMV8 is not configured
 		exit 1
 	fi
 	mkdir -p $TMPDIR/dist.armv8
 	pushd $BUILDDIRARMV8
 	make install DESTDIR=$TMPDIR/dist.armv8
 	popd
 	if [ ! -h $TMPDIR/dist.armv8/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib -a \
 		! -h $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib ]; then
 		$LIPO -create \
 			$PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib \
 			-arch arm64 $TMPDIR/dist.armv8/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib \
 			-output $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib
 	elif [ ! -h $TMPDIR/dist.armv8/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib -a \
 		! -h $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib ]; then
 		$LIPO -create \
 			$PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib \
 			-arch arm64 $TMPDIR/dist.armv8/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib \
 			-output $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib
 	fi
 	$LIPO -create \
 		$PKGROOT/$LIBDIR/libjpeg.a \
 		-arch arm64 $TMPDIR/dist.armv8/$LIBDIR/libjpeg.a \
 		-output $PKGROOT/$LIBDIR/libjpeg.a
 	$LIPO -create \
 		$PKGROOT/$LIBDIR/libturbojpeg.0.dylib \
 		-arch arm64 $TMPDIR/dist.armv8/$LIBDIR/libturbojpeg.0.dylib \
 		-output $PKGROOT/$LIBDIR/libturbojpeg.0.dylib
 	$LIPO -create \
 		$PKGROOT/$LIBDIR/libturbojpeg.a \
 		-arch arm64 $TMPDIR/dist.armv8/$LIBDIR/libturbojpeg.a \
 		-output $PKGROOT/$LIBDIR/libturbojpeg.a
 	$LIPO -create \
 		$PKGROOT/$BINDIR/cjpeg \
 		-arch arm64 $TMPDIR/dist.armv8/$BINDIR/cjpeg \
 		-output $PKGROOT/$BINDIR/cjpeg
 	$LIPO -create \
 		$PKGROOT/$BINDIR/djpeg \
 		-arch arm64 $TMPDIR/dist.armv8/$BINDIR/djpeg \
 		-output $PKGROOT/$BINDIR/djpeg
 	$LIPO -create \
 		$PKGROOT/$BINDIR/jpegtran \
 		-arch arm64 $TMPDIR/dist.armv8/$BINDIR/jpegtran \
 		-output $PKGROOT/$BINDIR/jpegtran
 	$LIPO -create \
 		$PKGROOT/$BINDIR/tjbench \
 		-arch arm64 $TMPDIR/dist.armv8/$BINDIR/tjbench \
 		-output $PKGROOT/$BINDIR/tjbench
 	$LIPO -create \
 		$PKGROOT/$BINDIR/rdjpgcom \
 		-arch arm64 $TMPDIR/dist.armv8/$BINDIR/rdjpgcom \
 		-output $PKGROOT/$BINDIR/rdjpgcom
 	$LIPO -create \
 		$PKGROOT/$BINDIR/wrjpgcom \
 		-arch arm64 $TMPDIR/dist.armv8/$BINDIR/wrjpgcom \
 		-output $PKGROOT/$BINDIR/wrjpgcom
 fi
 install_name_tool -id $LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib
@@ -235,7 +440,7 @@ install_name_tool -id $LIBDIR/libturbojpeg.0.dylib $PKGROOT/$LIBDIR/libturbojpeg
 if [ $WITH_JAVA = 1 ]; then
 	ln -fs libturbojpeg.0.dylib $PKGROOT/$LIBDIR/libturbojpeg.jnilib
 fi
-if [ "$PREFIX" = "/opt/libmozjpeg" -a "$LIBDIR" = "/opt/libmozjpeg/lib" ]; then
+if [ "$PREFIX" = "/opt/mozjpeg" -a "$LIBDIR" = "/opt/mozjpeg/lib" ]; then
 	if [ ! -h $PKGROOT/$PREFIX/lib32 ]; then
 		ln -fs lib $PKGROOT/$PREFIX/lib32
 	fi
@@ -254,7 +459,7 @@ cp $SRCDIR/release/License.rtf $SRCDIR/release/Welcome.rtf $SRCDIR/release/ReadM
 mkdir $TMPDIR/dmg
 pkgbuild --root $PKGROOT --version $VERSION.$BUILD \
-	--identifier com.libjpeg-turbo.libjpeg-turbo $TMPDIR/pkg/$PACKAGE_NAME.pkg
+	--identifier com.mozilla.$PACKAGE_NAME $TMPDIR/pkg/$PACKAGE_NAME.pkg
 productbuild --distribution $SRCDIR/release/Distribution.xml \
 	--package-path $TMPDIR/pkg/ --resources $TMPDIR/pkg/ \
 	$TMPDIR/dmg/$PACKAGE_NAME.pkg
--- a/release/libmozjpeg.nsi.in
+++ b/release/libmozjpeg.nsi.in
@@ -111,6 +111,7 @@ Section "Uninstall"
 !ifdef GCC
 	Delete $INSTDIR\bin\libjpeg-@DLL_VERSION@.dll
 	Delete $INSTDIR\bin\libturbojpeg.dll
 	Delete $SYSDIR\libturbojpeg.dll
 	Delete $INSTDIR\lib\libturbojpeg.dll.a"
 	Delete $INSTDIR\lib\libturbojpeg.a"
@@ -118,6 +119,7 @@ Section "Uninstall"
 	Delete $INSTDIR\lib\libjpeg.a"
 !else
 	Delete $INSTDIR\bin\jpeg@DLL_VERSION@.dll
 	Delete $INSTDIR\bin\turbojpeg.dll
 	Delete $SYSDIR\turbojpeg.dll
 	Delete $INSTDIR\lib\jpeg.lib
 	Delete $INSTDIR\lib\jpeg-static.lib
--- a/release/libmozjpeg.spec.in
+++ b/release/libmozjpeg.spec.in
@@ -1,4 +1,4 @@
-# Path under which libmozjpeg should be installed
+# Path under which mozjpeg should be installed
 %define _prefix %{__prefix}
 # Path under which executables should be installed
@@ -17,7 +17,7 @@
 %ifarch x86_64
 %define _lib lib64
 %else
-%if "%{_prefix}" == "/opt/libmozjpeg"
+%if "%{_prefix}" == "/opt/mozjpeg"
 %define _lib lib32
 %endif
 %endif
@@ -25,13 +25,13 @@
 # Path under which man pages should be installed
 %define _mandir %{__mandir}
-Summary: A SIMD-accelerated JPEG codec that provides both the libjpeg and TurboJPEG APIs
+Summary: A JPEG codec that provides increased compression for JPEG images (at the expense of compression performance)
 Name: @PKGNAME@
 Version: @VERSION@
-Vendor: The libmozjpeg Project
+Vendor: Mozilla Research
 URL: https://github.com/mozilla/mozjpeg
 Group: System Environment/Libraries
-#-->Source0: http://prdownloads.sourceforge.net/libmozjpeg/libmozjpeg-%{version}.tar.gz
+#-->Source0: https://github.com/mozilla/mozjpeg/archive/v%{version}.tar.gz
 Release: @BUILD@
 License: BSD-style
 BuildRoot: %{_blddir}/%{name}-buildroot-%{version}-%{release}
@@ -43,30 +43,23 @@ Provides: %{name} = %{version}-%{release}, @PACKAGE_NAME@ = %{version}-%{release
 %endif
 %description
-libmozjpeg is a JPEG image codec that uses SIMD instructions (MMX, SSE2,
+mozjpeg is a fork of libjpeg-turbo that aims to speed up load times of web
-NEON) to accelerate baseline JPEG compression and decompression on x86, x86-64,
+pages by reducing the size (and, by extension, the transmission time) of JPEG
-and ARM systems.  On such systems, libmozjpeg is generally 2-4x as fast as
+files.  It accomplishes this by enabling optimized Huffman trees and
-libjpeg, all else being equal.  On other types of systems, libmozjpeg can
+progressive entropy coding by default in the JPEG compressor, as well as
-still outperform libjpeg by a significant amount, by virtue of its
+splitting the spectrum of DCT coefficients into separate scans and using
-highly-optimized Huffman coding routines.  In many cases, the performance of
+Trellis quantisation.
 libmozjpeg rivals that of proprietary high-speed JPEG codecs.
-libmozjpeg implements both the traditional libjpeg API as well as the less
+Although it is based on libjpeg-turbo, mozjpeg is not intended to be a
-powerful but more straightforward TurboJPEG API.  libmozjpeg also features
+general-purpose or high-performance JPEG library.  Its performance is highly
-colorspace extensions that allow it to compress from/decompress to 32-bit and
+"asymmetric".  That is, the JPEG files it generates require much more time to
-big-endian pixel buffers (RGBX, XBGR, etc.), as well as a full-featured Java
+compress than to decompress.  When the default settings are used, mozjpeg is
-interface.
+considerably slower than libjpeg-turbo or even libjpeg at compressing images.
-
+Thus, it is not generally suitable for real-time compression.  It is best used
-libmozjpeg was forked from libjpeg-turbo.
+as part of a web encoding workflow.
 libjpeg-turbo was originally based on libjpeg/SIMD, an MMX-accelerated
 derivative of libjpeg v6b developed by Miyasaka Masaru.  The TigerVNC and
 VirtualGL projects made numerous enhancements to the codec in 2009, and in
 early 2010, libjpeg-turbo spun off into an independent project, with the goal
 of making high-speed JPEG compression/decompression technology available to a
 broader range of users and developers.
 #-->%prep
-#-->%setup -q -n libmozjpeg-%{version}
+#-->%setup -q -n mozjpeg-%{version}
 #-->%build
 #-->./configure prefix=%{_prefix} bindir=%{_bindir} datadir=%{_datadir} \
@@ -87,14 +80,14 @@ rm -f $RPM_BUILD_ROOT%{_libdir}/*.la
 LJT_LIBDIR=%{__libdir}
 if [ ! "$LJT_LIBDIR" = "%{_libdir}" ]; then
-	echo ERROR: libmozjpeg must be configured with libdir=%{_prefix}/%{_lib} when generating an in-tree RPM for this architecture.
+	echo ERROR: mozjpeg must be configured with libdir=%{_prefix}/%{_lib} when generating an in-tree RPM for this architecture.
 	exit 1
 fi
 #-->%endif
 LJT_DOCDIR=%{__docdir}
-if [ "%{_prefix}" = "/opt/libmozjpeg" -a "$LJT_DOCDIR" = "/opt/libmozjpeg/doc" ]; then
+if [ "%{_prefix}" = "/opt/mozjpeg" -a "$LJT_DOCDIR" = "/opt/mozjpeg/doc" ]; then
 	ln -fs %{_docdir} $RPM_BUILD_ROOT/$LJT_DOCDIR
 fi
@@ -110,7 +103,7 @@ rm -rf $RPM_BUILD_ROOT
 %dir %{_docdir}
 %doc %{_docdir}/*
 %dir %{_prefix}
-%if "%{_prefix}" == "/opt/libmozjpeg" && "%{_docdir}" != "%{_prefix}/doc"
+%if "%{_prefix}" == "/opt/mozjpeg" && "%{_docdir}" != "%{_prefix}/doc"
 %{_prefix}/doc
 %endif
 %dir %{_bindir}
--- a/release/uninstall.in
+++ b/release/uninstall.in
@@ -8,7 +8,7 @@
 # - Redistributions in binary form must reproduce the above copyright notice,
 #   this list of conditions and the following disclaimer in the documentation
 #   and/or other materials provided with the distribution.
-# - Neither the name of the libmozjpeg Project nor the names of its
+# - Neither the name of the mozjpeg Project nor the names of its
 #   contributors may be used to endorse or promote products derived from this
 #   software without specific prior written permission.
 #
@@ -32,7 +32,7 @@ if [ ! "`id -u`" = "0" ]; then
 fi
 PACKAGE=@PKGNAME@
-MACPACKAGE=com.$PACKAGE.$PACKAGE
+MACPACKAGE=com.mozilla.$PACKAGE
 RECEIPT=/Library/Receipts/$PACKAGE.pkg
 LSBOM=
@@ -72,7 +72,7 @@ fi
 if [ -d $INCLUDEDIR ]; then
 	rmdir $INCLUDEDIR 2>&1 || EXITSTATUS=-1
 fi
-if [ "$PREFIX" = "/opt/libmozjpeg" -a "$LIBDIR" = "/opt/libmozjpeg/lib" ]; then
+if [ "$PREFIX" = "/opt/mozjpeg" -a "$LIBDIR" = "/opt/mozjpeg/lib" ]; then
 	if [ -h $LIBDIR\32 ]; then
 		rm $LIBDIR\32 2>&1 || EXITSTATUS=-1
 	fi
@@ -92,7 +92,7 @@ fi
 if [ -d $DATADIR -a "$DATADIR" != "$PREFIX" ]; then
 	rmdir $DATADIR 2>&1 || EXITSTATUS=-1
 fi
-if [ "$PREFIX" = "/opt/libmozjpeg" -a -h "$PREFIX/doc" ]; then
+if [ "$PREFIX" = "/opt/mozjpeg" -a -h "$PREFIX/doc" ]; then
 	rm $PREFIX/doc 2>&1 || EXITSTATUS=-1
 fi
 rmdir $PREFIX 2>&1 || EXITSTATUS=-1
--- a/simd/Makefile.am
+++ b/simd/Makefile.am
@@ -72,9 +72,16 @@ endif
 if SIMD_POWERPC
-libsimd_la_SOURCES = jsimd_powerpc.c jsimd_powerpc_altivec.c
+libsimd_la_SOURCES = jsimd_powerpc.c \
 	jccolor-altivec.c     jcgray-altivec.c \
 	jfdctfst-altivec.c    jfdctint-altivec.c \
 	jidctfst-altivec.c    jidctint-altivec.c \
  jquanti-altivec.c
 libsimd_la_CFLAGS = -maltivec
 jccolor-altivec.lo:  jccolext-altivec.c
 jcgray-altivec.lo:   jcgryext-altivec.c
 endif
 AM_CPPFLAGS = -I$(top_srcdir)
--- a/simd/jccolext-altivec.c
+++ b/simd/jccolext-altivec.c
@@ -0,0 +1,250 @@
 /*
 * AltiVec optimizations for libjpeg-turbo
 *
 * Copyright (C) 2014, D. R. Commander.
 * Copyright (C) 2014, Jay Foad.
 * All rights reserved.
 * This software is provided 'as-is', without any express or implied
 * warranty.  In no event will the authors be held liable for any damages
 * arising from the use of this software.
 *
 * Permission is granted to anyone to use this software for any purpose,
 * including commercial applications, and to alter it and redistribute it
 * freely, subject to the following restrictions:
 *
 * 1. The origin of this software must not be misrepresented; you must not
 *    claim that you wrote the original software. If you use this software
 *    in a product, an acknowledgment in the product documentation would be
 *    appreciated but is not required.
 * 2. Altered source versions must be plainly marked as such, and must not be
 *    misrepresented as being the original software.
 * 3. This notice may not be removed or altered from any source distribution.
 */
 /* This file is included by jccolor-altivec.c */
 void jsimd_rgb_ycc_convert_altivec (JDIMENSION img_width, JSAMPARRAY input_buf,
                                    JSAMPIMAGE output_buf,
                                    JDIMENSION output_row, int num_rows)
 {
  JSAMPROW inptr;
  JSAMPROW outptr0, outptr1, outptr2;
  int pitch;
  __vector unsigned char rgb0, rgb1 = {0}, rgb2 = {0}, rgb3 = {0}, rgbg0,
    rgbg1, rgbg2, rgbg3, y, cb, cr;
 #if RGB_PIXELSIZE == 4
  __vector unsigned char rgb4;
 #endif
  __vector short rg0, rg1, rg2, rg3, bg0, bg1, bg2, bg3;
  __vector unsigned short y01, y23, cr01, cr23, cb01, cb23;
  __vector int y0, y1, y2, y3, cr0, cr1, cr2, cr3, cb0, cb1, cb2, cb3;
  /* Constants */
  __vector short pw_f0299_f0337 = { __4X2(F_0_299, F_0_337) },
    pw_f0114_f0250 = { __4X2(F_0_114, F_0_250) },
    pw_mf016_mf033 = { __4X2(-F_0_168, -F_0_331) },
    pw_mf008_mf041 = { __4X2(-F_0_081, -F_0_418) };
  __vector unsigned short pw_f050_f000 = { __4X2(F_0_500, 0) };
  __vector int pd_onehalf = { __4X(ONE_HALF) },
    pd_onehalfm1_cj = { __4X(ONE_HALF - 1 + (CENTERJSAMPLE << SCALEBITS)) };
  __vector unsigned char zero = { __16X(0) },
    shift_pack_index =
      { 0, 1, 4, 5, 8, 9, 12, 13, 16, 17, 20, 21, 24, 25, 28, 29};
  while (--num_rows >= 0) {
    inptr = *input_buf++;
    outptr0 = output_buf[0][output_row];
    outptr1 = output_buf[1][output_row];
    outptr2 = output_buf[2][output_row];
    output_row++;
    for (pitch = img_width * RGB_PIXELSIZE; pitch > 0;
         pitch -= RGB_PIXELSIZE * 16, inptr += RGB_PIXELSIZE * 16,
         outptr0 += 16, outptr1 += 16, outptr2 += 16) {
 #if RGB_PIXELSIZE == 3
      /* Load 16 pixels == 48 bytes */
      if ((size_t)inptr & 15) {
        __vector unsigned char unaligned_shift_index;
        rgb0 = vec_ld(0, inptr);
        if (pitch > 16)
          rgb1 = vec_ld(16, inptr);
        else
          rgb1 = vec_ld(-1, inptr + pitch);
        if (pitch > 32)
          rgb2 = vec_ld(32, inptr);
        else
          rgb2 = vec_ld(-1, inptr + pitch);
        if (pitch > 48)
          rgb3 = vec_ld(48, inptr);
        else
          rgb3 = vec_ld(-1, inptr + pitch);
        unaligned_shift_index = vec_lvsl(0, inptr);
        rgb0 = vec_perm(rgb0, rgb1, unaligned_shift_index);
        rgb1 = vec_perm(rgb1, rgb2, unaligned_shift_index);
        rgb2 = vec_perm(rgb2, rgb3, unaligned_shift_index);
      } else {
        rgb0 = vec_ld(0, inptr);
        if (pitch > 16)
          rgb1 = vec_ld(16, inptr);
        if (pitch > 32)
          rgb2 = vec_ld(32, inptr);
      }
      /* rgb0 = R0 G0 B0 R1 G1 B1 R2 G2 B2 R3 G3 B3 R4 G4 B4 R5
       * rgb1 = G5 B5 R6 G6 B6 R7 G7 B7 R8 G8 B8 R9 G9 B9 Ra Ga
       * rgb2 = Ba Rb Gb Bb Rc Gc Bc Rd Gd Bd Re Ge Be Rf Gf Bf
       *
       * rgbg0 = R0 G0 R1 G1 R2 G2 R3 G3 B0 G0 B1 G1 B2 G2 B3 G3
       * rgbg1 = R4 G4 R5 G5 R6 G6 R7 G7 B4 G4 B5 G5 B6 G6 B7 G7
       * rgbg2 = R8 G8 R9 G9 Ra Ga Rb Gb B8 G8 B9 G9 Ba Ga Bb Gb
       * rgbg3 = Rc Gc Rd Gd Re Ge Rf Gf Bc Gc Bd Gd Be Ge Bf Gf
       */
      rgbg0 = vec_perm(rgb0, rgb0, (__vector unsigned char)RGBG_INDEX0);
      rgbg1 = vec_perm(rgb0, rgb1, (__vector unsigned char)RGBG_INDEX1);
      rgbg2 = vec_perm(rgb1, rgb2, (__vector unsigned char)RGBG_INDEX2);
      rgbg3 = vec_perm(rgb2, rgb2, (__vector unsigned char)RGBG_INDEX3);
 #else
      /* Load 16 pixels == 64 bytes */
      if ((size_t)inptr & 15) {
        __vector unsigned char unaligned_shift_index;
        rgb0 = vec_ld(0, inptr);
        if (pitch > 16)
          rgb1 = vec_ld(16, inptr);
        else
          rgb1 = vec_ld(-1, inptr + pitch);
        if (pitch > 32)
          rgb2 = vec_ld(32, inptr);
        else
          rgb2 = vec_ld(-1, inptr + pitch);
        if (pitch > 48)
          rgb3 = vec_ld(48, inptr);
        else
          rgb3 = vec_ld(-1, inptr + pitch);
        if (pitch > 64)
          rgb4 = vec_ld(64, inptr);
        else
          rgb4 = vec_ld(-1, inptr + pitch);
        unaligned_shift_index = vec_lvsl(0, inptr);
        rgb0 = vec_perm(rgb0, rgb1, unaligned_shift_index);
        rgb1 = vec_perm(rgb1, rgb2, unaligned_shift_index);
        rgb2 = vec_perm(rgb2, rgb3, unaligned_shift_index);
        rgb3 = vec_perm(rgb3, rgb4, unaligned_shift_index);
      } else {
        rgb0 = vec_ld(0, inptr);
        if (pitch > 16)
          rgb1 = vec_ld(16, inptr);
        if (pitch > 32)
          rgb2 = vec_ld(32, inptr);
        if (pitch > 48)
          rgb3 = vec_ld(48, inptr);
      }
      /* rgb0 = R0 G0 B0 X0 R1 G1 B1 X1 R2 G2 B2 X2 R3 G3 B3 X3
       * rgb0 = R4 G4 B4 X4 R5 G5 B5 X5 R6 G6 B6 X6 R7 G7 B7 X7
       * rgb0 = R8 G8 B8 X8 R9 G9 B9 X9 Ra Ga Ba Xa Rb Gb Bb Xb
       * rgb0 = Rc Gc Bc Xc Rd Gd Bd Xd Re Ge Be Xe Rf Gf Bf Xf
       *
       * rgbg0 = R0 G0 R1 G1 R2 G2 R3 G3 B0 G0 B1 G1 B2 G2 B3 G3
       * rgbg1 = R4 G4 R5 G5 R6 G6 R7 G7 B4 G4 B5 G5 B6 G6 B7 G7
       * rgbg2 = R8 G8 R9 G9 Ra Ga Rb Gb B8 G8 B9 G9 Ba Ga Bb Gb
       * rgbg3 = Rc Gc Rd Gd Re Ge Rf Gf Bc Gc Bd Gd Be Ge Bf Gf
       */
      rgbg0 = vec_perm(rgb0, rgb0, (__vector unsigned char)RGBG_INDEX);
      rgbg1 = vec_perm(rgb1, rgb1, (__vector unsigned char)RGBG_INDEX);
      rgbg2 = vec_perm(rgb2, rgb2, (__vector unsigned char)RGBG_INDEX);
      rgbg3 = vec_perm(rgb3, rgb3, (__vector unsigned char)RGBG_INDEX);
 #endif
      /* rg0 = R0 G0 R1 G1 R2 G2 R3 G3
       * bg0 = B0 G0 B1 G1 B2 G2 B3 G3
       * ...
       *
       * NOTE: We have to use vec_merge*() here because vec_unpack*() doesn't
       * support unsigned vectors.
       */
      rg0 = (__vector signed short)vec_mergeh(zero, rgbg0);
      bg0 = (__vector signed short)vec_mergel(zero, rgbg0);
      rg1 = (__vector signed short)vec_mergeh(zero, rgbg1);
      bg1 = (__vector signed short)vec_mergel(zero, rgbg1);
      rg2 = (__vector signed short)vec_mergeh(zero, rgbg2);
      bg2 = (__vector signed short)vec_mergel(zero, rgbg2);
      rg3 = (__vector signed short)vec_mergeh(zero, rgbg3);
      bg3 = (__vector signed short)vec_mergel(zero, rgbg3);
      /* (Original)
       * Y  =  0.29900 * R + 0.58700 * G + 0.11400 * B
       * Cb = -0.16874 * R - 0.33126 * G + 0.50000 * B + CENTERJSAMPLE
       * Cr =  0.50000 * R - 0.41869 * G - 0.08131 * B + CENTERJSAMPLE
       *
       * (This implementation)
       * Y  =  0.29900 * R + 0.33700 * G + 0.11400 * B + 0.25000 * G
       * Cb = -0.16874 * R - 0.33126 * G + 0.50000 * B + CENTERJSAMPLE
       * Cr =  0.50000 * R - 0.41869 * G - 0.08131 * B + CENTERJSAMPLE
       */
      /* Calculate Y values */
      y0 = vec_msums(rg0, pw_f0299_f0337, pd_onehalf);
      y1 = vec_msums(rg1, pw_f0299_f0337, pd_onehalf);
      y2 = vec_msums(rg2, pw_f0299_f0337, pd_onehalf);
      y3 = vec_msums(rg3, pw_f0299_f0337, pd_onehalf);
      y0 = vec_msums(bg0, pw_f0114_f0250, y0);
      y1 = vec_msums(bg1, pw_f0114_f0250, y1);
      y2 = vec_msums(bg2, pw_f0114_f0250, y2);
      y3 = vec_msums(bg3, pw_f0114_f0250, y3);
      /* Clever way to avoid 4 shifts + 2 packs.  This packs the high word from
       * each dword into a new 16-bit vector, which is the equivalent of
       * descaling the 32-bit results (right-shifting by 16 bits) and then
       * packing them.
       */
      y01 = vec_perm((__vector unsigned short)y0, (__vector unsigned short)y1,
                     shift_pack_index);
      y23 = vec_perm((__vector unsigned short)y2, (__vector unsigned short)y3,
                     shift_pack_index);
      y = vec_pack(y01, y23);
      vec_st(y, 0, outptr0);
      /* Calculate Cb values */
      cb0 = vec_msums(rg0, pw_mf016_mf033, pd_onehalfm1_cj);
      cb1 = vec_msums(rg1, pw_mf016_mf033, pd_onehalfm1_cj);
      cb2 = vec_msums(rg2, pw_mf016_mf033, pd_onehalfm1_cj);
      cb3 = vec_msums(rg3, pw_mf016_mf033, pd_onehalfm1_cj);
      cb0 = (__vector int)vec_msum((__vector unsigned short)bg0, pw_f050_f000,
                                   (__vector unsigned int)cb0);
      cb1 = (__vector int)vec_msum((__vector unsigned short)bg1, pw_f050_f000,
                                   (__vector unsigned int)cb1);
      cb2 = (__vector int)vec_msum((__vector unsigned short)bg2, pw_f050_f000,
                                   (__vector unsigned int)cb2);
      cb3 = (__vector int)vec_msum((__vector unsigned short)bg3, pw_f050_f000,
                                   (__vector unsigned int)cb3);
      cb01 = vec_perm((__vector unsigned short)cb0,
                      (__vector unsigned short)cb1, shift_pack_index);
      cb23 = vec_perm((__vector unsigned short)cb2,
                      (__vector unsigned short)cb3, shift_pack_index);
      cb = vec_pack(cb01, cb23);
      vec_st(cb, 0, outptr1);
      /* Calculate Cr values */
      cr0 = vec_msums(bg0, pw_mf008_mf041, pd_onehalfm1_cj);
      cr1 = vec_msums(bg1, pw_mf008_mf041, pd_onehalfm1_cj);
      cr2 = vec_msums(bg2, pw_mf008_mf041, pd_onehalfm1_cj);
      cr3 = vec_msums(bg3, pw_mf008_mf041, pd_onehalfm1_cj);
      cr0 = (__vector int)vec_msum((__vector unsigned short)rg0, pw_f050_f000,
                                   (__vector unsigned int)cr0);
      cr1 = (__vector int)vec_msum((__vector unsigned short)rg1, pw_f050_f000,
                                   (__vector unsigned int)cr1);
      cr2 = (__vector int)vec_msum((__vector unsigned short)rg2, pw_f050_f000,
                                   (__vector unsigned int)cr2);
      cr3 = (__vector int)vec_msum((__vector unsigned short)rg3, pw_f050_f000,
                                   (__vector unsigned int)cr3);
      cr01 = vec_perm((__vector unsigned short)cr0,
                      (__vector unsigned short)cr1, shift_pack_index);
      cr23 = vec_perm((__vector unsigned short)cr2,
                      (__vector unsigned short)cr3, shift_pack_index);
      cr = vec_pack(cr01, cr23);
      vec_st(cr, 0, outptr2);
    }
  }
 }
--- a/simd/jccolor-altivec.c
+++ b/simd/jccolor-altivec.c
@@ -0,0 +1,104 @@
 /*
 * AltiVec optimizations for libjpeg-turbo
 *
 * Copyright (C) 2014, D. R. Commander.
 * All rights reserved.
 * This software is provided 'as-is', without any express or implied
 * warranty.  In no event will the authors be held liable for any damages
 * arising from the use of this software.
 *
 * Permission is granted to anyone to use this software for any purpose,
 * including commercial applications, and to alter it and redistribute it
 * freely, subject to the following restrictions:
 *
 * 1. The origin of this software must not be misrepresented; you must not
 *    claim that you wrote the original software. If you use this software
 *    in a product, an acknowledgment in the product documentation would be
 *    appreciated but is not required.
 * 2. Altered source versions must be plainly marked as such, and must not be
 *    misrepresented as being the original software.
 * 3. This notice may not be removed or altered from any source distribution.
 */
 /* RGB --> YCC CONVERSION */
 #include "jsimd_altivec.h"
 #define F_0_081 5329                 /* FIX(0.08131) */
 #define F_0_114 7471                 /* FIX(0.11400) */
 #define F_0_168 11059                /* FIX(0.16874) */
 #define F_0_250 16384                /* FIX(0.25000) */
 #define F_0_299 19595                /* FIX(0.29900) */
 #define F_0_331 21709                /* FIX(0.33126) */
 #define F_0_418 27439                /* FIX(0.41869) */
 #define F_0_500 32768                /* FIX(0.50000) */
 #define F_0_587 38470                /* FIX(0.58700) */
 #define F_0_337 (F_0_587 - F_0_250)  /* FIX(0.58700) - FIX(0.25000) */
 #define SCALEBITS 16
 #define ONE_HALF (1 << (SCALEBITS - 1))
 #define RGBG_INDEX0 {0,1,3,4,6,7,9,10,2,1,5,4,8,7,11,10}
 #define RGBG_INDEX1 {12,13,15,16,18,19,21,22,14,13,17,16,20,19,23,22}
 #define RGBG_INDEX2 {8,9,11,12,14,15,17,18,10,9,13,12,16,15,19,18}
 #define RGBG_INDEX3 {4,5,7,8,10,11,13,14,6,5,9,8,12,11,15,14}
 #include "jccolext-altivec.c"
 #undef RGB_PIXELSIZE
 #define RGB_PIXELSIZE EXT_RGB_PIXELSIZE
 #define jsimd_rgb_ycc_convert_altivec jsimd_extrgb_ycc_convert_altivec
 #include "jccolext-altivec.c"
 #undef RGB_PIXELSIZE
 #undef RGBG_INDEX0
 #undef RGBG_INDEX1
 #undef RGBG_INDEX2
 #undef RGBG_INDEX3
 #undef jsimd_rgb_ycc_convert_altivec
 #define RGB_PIXELSIZE EXT_RGBX_PIXELSIZE
 #define RGBG_INDEX {0,1,4,5,8,9,12,13,2,1,6,5,10,9,14,13}
 #define jsimd_rgb_ycc_convert_altivec jsimd_extrgbx_ycc_convert_altivec
 #include "jccolext-altivec.c"
 #undef RGB_PIXELSIZE
 #undef RGBG_INDEX
 #undef jsimd_rgb_ycc_convert_altivec
 #define RGB_PIXELSIZE EXT_BGR_PIXELSIZE
 #define RGBG_INDEX0 {2,1,5,4,8,7,11,10,0,1,3,4,6,7,9,10}
 #define RGBG_INDEX1 {14,13,17,16,20,19,23,22,12,13,15,16,18,19,21,22}
 #define RGBG_INDEX2 {10,9,13,12,16,15,19,18,8,9,11,12,14,15,17,18}
 #define RGBG_INDEX3 {6,5,9,8,12,11,15,14,4,5,7,8,10,11,13,14}
 #define jsimd_rgb_ycc_convert_altivec jsimd_extbgr_ycc_convert_altivec
 #include "jccolext-altivec.c"
 #undef RGB_PIXELSIZE
 #undef RGBG_INDEX0
 #undef RGBG_INDEX1
 #undef RGBG_INDEX2
 #undef RGBG_INDEX3
 #undef jsimd_rgb_ycc_convert_altivec
 #define RGB_PIXELSIZE EXT_BGRX_PIXELSIZE
 #define RGBG_INDEX {2,1,6,5,10,9,14,13,0,1,4,5,8,9,12,13}
 #define jsimd_rgb_ycc_convert_altivec jsimd_extbgrx_ycc_convert_altivec
 #include "jccolext-altivec.c"
 #undef RGB_PIXELSIZE
 #undef RGBG_INDEX
 #undef jsimd_rgb_ycc_convert_altivec
 #define RGB_PIXELSIZE EXT_XBGR_PIXELSIZE
 #define RGBG_INDEX {3,2,7,6,11,10,15,14,1,2,5,6,9,10,13,14}
 #define jsimd_rgb_ycc_convert_altivec jsimd_extxbgr_ycc_convert_altivec
 #include "jccolext-altivec.c"
 #undef RGB_PIXELSIZE
 #undef RGBG_INDEX
 #undef jsimd_rgb_ycc_convert_altivec
 #define RGB_PIXELSIZE EXT_XRGB_PIXELSIZE
 #define RGBG_INDEX {1,2,5,6,9,10,13,14,3,2,7,6,11,10,15,14}
 #define jsimd_rgb_ycc_convert_altivec jsimd_extxrgb_ycc_convert_altivec
 #include "jccolext-altivec.c"
 #undef RGB_PIXELSIZE
 #undef RGBG_INDEX
 #undef jsimd_rgb_ycc_convert_altivec
--- a/simd/jcgray-altivec.c
+++ b/simd/jcgray-altivec.c
@@ -0,0 +1,99 @@
 /*
 * AltiVec optimizations for libjpeg-turbo
 *
 * Copyright (C) 2014, D. R. Commander.
 * All rights reserved.
 * This software is provided 'as-is', without any express or implied
 * warranty.  In no event will the authors be held liable for any damages
 * arising from the use of this software.
 *
 * Permission is granted to anyone to use this software for any purpose,
 * including commercial applications, and to alter it and redistribute it
 * freely, subject to the following restrictions:
 *
 * 1. The origin of this software must not be misrepresented; you must not
 *    claim that you wrote the original software. If you use this software
 *    in a product, an acknowledgment in the product documentation would be
 *    appreciated but is not required.
 * 2. Altered source versions must be plainly marked as such, and must not be
 *    misrepresented as being the original software.
 * 3. This notice may not be removed or altered from any source distribution.
 */
 /* RGB --> GRAYSCALE CONVERSION */
 #include "jsimd_altivec.h"
 #define F_0_114 7471                 /* FIX(0.11400) */
 #define F_0_250 16384                /* FIX(0.25000) */
 #define F_0_299 19595                /* FIX(0.29900) */
 #define F_0_587 38470                /* FIX(0.58700) */
 #define F_0_337 (F_0_587 - F_0_250)  /* FIX(0.58700) - FIX(0.25000) */
 #define SCALEBITS 16
 #define ONE_HALF (1 << (SCALEBITS - 1))
 #define RGBG_INDEX0 {0,1,3,4,6,7,9,10,2,1,5,4,8,7,11,10}
 #define RGBG_INDEX1 {12,13,15,16,18,19,21,22,14,13,17,16,20,19,23,22}
 #define RGBG_INDEX2 {8,9,11,12,14,15,17,18,10,9,13,12,16,15,19,18}
 #define RGBG_INDEX3 {4,5,7,8,10,11,13,14,6,5,9,8,12,11,15,14}
 #include "jcgryext-altivec.c"
 #undef RGB_PIXELSIZE
 #define RGB_PIXELSIZE EXT_RGB_PIXELSIZE
 #define jsimd_rgb_gray_convert_altivec jsimd_extrgb_gray_convert_altivec
 #include "jcgryext-altivec.c"
 #undef RGB_PIXELSIZE
 #undef RGBG_INDEX0
 #undef RGBG_INDEX1
 #undef RGBG_INDEX2
 #undef RGBG_INDEX3
 #undef jsimd_rgb_gray_convert_altivec
 #define RGB_PIXELSIZE EXT_RGBX_PIXELSIZE
 #define RGBG_INDEX {0,1,4,5,8,9,12,13,2,1,6,5,10,9,14,13}
 #define jsimd_rgb_gray_convert_altivec jsimd_extrgbx_gray_convert_altivec
 #include "jcgryext-altivec.c"
 #undef RGB_PIXELSIZE
 #undef RGBG_INDEX
 #undef jsimd_rgb_gray_convert_altivec
 #define RGB_PIXELSIZE EXT_BGR_PIXELSIZE
 #define RGBG_INDEX0 {2,1,5,4,8,7,11,10,0,1,3,4,6,7,9,10}
 #define RGBG_INDEX1 {14,13,17,16,20,19,23,22,12,13,15,16,18,19,21,22}
 #define RGBG_INDEX2 {10,9,13,12,16,15,19,18,8,9,11,12,14,15,17,18}
 #define RGBG_INDEX3 {6,5,9,8,12,11,15,14,4,5,7,8,10,11,13,14}
 #define jsimd_rgb_gray_convert_altivec jsimd_extbgr_gray_convert_altivec
 #include "jcgryext-altivec.c"
 #undef RGB_PIXELSIZE
 #undef RGBG_INDEX0
 #undef RGBG_INDEX1
 #undef RGBG_INDEX2
 #undef RGBG_INDEX3
 #undef jsimd_rgb_gray_convert_altivec
 #define RGB_PIXELSIZE EXT_BGRX_PIXELSIZE
 #define RGBG_INDEX {2,1,6,5,10,9,14,13,0,1,4,5,8,9,12,13}
 #define jsimd_rgb_gray_convert_altivec jsimd_extbgrx_gray_convert_altivec
 #include "jcgryext-altivec.c"
 #undef RGB_PIXELSIZE
 #undef RGBG_INDEX
 #undef jsimd_rgb_gray_convert_altivec
 #define RGB_PIXELSIZE EXT_XBGR_PIXELSIZE
 #define RGBG_INDEX {3,2,7,6,11,10,15,14,1,2,5,6,9,10,13,14}
 #define jsimd_rgb_gray_convert_altivec jsimd_extxbgr_gray_convert_altivec
 #include "jcgryext-altivec.c"
 #undef RGB_PIXELSIZE
 #undef RGBG_INDEX
 #undef jsimd_rgb_gray_convert_altivec
 #define RGB_PIXELSIZE EXT_XRGB_PIXELSIZE
 #define RGBG_INDEX {1,2,5,6,9,10,13,14,3,2,7,6,11,10,15,14}
 #define jsimd_rgb_gray_convert_altivec jsimd_extxrgb_gray_convert_altivec
 #include "jcgryext-altivec.c"
 #undef RGB_PIXELSIZE
 #undef RGBG_INDEX
 #undef jsimd_rgb_gray_convert_altivec
--- a/simd/jcgryext-altivec.c
+++ b/simd/jcgryext-altivec.c
@@ -0,0 +1,200 @@
 /*
 * AltiVec optimizations for libjpeg-turbo
 *
 * Copyright (C) 2014, D. R. Commander.
 * Copyright (C) 2014, Jay Foad.
 * All rights reserved.
 * This software is provided 'as-is', without any express or implied
 * warranty.  In no event will the authors be held liable for any damages
 * arising from the use of this software.
 *
 * Permission is granted to anyone to use this software for any purpose,
 * including commercial applications, and to alter it and redistribute it
 * freely, subject to the following restrictions:
 *
 * 1. The origin of this software must not be misrepresented; you must not
 *    claim that you wrote the original software. If you use this software
 *    in a product, an acknowledgment in the product documentation would be
 *    appreciated but is not required.
 * 2. Altered source versions must be plainly marked as such, and must not be
 *    misrepresented as being the original software.
 * 3. This notice may not be removed or altered from any source distribution.
 */
 /* This file is included by jcgray-altivec.c */
 void jsimd_rgb_gray_convert_altivec (JDIMENSION img_width,
                                     JSAMPARRAY input_buf,
                                     JSAMPIMAGE output_buf,
                                     JDIMENSION output_row, int num_rows)
 {
  JSAMPROW inptr, outptr;
  int pitch;
  __vector unsigned char rgb0, rgb1 = {0}, rgb2 = {0}, rgb3 = {0}, rgbg0,
    rgbg1, rgbg2, rgbg3, y;
 #if RGB_PIXELSIZE == 4
  __vector unsigned char rgb4;
 #endif
  __vector short rg0, rg1, rg2, rg3, bg0, bg1, bg2, bg3;
  __vector unsigned short y01, y23;
  __vector int y0, y1, y2, y3;
  /* Constants */
  __vector short pw_f0299_f0337 = { __4X2(F_0_299, F_0_337) },
    pw_f0114_f0250 = { __4X2(F_0_114, F_0_250) };
  __vector int pd_onehalf = { __4X(ONE_HALF) };
  __vector unsigned char zero = { __16X(0) },
    shift_pack_index =
      { 0, 1, 4, 5, 8, 9, 12, 13, 16, 17, 20, 21, 24, 25, 28, 29};
  while (--num_rows >= 0) {
    inptr = *input_buf++;
    outptr = output_buf[0][output_row];
    output_row++;
    for (pitch = img_width * RGB_PIXELSIZE; pitch > 0;
         pitch -= RGB_PIXELSIZE * 16, inptr += RGB_PIXELSIZE * 16,
         outptr += 16) {
 #if RGB_PIXELSIZE == 3
      /* Load 16 pixels == 48 bytes */
      if ((size_t)inptr & 15) {
        __vector unsigned char unaligned_shift_index;
        rgb0 = vec_ld(0, inptr);
        if (pitch > 16)
          rgb1 = vec_ld(16, inptr);
        else
          rgb1 = vec_ld(-1, inptr + pitch);
        if (pitch > 32)
          rgb2 = vec_ld(32, inptr);
        else
          rgb2 = vec_ld(-1, inptr + pitch);
        if (pitch > 48)
          rgb3 = vec_ld(48, inptr);
        else
          rgb3 = vec_ld(-1, inptr + pitch);
        unaligned_shift_index = vec_lvsl(0, inptr);
        rgb0 = vec_perm(rgb0, rgb1, unaligned_shift_index);
        rgb1 = vec_perm(rgb1, rgb2, unaligned_shift_index);
        rgb2 = vec_perm(rgb2, rgb3, unaligned_shift_index);
      } else {
        rgb0 = vec_ld(0, inptr);
        if (pitch > 16)
          rgb1 = vec_ld(16, inptr);
        if (pitch > 32)
          rgb2 = vec_ld(32, inptr);
      }
      /* rgb0 = R0 G0 B0 R1 G1 B1 R2 G2 B2 R3 G3 B3 R4 G4 B4 R5
       * rgb1 = G5 B5 R6 G6 B6 R7 G7 B7 R8 G8 B8 R9 G9 B9 Ra Ga
       * rgb2 = Ba Rb Gb Bb Rc Gc Bc Rd Gd Bd Re Ge Be Rf Gf Bf
       *
       * rgbg0 = R0 G0 R1 G1 R2 G2 R3 G3 B0 G0 B1 G1 B2 G2 B3 G3
       * rgbg1 = R4 G4 R5 G5 R6 G6 R7 G7 B4 G4 B5 G5 B6 G6 B7 G7
       * rgbg2 = R8 G8 R9 G9 Ra Ga Rb Gb B8 G8 B9 G9 Ba Ga Bb Gb
       * rgbg3 = Rc Gc Rd Gd Re Ge Rf Gf Bc Gc Bd Gd Be Ge Bf Gf
       */
      rgbg0 = vec_perm(rgb0, rgb0, (__vector unsigned char)RGBG_INDEX0);
      rgbg1 = vec_perm(rgb0, rgb1, (__vector unsigned char)RGBG_INDEX1);
      rgbg2 = vec_perm(rgb1, rgb2, (__vector unsigned char)RGBG_INDEX2);
      rgbg3 = vec_perm(rgb2, rgb2, (__vector unsigned char)RGBG_INDEX3);
 #else
      /* Load 16 pixels == 64 bytes */
      if ((size_t)inptr & 15) {
        __vector unsigned char unaligned_shift_index;
        rgb0 = vec_ld(0, inptr);
        if (pitch > 16)
          rgb1 = vec_ld(16, inptr);
        else
          rgb1 = vec_ld(-1, inptr + pitch);
        if (pitch > 32)
          rgb2 = vec_ld(32, inptr);
        else
          rgb2 = vec_ld(-1, inptr + pitch);
        if (pitch > 48)
          rgb3 = vec_ld(48, inptr);
        else
          rgb3 = vec_ld(-1, inptr + pitch);
        if (pitch > 64)
          rgb4 = vec_ld(64, inptr);
        else
          rgb4 = vec_ld(-1, inptr + pitch);
        unaligned_shift_index = vec_lvsl(0, inptr);
        rgb0 = vec_perm(rgb0, rgb1, unaligned_shift_index);
        rgb1 = vec_perm(rgb1, rgb2, unaligned_shift_index);
        rgb2 = vec_perm(rgb2, rgb3, unaligned_shift_index);
        rgb3 = vec_perm(rgb3, rgb4, unaligned_shift_index);
      } else {
        rgb0 = vec_ld(0, inptr);
        if (pitch > 16)
          rgb1 = vec_ld(16, inptr);
        if (pitch > 32)
          rgb2 = vec_ld(32, inptr);
        if (pitch > 48)
          rgb3 = vec_ld(48, inptr);
      }
      /* rgb0 = R0 G0 B0 X0 R1 G1 B1 X1 R2 G2 B2 X2 R3 G3 B3 X3
       * rgb0 = R4 G4 B4 X4 R5 G5 B5 X5 R6 G6 B6 X6 R7 G7 B7 X7
       * rgb0 = R8 G8 B8 X8 R9 G9 B9 X9 Ra Ga Ba Xa Rb Gb Bb Xb
       * rgb0 = Rc Gc Bc Xc Rd Gd Bd Xd Re Ge Be Xe Rf Gf Bf Xf
       *
       * rgbg0 = R0 G0 R1 G1 R2 G2 R3 G3 B0 G0 B1 G1 B2 G2 B3 G3
       * rgbg1 = R4 G4 R5 G5 R6 G6 R7 G7 B4 G4 B5 G5 B6 G6 B7 G7
       * rgbg2 = R8 G8 R9 G9 Ra Ga Rb Gb B8 G8 B9 G9 Ba Ga Bb Gb
       * rgbg3 = Rc Gc Rd Gd Re Ge Rf Gf Bc Gc Bd Gd Be Ge Bf Gf
       */
      rgbg0 = vec_perm(rgb0, rgb0, (__vector unsigned char)RGBG_INDEX);
      rgbg1 = vec_perm(rgb1, rgb1, (__vector unsigned char)RGBG_INDEX);
      rgbg2 = vec_perm(rgb2, rgb2, (__vector unsigned char)RGBG_INDEX);
      rgbg3 = vec_perm(rgb3, rgb3, (__vector unsigned char)RGBG_INDEX);
 #endif
      /* rg0 = R0 G0 R1 G1 R2 G2 R3 G3
       * bg0 = B0 G0 B1 G1 B2 G2 B3 G3
       * ...
       *
       * NOTE: We have to use vec_merge*() here because vec_unpack*() doesn't
       * support unsigned vectors.
       */
      rg0 = (__vector signed short)vec_mergeh(zero, rgbg0);
      bg0 = (__vector signed short)vec_mergel(zero, rgbg0);
      rg1 = (__vector signed short)vec_mergeh(zero, rgbg1);
      bg1 = (__vector signed short)vec_mergel(zero, rgbg1);
      rg2 = (__vector signed short)vec_mergeh(zero, rgbg2);
      bg2 = (__vector signed short)vec_mergel(zero, rgbg2);
      rg3 = (__vector signed short)vec_mergeh(zero, rgbg3);
      bg3 = (__vector signed short)vec_mergel(zero, rgbg3);
      /* (Original)
       * Y  =  0.29900 * R + 0.58700 * G + 0.11400 * B
       *
       * (This implementation)
       * Y  =  0.29900 * R + 0.33700 * G + 0.11400 * B + 0.25000 * G
       */
      /* Calculate Y values */
      y0 = vec_msums(rg0, pw_f0299_f0337, pd_onehalf);
      y1 = vec_msums(rg1, pw_f0299_f0337, pd_onehalf);
      y2 = vec_msums(rg2, pw_f0299_f0337, pd_onehalf);
      y3 = vec_msums(rg3, pw_f0299_f0337, pd_onehalf);
      y0 = vec_msums(bg0, pw_f0114_f0250, y0);
      y1 = vec_msums(bg1, pw_f0114_f0250, y1);
      y2 = vec_msums(bg2, pw_f0114_f0250, y2);
      y3 = vec_msums(bg3, pw_f0114_f0250, y3);
      /* Clever way to avoid 4 shifts + 2 packs.  This packs the high word from
       * each dword into a new 16-bit vector, which is the equivalent of
       * descaling the 32-bit results (right-shifting by 16 bits) and then
       * packing them.
       */
      y01 = vec_perm((__vector unsigned short)y0, (__vector unsigned short)y1,
                     shift_pack_index);
      y23 = vec_perm((__vector unsigned short)y2, (__vector unsigned short)y3,
                     shift_pack_index);
      y = vec_pack(y01, y23);
      vec_st(y, 0, outptr);
    }
  }
 }
--- a/simd/jfdctfst-altivec.c
+++ b/simd/jfdctfst-altivec.c
@@ -0,0 +1,156 @@
 /*
 * AltiVec optimizations for libjpeg-turbo
 *
 * Copyright (C) 2014, D. R. Commander.
 * All rights reserved.
 * This software is provided 'as-is', without any express or implied
 * warranty.  In no event will the authors be held liable for any damages
 * arising from the use of this software.
 *
 * Permission is granted to anyone to use this software for any purpose,
 * including commercial applications, and to alter it and redistribute it
 * freely, subject to the following restrictions:
 *
 * 1. The origin of this software must not be misrepresented; you must not
 *    claim that you wrote the original software. If you use this software
 *    in a product, an acknowledgment in the product documentation would be
 *    appreciated but is not required.
 * 2. Altered source versions must be plainly marked as such, and must not be
 *    misrepresented as being the original software.
 * 3. This notice may not be removed or altered from any source distribution.
 */
 /* FAST INTEGER FORWARD DCT
 *
 * This is similar to the SSE2 implementation, except that we left-shift the
 * constants by 1 less bit (the -1 in CONST_SHIFT.)  This is because
 * vec_madds(arg1, arg2, arg3) generates the 16-bit saturated sum of:
 *   the elements in arg3 + the most significant 17 bits of
 *     (the elements in arg1 * the elements in arg2).
 */
 #include "jsimd_altivec.h"
 #define F_0_382 98   /* FIX(0.382683433) */
 #define F_0_541 139  /* FIX(0.541196100) */
 #define F_0_707 181  /* FIX(0.707106781) */
 #define F_1_306 334  /* FIX(1.306562965) */
 #define CONST_BITS 8
 #define PRE_MULTIPLY_SCALE_BITS 2
 #define CONST_SHIFT (16 - PRE_MULTIPLY_SCALE_BITS - CONST_BITS - 1)
 #define DO_FDCT()  \
 {  \
  /* Even part */  \
  \
  tmp10 = vec_add(tmp0, tmp3);  \
  tmp13 = vec_sub(tmp0, tmp3);  \
  tmp11 = vec_add(tmp1, tmp2);  \
  tmp12 = vec_sub(tmp1, tmp2);  \
  \
  out0  = vec_add(tmp10, tmp11);  \
  out4  = vec_sub(tmp10, tmp11);  \
  \
  z1 = vec_add(tmp12, tmp13);  \
  z1 = vec_sl(z1, pre_multiply_scale_bits);  \
  z1 = vec_madds(z1, pw_0707, zero);  \
  \
  out2 = vec_add(tmp13, z1);  \
  out6 = vec_sub(tmp13, z1);  \
  \
  /* Odd part */  \
  \
  tmp10 = vec_add(tmp4, tmp5);  \
  tmp11 = vec_add(tmp5, tmp6);  \
  tmp12 = vec_add(tmp6, tmp7);  \
  \
  tmp10 = vec_sl(tmp10, pre_multiply_scale_bits);  \
  tmp12 = vec_sl(tmp12, pre_multiply_scale_bits);  \
  z5 = vec_sub(tmp10, tmp12);  \
  z5 = vec_madds(z5, pw_0382, zero);  \
  \
  z2 = vec_madds(tmp10, pw_0541, z5);  \
  z4 = vec_madds(tmp12, pw_1306, z5);  \
  \
  tmp11 = vec_sl(tmp11, pre_multiply_scale_bits);  \
  z3 = vec_madds(tmp11, pw_0707, zero);  \
  \
  z11 = vec_add(tmp7, z3);  \
  z13 = vec_sub(tmp7, z3);  \
  \
  out5 = vec_add(z13, z2);  \
  out3 = vec_sub(z13, z2);  \
  out1 = vec_add(z11, z4);  \
  out7 = vec_sub(z11, z4);  \
 }
 void
 jsimd_fdct_ifast_altivec (DCTELEM *data)
 {
  __vector short row0, row1, row2, row3, row4, row5, row6, row7,
    col0, col1, col2, col3, col4, col5, col6, col7,
    tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, tmp10, tmp11, tmp12, tmp13,
    z1, z2, z3, z4, z5, z11, z13,
    out0, out1, out2, out3, out4, out5, out6, out7;
  /* Constants */
  __vector short zero = vec_splat_s16(0),
    pw_0382 = { __8X(F_0_382 << CONST_SHIFT) },
    pw_0541 = { __8X(F_0_541 << CONST_SHIFT) },
    pw_0707 = { __8X(F_0_707 << CONST_SHIFT) },
    pw_1306 = { __8X(F_1_306 << CONST_SHIFT) };
  __vector unsigned short
    pre_multiply_scale_bits = { __8X(PRE_MULTIPLY_SCALE_BITS) };
  /* Pass 1: process rows */
  row0 = vec_ld(0, data);
  row1 = vec_ld(16, data);
  row2 = vec_ld(32, data);
  row3 = vec_ld(48, data);
  row4 = vec_ld(64, data);
  row5 = vec_ld(80, data);
  row6 = vec_ld(96, data);
  row7 = vec_ld(112, data);
  TRANSPOSE(row, col);
  tmp0 = vec_add(col0, col7);
  tmp7 = vec_sub(col0, col7);
  tmp1 = vec_add(col1, col6);
  tmp6 = vec_sub(col1, col6);
  tmp2 = vec_add(col2, col5);
  tmp5 = vec_sub(col2, col5);
  tmp3 = vec_add(col3, col4);
  tmp4 = vec_sub(col3, col4);
  DO_FDCT();
  /* Pass 2: process columns */
  TRANSPOSE(out, row);
  tmp0 = vec_add(row0, row7);
  tmp7 = vec_sub(row0, row7);
  tmp1 = vec_add(row1, row6);
  tmp6 = vec_sub(row1, row6);
  tmp2 = vec_add(row2, row5);
  tmp5 = vec_sub(row2, row5);
  tmp3 = vec_add(row3, row4);
  tmp4 = vec_sub(row3, row4);
  DO_FDCT();
  vec_st(out0, 0, data);
  vec_st(out1, 16, data);
  vec_st(out2, 32, data);
  vec_st(out3, 48, data);
  vec_st(out4, 64, data);
  vec_st(out5, 80, data);
  vec_st(out6, 96, data);
  vec_st(out7, 112, data);
 }
--- a/simd/jfdctint-altivec.c
+++ b/simd/jfdctint-altivec.c
@@ -0,0 +1,262 @@
 /*
 * AltiVec optimizations for libjpeg-turbo
 *
 * Copyright (C) 2014, D. R. Commander.
 * All rights reserved.
 * This software is provided 'as-is', without any express or implied
 * warranty.  In no event will the authors be held liable for any damages
 * arising from the use of this software.
 *
 * Permission is granted to anyone to use this software for any purpose,
 * including commercial applications, and to alter it and redistribute it
 * freely, subject to the following restrictions:
 *
 * 1. The origin of this software must not be misrepresented; you must not
 *    claim that you wrote the original software. If you use this software
 *    in a product, an acknowledgment in the product documentation would be
 *    appreciated but is not required.
 * 2. Altered source versions must be plainly marked as such, and must not be
 *    misrepresented as being the original software.
 * 3. This notice may not be removed or altered from any source distribution.
 */
 /* SLOW INTEGER FORWARD DCT */
 #include "jsimd_altivec.h"
 #define F_0_298 2446   /* FIX(0.298631336) */
 #define F_0_390 3196   /* FIX(0.390180644) */
 #define F_0_541 4433   /* FIX(0.541196100) */
 #define F_0_765 6270   /* FIX(0.765366865) */
 #define F_0_899 7373   /* FIX(0.899976223) */
 #define F_1_175 9633   /* FIX(1.175875602) */
 #define F_1_501 12299  /* FIX(1.501321110) */
 #define F_1_847 15137  /* FIX(1.847759065) */
 #define F_1_961 16069  /* FIX(1.961570560) */
 #define F_2_053 16819  /* FIX(2.053119869) */
 #define F_2_562 20995  /* FIX(2.562915447) */
 #define F_3_072 25172  /* FIX(3.072711026) */
 #define CONST_BITS 13
 #define PASS1_BITS 2
 #define DESCALE_P1 (CONST_BITS - PASS1_BITS)
 #define DESCALE_P2 (CONST_BITS + PASS1_BITS)
 #define DO_FDCT_COMMON(PASS)  \
 {  \
  /* (Original)  \
   * z1 = (tmp12 + tmp13) * 0.541196100;  \
   * data2 = z1 + tmp13 * 0.765366865;  \
   * data6 = z1 + tmp12 * -1.847759065;  \
   *  \
   * (This implementation)  \
   * data2 = tmp13 * (0.541196100 + 0.765366865) + tmp12 * 0.541196100;  \
   * data6 = tmp13 * 0.541196100 + tmp12 * (0.541196100 - 1.847759065);  \
   */  \
  \
  tmp1312l = vec_mergeh(tmp13, tmp12);  \
  tmp1312h = vec_mergel(tmp13, tmp12);  \
  \
  out2l = vec_msums(tmp1312l, pw_f130_f054, pd_descale_p##PASS);  \
  out2h = vec_msums(tmp1312h, pw_f130_f054, pd_descale_p##PASS);  \
  out6l = vec_msums(tmp1312l, pw_f054_mf130, pd_descale_p##PASS);  \
  out6h = vec_msums(tmp1312h, pw_f054_mf130, pd_descale_p##PASS);  \
  \
  out2l = vec_sra(out2l, descale_p##PASS);  \
  out2h = vec_sra(out2h, descale_p##PASS);  \
  out6l = vec_sra(out6l, descale_p##PASS);  \
  out6h = vec_sra(out6h, descale_p##PASS);  \
  \
  out2 = vec_pack(out2l, out2h);  \
  out6 = vec_pack(out6l, out6h);  \
  \
  /* Odd part */  \
  \
  z3 = vec_add(tmp4, tmp6);  \
  z4 = vec_add(tmp5, tmp7);  \
  \
  /* (Original)  \
   * z5 = (z3 + z4) * 1.175875602;  \
   * z3 = z3 * -1.961570560;  z4 = z4 * -0.390180644;  \
   * z3 += z5;  z4 += z5;  \
   *  \
   * (This implementation)  \
   * z3 = z3 * (1.175875602 - 1.961570560) + z4 * 1.175875602;  \
   * z4 = z3 * 1.175875602 + z4 * (1.175875602 - 0.390180644);  \
   */  \
  \
  z34l = vec_mergeh(z3, z4);  \
  z34h = vec_mergel(z3, z4);  \
  \
  z3l = vec_msums(z34l, pw_mf078_f117, pd_descale_p##PASS);  \
  z3h = vec_msums(z34h, pw_mf078_f117, pd_descale_p##PASS);  \
  z4l = vec_msums(z34l, pw_f117_f078, pd_descale_p##PASS);  \
  z4h = vec_msums(z34h, pw_f117_f078, pd_descale_p##PASS);  \
  \
  /* (Original)  \
   * z1 = tmp4 + tmp7;  z2 = tmp5 + tmp6;  \
   * tmp4 = tmp4 * 0.298631336;  tmp5 = tmp5 * 2.053119869;  \
   * tmp6 = tmp6 * 3.072711026;  tmp7 = tmp7 * 1.501321110;  \
   * z1 = z1 * -0.899976223;  z2 = z2 * -2.562915447;  \
   * data7 = tmp4 + z1 + z3;  data5 = tmp5 + z2 + z4;  \
   * data3 = tmp6 + z2 + z3;  data1 = tmp7 + z1 + z4;  \
   *  \
   * (This implementation)  \
   * tmp4 = tmp4 * (0.298631336 - 0.899976223) + tmp7 * -0.899976223;  \
   * tmp5 = tmp5 * (2.053119869 - 2.562915447) + tmp6 * -2.562915447;  \
   * tmp6 = tmp5 * -2.562915447 + tmp6 * (3.072711026 - 2.562915447);  \
   * tmp7 = tmp4 * -0.899976223 + tmp7 * (1.501321110 - 0.899976223);  \
   * data7 = tmp4 + z3;  data5 = tmp5 + z4;  \
   * data3 = tmp6 + z3;  data1 = tmp7 + z4;  \
   */  \
  \
  tmp47l = vec_mergeh(tmp4, tmp7);  \
  tmp47h = vec_mergel(tmp4, tmp7);  \
  \
  out7l = vec_msums(tmp47l, pw_mf060_mf089, z3l);  \
  out7h = vec_msums(tmp47h, pw_mf060_mf089, z3h);  \
  out1l = vec_msums(tmp47l, pw_mf089_f060, z4l);  \
  out1h = vec_msums(tmp47h, pw_mf089_f060, z4h);  \
  \
  out7l = vec_sra(out7l, descale_p##PASS);  \
  out7h = vec_sra(out7h, descale_p##PASS);  \
  out1l = vec_sra(out1l, descale_p##PASS);  \
  out1h = vec_sra(out1h, descale_p##PASS);  \
  \
  out7 = vec_pack(out7l, out7h);  \
  out1 = vec_pack(out1l, out1h);  \
  \
  tmp56l = vec_mergeh(tmp5, tmp6);  \
  tmp56h = vec_mergel(tmp5, tmp6);  \
  \
  out5l = vec_msums(tmp56l, pw_mf050_mf256, z4l);  \
  out5h = vec_msums(tmp56h, pw_mf050_mf256, z4h);  \
  out3l = vec_msums(tmp56l, pw_mf256_f050, z3l);  \
  out3h = vec_msums(tmp56h, pw_mf256_f050, z3h);  \
  \
  out5l = vec_sra(out5l, descale_p##PASS);  \
  out5h = vec_sra(out5h, descale_p##PASS);  \
  out3l = vec_sra(out3l, descale_p##PASS);  \
  out3h = vec_sra(out3h, descale_p##PASS);  \
  \
  out5 = vec_pack(out5l, out5h);  \
  out3 = vec_pack(out3l, out3h);  \
 }
 #define DO_FDCT_ROWS()  \
 {  \
  /* Even part */  \
  \
  tmp10 = vec_add(tmp0, tmp3);  \
  tmp13 = vec_sub(tmp0, tmp3);  \
  tmp11 = vec_add(tmp1, tmp2);  \
  tmp12 = vec_sub(tmp1, tmp2);  \
  \
  out0  = vec_add(tmp10, tmp11);  \
  out0  = vec_sl(out0, pass1_bits);  \
  out4  = vec_sub(tmp10, tmp11);  \
  out4  = vec_sl(out4, pass1_bits);  \
  \
  DO_FDCT_COMMON(1);  \
 }
 #define DO_FDCT_COLS()  \
 {  \
  /* Even part */  \
  \
  tmp10 = vec_add(tmp0, tmp3);  \
  tmp13 = vec_sub(tmp0, tmp3);  \
  tmp11 = vec_add(tmp1, tmp2);  \
  tmp12 = vec_sub(tmp1, tmp2);  \
  \
  out0  = vec_add(tmp10, tmp11);  \
  out0  = vec_add(out0, pw_descale_p2x);  \
  out0  = vec_sra(out0, pass1_bits);  \
  out4  = vec_sub(tmp10, tmp11);  \
  out4  = vec_add(out4, pw_descale_p2x);  \
  out4  = vec_sra(out4, pass1_bits);  \
  \
  DO_FDCT_COMMON(2);  \
 }
 void
 jsimd_fdct_islow_altivec (DCTELEM *data)
 {
  __vector short row0, row1, row2, row3, row4, row5, row6, row7,
    col0, col1, col2, col3, col4, col5, col6, col7,
    tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, tmp10, tmp11, tmp12, tmp13,
    tmp47l, tmp47h, tmp56l, tmp56h, tmp1312l, tmp1312h,
    z3, z4, z34l, z34h,
    out0, out1, out2, out3, out4, out5, out6, out7;
  __vector int z3l, z3h, z4l, z4h,
    out1l, out1h, out2l, out2h, out3l, out3h, out5l, out5h, out6l, out6h,
    out7l, out7h;
  /* Constants */
  __vector short
    pw_f130_f054 = { __4X2(F_0_541 + F_0_765, F_0_541) },
    pw_f054_mf130 = { __4X2(F_0_541, F_0_541 - F_1_847) },
    pw_mf078_f117 = { __4X2(F_1_175 - F_1_961, F_1_175) },
    pw_f117_f078 = { __4X2(F_1_175, F_1_175 - F_0_390) },
    pw_mf060_mf089 = { __4X2(F_0_298 - F_0_899, -F_0_899) },
    pw_mf089_f060 = { __4X2(-F_0_899, F_1_501 - F_0_899) },
    pw_mf050_mf256 = { __4X2(F_2_053 - F_2_562, -F_2_562) },
    pw_mf256_f050 = { __4X2(-F_2_562, F_3_072 - F_2_562) },
    pw_descale_p2x = { __8X(1 << (PASS1_BITS - 1)) };
  __vector unsigned short pass1_bits = { __8X(PASS1_BITS) };
  __vector int pd_descale_p1 = { __4X(1 << (DESCALE_P1 - 1)) },
    pd_descale_p2 = { __4X(1 << (DESCALE_P2 - 1)) };
  __vector unsigned int descale_p1 = { __4X(DESCALE_P1) },
    descale_p2 = { __4X(DESCALE_P2) };
  /* Pass 1: process rows */
  row0 = vec_ld(0, data);
  row1 = vec_ld(16, data);
  row2 = vec_ld(32, data);
  row3 = vec_ld(48, data);
  row4 = vec_ld(64, data);
  row5 = vec_ld(80, data);
  row6 = vec_ld(96, data);
  row7 = vec_ld(112, data);
  TRANSPOSE(row, col);
  tmp0 = vec_add(col0, col7);
  tmp7 = vec_sub(col0, col7);
  tmp1 = vec_add(col1, col6);
  tmp6 = vec_sub(col1, col6);
  tmp2 = vec_add(col2, col5);
  tmp5 = vec_sub(col2, col5);
  tmp3 = vec_add(col3, col4);
  tmp4 = vec_sub(col3, col4);
  DO_FDCT_ROWS();
  /* Pass 2: process columns */
  TRANSPOSE(out, row);
  tmp0 = vec_add(row0, row7);
  tmp7 = vec_sub(row0, row7);
  tmp1 = vec_add(row1, row6);
  tmp6 = vec_sub(row1, row6);
  tmp2 = vec_add(row2, row5);
  tmp5 = vec_sub(row2, row5);
  tmp3 = vec_add(row3, row4);
  tmp4 = vec_sub(row3, row4);
  DO_FDCT_COLS();
  vec_st(out0, 0, data);
  vec_st(out1, 16, data);
  vec_st(out2, 32, data);
  vec_st(out3, 48, data);
  vec_st(out4, 64, data);
  vec_st(out5, 80, data);
  vec_st(out6, 96, data);
  vec_st(out7, 112, data);
 }
--- a/simd/jfdctint-mmx.asm
+++ b/simd/jfdctint-mmx.asm
@@ -444,11 +444,11 @@ EXTN(jsimd_fdct_islow_mmx):
        psubw   mm6,mm4                 ; mm6=tmp12
        movq    mm7,mm5
-        paddw   mm5,mm0                 ; mm5=tmp10+tmp11
+        paddsw   mm5,mm0                 ; mm5=tmp10+tmp11
-        psubw   mm7,mm0                 ; mm7=tmp10-tmp11
+        psubsw   mm7,mm0                 ; mm7=tmp10-tmp11
-        paddw   mm5,[GOTOFF(ebx,PW_DESCALE_P2X)]
+        paddsw   mm5,[GOTOFF(ebx,PW_DESCALE_P2X)]
-        paddw   mm7,[GOTOFF(ebx,PW_DESCALE_P2X)]
+        paddsw   mm7,[GOTOFF(ebx,PW_DESCALE_P2X)]
        psraw   mm5,PASS1_BITS          ; mm5=data0
        psraw   mm7,PASS1_BITS          ; mm7=data4
--- a/simd/jfdctint-sse2-64.asm
+++ b/simd/jfdctint-sse2-64.asm
@@ -454,11 +454,11 @@ EXTN(jsimd_fdct_islow_sse2):
        psubw   xmm6,xmm4               ; xmm6=tmp12
        movdqa  xmm5,xmm7
-        paddw   xmm7,xmm2               ; xmm7=tmp10+tmp11
+        paddsw   xmm7,xmm2               ; xmm7=tmp10+tmp11
-        psubw   xmm5,xmm2               ; xmm5=tmp10-tmp11
+        psubsw   xmm5,xmm2               ; xmm5=tmp10-tmp11
-        paddw   xmm7,[rel PW_DESCALE_P2X]
+        paddsw   xmm7,[rel PW_DESCALE_P2X]
-        paddw   xmm5,[rel PW_DESCALE_P2X]
+        paddsw   xmm5,[rel PW_DESCALE_P2X]
        psraw   xmm7,PASS1_BITS         ; xmm7=data0
        psraw   xmm5,PASS1_BITS         ; xmm5=data4
--- a/simd/jfdctint-sse2.asm
+++ b/simd/jfdctint-sse2.asm
@@ -462,11 +462,11 @@ EXTN(jsimd_fdct_islow_sse2):
        psubw   xmm6,xmm4               ; xmm6=tmp12
        movdqa  xmm5,xmm7
-        paddw   xmm7,xmm2               ; xmm7=tmp10+tmp11
+        paddsw   xmm7,xmm2               ; xmm7=tmp10+tmp11
-        psubw   xmm5,xmm2               ; xmm5=tmp10-tmp11
+        psubsw   xmm5,xmm2               ; xmm5=tmp10-tmp11
-        paddw   xmm7,[GOTOFF(ebx,PW_DESCALE_P2X)]
+        paddsw   xmm7,[GOTOFF(ebx,PW_DESCALE_P2X)]
-        paddw   xmm5,[GOTOFF(ebx,PW_DESCALE_P2X)]
+        paddsw   xmm5,[GOTOFF(ebx,PW_DESCALE_P2X)]
        psraw   xmm7,PASS1_BITS         ; xmm7=data0
        psraw   xmm5,PASS1_BITS         ; xmm5=data4
--- a/simd/jidctfst-altivec.c
+++ b/simd/jidctfst-altivec.c
@@ -0,0 +1,256 @@
 /*
 * AltiVec optimizations for libjpeg-turbo
 *
 * Copyright (C) 2014, D. R. Commander.
 * All rights reserved.
 * This software is provided 'as-is', without any express or implied
 * warranty.  In no event will the authors be held liable for any damages
 * arising from the use of this software.
 *
 * Permission is granted to anyone to use this software for any purpose,
 * including commercial applications, and to alter it and redistribute it
 * freely, subject to the following restrictions:
 *
 * 1. The origin of this software must not be misrepresented; you must not
 *    claim that you wrote the original software. If you use this software
 *    in a product, an acknowledgment in the product documentation would be
 *    appreciated but is not required.
 * 2. Altered source versions must be plainly marked as such, and must not be
 *    misrepresented as being the original software.
 * 3. This notice may not be removed or altered from any source distribution.
 */
 /* FAST INTEGER INVERSE DCT
 *
 * This is similar to the SSE2 implementation, except that we left-shift the
 * constants by 1 less bit (the -1 in CONST_SHIFT.)  This is because
 * vec_madds(arg1, arg2, arg3) generates the 16-bit saturated sum of:
 *   the elements in arg3 + the most significant 17 bits of
 *     (the elements in arg1 * the elements in arg2).
 */
 #include "jsimd_altivec.h"
 #define F_1_082 277              /* FIX(1.082392200) */
 #define F_1_414 362              /* FIX(1.414213562) */
 #define F_1_847 473              /* FIX(1.847759065) */
 #define F_2_613 669              /* FIX(2.613125930) */
 #define F_1_613 (F_2_613 - 256)  /* FIX(2.613125930) - FIX(1) */
 #define CONST_BITS 8
 #define PASS1_BITS 2
 #define PRE_MULTIPLY_SCALE_BITS 2
 #define CONST_SHIFT (16 - PRE_MULTIPLY_SCALE_BITS - CONST_BITS - 1)
 #define DO_IDCT(in)  \
 {  \
  /* Even part */  \
  \
  tmp10 = vec_add(in##0, in##4);  \
  tmp11 = vec_sub(in##0, in##4);  \
  tmp13 = vec_add(in##2, in##6);  \
  \
  tmp12 = vec_sub(in##2, in##6);  \
  tmp12 = vec_sl(tmp12, pre_multiply_scale_bits);  \
  tmp12 = vec_madds(tmp12, pw_F1414, zero);  \
  tmp12 = vec_sub(tmp12, tmp13);  \
  \
  tmp0 = vec_add(tmp10, tmp13);  \
  tmp3 = vec_sub(tmp10, tmp13);  \
  tmp1 = vec_add(tmp11, tmp12);  \
  tmp2 = vec_sub(tmp11, tmp12);  \
  \
  /* Odd part */  \
  \
  z13 = vec_add(in##5, in##3);  \
  z10 = vec_sub(in##5, in##3);  \
  z10s = vec_sl(z10, pre_multiply_scale_bits);  \
  z11 = vec_add(in##1, in##7);  \
  z12s = vec_sub(in##1, in##7);  \
  z12s = vec_sl(z12s, pre_multiply_scale_bits);  \
  \
  tmp11 = vec_sub(z11, z13);  \
  tmp11 = vec_sl(tmp11, pre_multiply_scale_bits);  \
  tmp11 = vec_madds(tmp11, pw_F1414, zero);  \
  \
  tmp7 = vec_add(z11, z13);  \
  \
  /* To avoid overflow...  \
   *  \
   * (Original)  \
   * tmp12 = -2.613125930 * z10 + z5;  \
   *  \
   * (This implementation)  \
   * tmp12 = (-1.613125930 - 1) * z10 + z5;  \
   *       = -1.613125930 * z10 - z10 + z5;  \
   */  \
  \
  z5 = vec_add(z10s, z12s);  \
  z5 = vec_madds(z5, pw_F1847, zero);  \
  \
  tmp10 = vec_madds(z12s, pw_F1082, zero);  \
  tmp10 = vec_sub(tmp10, z5);  \
  tmp12 = vec_madds(z10s, pw_MF1613, z5);  \
  tmp12 = vec_sub(tmp12, z10);  \
  \
  tmp6 = vec_sub(tmp12, tmp7);  \
  tmp5 = vec_sub(tmp11, tmp6);  \
  tmp4 = vec_add(tmp10, tmp5);  \
  \
  out0 = vec_add(tmp0, tmp7);  \
  out1 = vec_add(tmp1, tmp6);  \
  out2 = vec_add(tmp2, tmp5);  \
  out3 = vec_sub(tmp3, tmp4);  \
  out4 = vec_add(tmp3, tmp4);  \
  out5 = vec_sub(tmp2, tmp5);  \
  out6 = vec_sub(tmp1, tmp6);  \
  out7 = vec_sub(tmp0, tmp7);  \
 }
 void
 jsimd_idct_ifast_altivec (void * dct_table_, JCOEFPTR coef_block,
                          JSAMPARRAY output_buf, JDIMENSION output_col)
 {
  short *dct_table = (short *)dct_table_;
  __vector short row0, row1, row2, row3, row4, row5, row6, row7,
    col0, col1, col2, col3, col4, col5, col6, col7,
    quant0, quant1, quant2, quant3, quant4, quant5, quant6, quant7,
    tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, tmp10, tmp11, tmp12, tmp13,
    z5, z10, z10s, z11, z12s, z13,
    out0, out1, out2, out3, out4, out5, out6, out7;
  __vector signed char outb;
  int *outptr;
  /* Constants */
  __vector short zero = { __8X(0) },
    pw_F1414 = { __8X(F_1_414 << CONST_SHIFT) },
    pw_F1847 = { __8X(F_1_847 << CONST_SHIFT) },
    pw_MF1613 = { __8X(-F_1_613 << CONST_SHIFT) },
    pw_F1082 = { __8X(F_1_082 << CONST_SHIFT) };
  __vector unsigned short
    pre_multiply_scale_bits = { __8X(PRE_MULTIPLY_SCALE_BITS) },
    pass1_bits3 = { __8X(PASS1_BITS + 3) };
  __vector signed char pb_centerjsamp = { __16X(CENTERJSAMPLE) };
  /* Pass 1: process columns */
  col0 = vec_ld(0, coef_block);
  col1 = vec_ld(16, coef_block);
  col2 = vec_ld(32, coef_block);
  col3 = vec_ld(48, coef_block);
  col4 = vec_ld(64, coef_block);
  col5 = vec_ld(80, coef_block);
  col6 = vec_ld(96, coef_block);
  col7 = vec_ld(112, coef_block);
  tmp1 = vec_or(col1, col2);
  tmp2 = vec_or(col3, col4);
  tmp1 = vec_or(tmp1, tmp2);
  tmp3 = vec_or(col5, col6);
  tmp3 = vec_or(tmp3, col7);
  tmp1 = vec_or(tmp1, tmp3);
  quant0 = *(__vector short *)&dct_table[0];
  col0 = vec_mladd(col0, quant0, zero);
  if (vec_all_eq(tmp1, zero)) {
    /* AC terms all zero */
    row0 = vec_splat(col0, 0);
    row1 = vec_splat(col0, 1);
    row2 = vec_splat(col0, 2);
    row3 = vec_splat(col0, 3);
    row4 = vec_splat(col0, 4);
    row5 = vec_splat(col0, 5);
    row6 = vec_splat(col0, 6);
    row7 = vec_splat(col0, 7);
  } else {
    quant1 = *(__vector short *)&dct_table[8];
    quant2 = *(__vector short *)&dct_table[16];
    quant3 = *(__vector short *)&dct_table[24];
    quant4 = *(__vector short *)&dct_table[32];
    quant5 = *(__vector short *)&dct_table[40];
    quant6 = *(__vector short *)&dct_table[48];
    quant7 = *(__vector short *)&dct_table[56];
    col1 = vec_mladd(col1, quant1, zero);
    col2 = vec_mladd(col2, quant2, zero);
    col3 = vec_mladd(col3, quant3, zero);
    col4 = vec_mladd(col4, quant4, zero);
    col5 = vec_mladd(col5, quant5, zero);
    col6 = vec_mladd(col6, quant6, zero);
    col7 = vec_mladd(col7, quant7, zero);
    DO_IDCT(col);
    TRANSPOSE(out, row);
  }
  /* Pass 2: process rows */
  DO_IDCT(row);
  out0 = vec_sra(out0, pass1_bits3);
  out1 = vec_sra(out1, pass1_bits3);
  out2 = vec_sra(out2, pass1_bits3);
  out3 = vec_sra(out3, pass1_bits3);
  out4 = vec_sra(out4, pass1_bits3);
  out5 = vec_sra(out5, pass1_bits3);
  out6 = vec_sra(out6, pass1_bits3);
  out7 = vec_sra(out7, pass1_bits3);
  TRANSPOSE(out, col);
  outb = vec_packs(col0, col0);
  outb = vec_add(outb, pb_centerjsamp);
  outptr = (int *)(output_buf[0] + output_col);
  vec_ste((__vector int)outb, 0, outptr);
  vec_ste((__vector int)outb, 4, outptr);
  outb = vec_packs(col1, col1);
  outb = vec_add(outb, pb_centerjsamp);
  outptr = (int *)(output_buf[1] + output_col);
  vec_ste((__vector int)outb, 0, outptr);
  vec_ste((__vector int)outb, 4, outptr);
  outb = vec_packs(col2, col2);
  outb = vec_add(outb, pb_centerjsamp);
  outptr = (int *)(output_buf[2] + output_col);
  vec_ste((__vector int)outb, 0, outptr);
  vec_ste((__vector int)outb, 4, outptr);
  outb = vec_packs(col3, col3);
  outb = vec_add(outb, pb_centerjsamp);
  outptr = (int *)(output_buf[3] + output_col);
  vec_ste((__vector int)outb, 0, outptr);
  vec_ste((__vector int)outb, 4, outptr);
  outb = vec_packs(col4, col4);
  outb = vec_add(outb, pb_centerjsamp);
  outptr = (int *)(output_buf[4] + output_col);
  vec_ste((__vector int)outb, 0, outptr);
  vec_ste((__vector int)outb, 4, outptr);
  outb = vec_packs(col5, col5);
  outb = vec_add(outb, pb_centerjsamp);
  outptr = (int *)(output_buf[5] + output_col);
  vec_ste((__vector int)outb, 0, outptr);
  vec_ste((__vector int)outb, 4, outptr);
  outb = vec_packs(col6, col6);
  outb = vec_add(outb, pb_centerjsamp);
  outptr = (int *)(output_buf[6] + output_col);
  vec_ste((__vector int)outb, 0, outptr);
  vec_ste((__vector int)outb, 4, outptr);
  outb = vec_packs(col7, col7);
  outb = vec_add(outb, pb_centerjsamp);
  outptr = (int *)(output_buf[7] + output_col);
  vec_ste((__vector int)outb, 0, outptr);
  vec_ste((__vector int)outb, 4, outptr);
 }
--- a/simd/jidctint-altivec.c
+++ b/simd/jidctint-altivec.c
@@ -0,0 +1,358 @@
 /*
 * AltiVec optimizations for libjpeg-turbo
 *
 * Copyright (C) 2014, D. R. Commander.
 * All rights reserved.
 * This software is provided 'as-is', without any express or implied
 * warranty.  In no event will the authors be held liable for any damages
 * arising from the use of this software.
 *
 * Permission is granted to anyone to use this software for any purpose,
 * including commercial applications, and to alter it and redistribute it
 * freely, subject to the following restrictions:
 *
 * 1. The origin of this software must not be misrepresented; you must not
 *    claim that you wrote the original software. If you use this software
 *    in a product, an acknowledgment in the product documentation would be
 *    appreciated but is not required.
 * 2. Altered source versions must be plainly marked as such, and must not be
 *    misrepresented as being the original software.
 * 3. This notice may not be removed or altered from any source distribution.
 */
 /* SLOW INTEGER INVERSE DCT */
 #include "jsimd_altivec.h"
 #define F_0_298 2446   /* FIX(0.298631336) */
 #define F_0_390 3196   /* FIX(0.390180644) */
 #define F_0_541 4433   /* FIX(0.541196100) */
 #define F_0_765 6270   /* FIX(0.765366865) */
 #define F_0_899 7373   /* FIX(0.899976223) */
 #define F_1_175 9633   /* FIX(1.175875602) */
 #define F_1_501 12299  /* FIX(1.501321110) */
 #define F_1_847 15137  /* FIX(1.847759065) */
 #define F_1_961 16069  /* FIX(1.961570560) */
 #define F_2_053 16819  /* FIX(2.053119869) */
 #define F_2_562 20995  /* FIX(2.562915447) */
 #define F_3_072 25172  /* FIX(3.072711026) */
 #define CONST_BITS 13
 #define PASS1_BITS 2
 #define DESCALE_P1 (CONST_BITS - PASS1_BITS)
 #define DESCALE_P2 (CONST_BITS + PASS1_BITS + 3)
 #define DO_IDCT(in, PASS)  \
 {  \
  /* Even part  \
   *  \
   * (Original)  \
   * z1 = (z2 + z3) * 0.541196100;  \
   * tmp2 = z1 + z3 * -1.847759065;  \
   * tmp3 = z1 + z2 * 0.765366865;  \
   *  \
   * (This implementation)  \
   * tmp2 = z2 * 0.541196100 + z3 * (0.541196100 - 1.847759065);  \
   * tmp3 = z2 * (0.541196100 + 0.765366865) + z3 * 0.541196100;  \
   */  \
  \
  in##26l = vec_mergeh(in##2, in##6);  \
  in##26h = vec_mergel(in##2, in##6);  \
  \
  tmp3l = vec_msums(in##26l, pw_f130_f054, zero32);  \
  tmp3h = vec_msums(in##26h, pw_f130_f054, zero32);  \
  tmp2l = vec_msums(in##26l, pw_f054_mf130, zero32);  \
  tmp2h = vec_msums(in##26h, pw_f054_mf130, zero32);  \
  \
  tmp0 = vec_add(in##0, in##4);  \
  tmp1 = vec_sub(in##0, in##4);  \
  \
  tmp0l = vec_unpackh(tmp0);  \
  tmp0h = vec_unpackl(tmp0);  \
  tmp0l = vec_sl(tmp0l, const_bits);  \
  tmp0h = vec_sl(tmp0h, const_bits);  \
  tmp0l = vec_add(tmp0l, pd_descale_p##PASS);  \
  tmp0h = vec_add(tmp0h, pd_descale_p##PASS);  \
  \
  tmp10l = vec_add(tmp0l, tmp3l);  \
  tmp10h = vec_add(tmp0h, tmp3h);  \
  tmp13l = vec_sub(tmp0l, tmp3l);  \
  tmp13h = vec_sub(tmp0h, tmp3h);  \
  \
  tmp1l = vec_unpackh(tmp1);  \
  tmp1h = vec_unpackl(tmp1);  \
  tmp1l = vec_sl(tmp1l, const_bits);  \
  tmp1h = vec_sl(tmp1h, const_bits);  \
  tmp1l = vec_add(tmp1l, pd_descale_p##PASS);  \
  tmp1h = vec_add(tmp1h, pd_descale_p##PASS);  \
  \
  tmp11l = vec_add(tmp1l, tmp2l);  \
  tmp11h = vec_add(tmp1h, tmp2h);  \
  tmp12l = vec_sub(tmp1l, tmp2l);  \
  tmp12h = vec_sub(tmp1h, tmp2h);  \
  \
  /* Odd part */  \
  \
  z3 = vec_add(in##3, in##7);  \
  z4 = vec_add(in##1, in##5);  \
  \
  /* (Original)  \
   * z5 = (z3 + z4) * 1.175875602;  \
   * z3 = z3 * -1.961570560;  z4 = z4 * -0.390180644;  \
   * z3 += z5;  z4 += z5;  \
   *  \
   * (This implementation)  \
   * z3 = z3 * (1.175875602 - 1.961570560) + z4 * 1.175875602;  \
   * z4 = z3 * 1.175875602 + z4 * (1.175875602 - 0.390180644);  \
   */  \
  \
  z34l = vec_mergeh(z3, z4);  \
  z34h = vec_mergel(z3, z4);  \
  \
  z3l = vec_msums(z34l, pw_mf078_f117, zero32);  \
  z3h = vec_msums(z34h, pw_mf078_f117, zero32);  \
  z4l = vec_msums(z34l, pw_f117_f078, zero32);  \
  z4h = vec_msums(z34h, pw_f117_f078, zero32);  \
  \
  /* (Original)  \
   * z1 = tmp0 + tmp3;  z2 = tmp1 + tmp2;  \
   * tmp0 = tmp0 * 0.298631336;  tmp1 = tmp1 * 2.053119869;  \
   * tmp2 = tmp2 * 3.072711026;  tmp3 = tmp3 * 1.501321110;  \
   * z1 = z1 * -0.899976223;  z2 = z2 * -2.562915447;  \
   * tmp0 += z1 + z3;  tmp1 += z2 + z4;  \
   * tmp2 += z2 + z3;  tmp3 += z1 + z4;  \
   *  \
   * (This implementation)  \
   * tmp0 = tmp0 * (0.298631336 - 0.899976223) + tmp3 * -0.899976223;  \
   * tmp1 = tmp1 * (2.053119869 - 2.562915447) + tmp2 * -2.562915447;  \
   * tmp2 = tmp1 * -2.562915447 + tmp2 * (3.072711026 - 2.562915447);  \
   * tmp3 = tmp0 * -0.899976223 + tmp3 * (1.501321110 - 0.899976223);  \
   * tmp0 += z3;  tmp1 += z4;  \
   * tmp2 += z3;  tmp3 += z4;  \
   */  \
  \
  in##71l = vec_mergeh(in##7, in##1);  \
  in##71h = vec_mergel(in##7, in##1);  \
  \
  tmp0l = vec_msums(in##71l, pw_mf060_mf089, z3l);  \
  tmp0h = vec_msums(in##71h, pw_mf060_mf089, z3h);  \
  tmp3l = vec_msums(in##71l, pw_mf089_f060, z4l);  \
  tmp3h = vec_msums(in##71h, pw_mf089_f060, z4h);  \
  \
  in##53l = vec_mergeh(in##5, in##3);  \
  in##53h = vec_mergel(in##5, in##3);  \
  \
  tmp1l = vec_msums(in##53l, pw_mf050_mf256, z4l);  \
  tmp1h = vec_msums(in##53h, pw_mf050_mf256, z4h);  \
  tmp2l = vec_msums(in##53l, pw_mf256_f050, z3l);  \
  tmp2h = vec_msums(in##53h, pw_mf256_f050, z3h);  \
  \
  /* Final output stage */  \
  \
  out0l = vec_add(tmp10l, tmp3l);  \
  out0h = vec_add(tmp10h, tmp3h);  \
  out7l = vec_sub(tmp10l, tmp3l);  \
  out7h = vec_sub(tmp10h, tmp3h);  \
  \
  out0l = vec_sra(out0l, descale_p##PASS);  \
  out0h = vec_sra(out0h, descale_p##PASS);  \
  out7l = vec_sra(out7l, descale_p##PASS);  \
  out7h = vec_sra(out7h, descale_p##PASS);  \
  \
  out0 = vec_pack(out0l, out0h);  \
  out7 = vec_pack(out7l, out7h);  \
  \
  out1l = vec_add(tmp11l, tmp2l);  \
  out1h = vec_add(tmp11h, tmp2h);  \
  out6l = vec_sub(tmp11l, tmp2l);  \
  out6h = vec_sub(tmp11h, tmp2h);  \
  \
  out1l = vec_sra(out1l, descale_p##PASS);  \
  out1h = vec_sra(out1h, descale_p##PASS);  \
  out6l = vec_sra(out6l, descale_p##PASS);  \
  out6h = vec_sra(out6h, descale_p##PASS);  \
  \
  out1 = vec_pack(out1l, out1h);  \
  out6 = vec_pack(out6l, out6h);  \
  \
  out2l = vec_add(tmp12l, tmp1l);  \
  out2h = vec_add(tmp12h, tmp1h);  \
  out5l = vec_sub(tmp12l, tmp1l);  \
  out5h = vec_sub(tmp12h, tmp1h);  \
  \
  out2l = vec_sra(out2l, descale_p##PASS);  \
  out2h = vec_sra(out2h, descale_p##PASS);  \
  out5l = vec_sra(out5l, descale_p##PASS);  \
  out5h = vec_sra(out5h, descale_p##PASS);  \
  \
  out2 = vec_pack(out2l, out2h);  \
  out5 = vec_pack(out5l, out5h);  \
  \
  out3l = vec_add(tmp13l, tmp0l);  \
  out3h = vec_add(tmp13h, tmp0h);  \
  out4l = vec_sub(tmp13l, tmp0l);  \
  out4h = vec_sub(tmp13h, tmp0h);  \
  \
  out3l = vec_sra(out3l, descale_p##PASS);  \
  out3h = vec_sra(out3h, descale_p##PASS);  \
  out4l = vec_sra(out4l, descale_p##PASS);  \
  out4h = vec_sra(out4h, descale_p##PASS);  \
  \
  out3 = vec_pack(out3l, out3h);  \
  out4 = vec_pack(out4l, out4h);  \
 }
 void
 jsimd_idct_islow_altivec (void * dct_table_, JCOEFPTR coef_block,
                          JSAMPARRAY output_buf, JDIMENSION output_col)
 {
  short *dct_table = (short *)dct_table_;
  __vector short row0, row1, row2, row3, row4, row5, row6, row7,
    col0, col1, col2, col3, col4, col5, col6, col7,
    quant0, quant1, quant2, quant3, quant4, quant5, quant6, quant7,
    tmp0, tmp1, tmp2, tmp3, z3, z4,
    z34l, z34h, col71l, col71h, col26l, col26h, col53l, col53h,
    row71l, row71h, row26l, row26h, row53l, row53h,
    out0, out1, out2, out3, out4, out5, out6, out7;
  __vector int tmp0l, tmp0h, tmp1l, tmp1h, tmp2l, tmp2h, tmp3l, tmp3h,
    tmp10l, tmp10h, tmp11l, tmp11h, tmp12l, tmp12h, tmp13l, tmp13h,
    z3l, z3h, z4l, z4h,
    out0l, out0h, out1l, out1h, out2l, out2h, out3l, out3h, out4l, out4h,
    out5l, out5h, out6l, out6h, out7l, out7h;
  __vector signed char outb;
  int *outptr;
  /* Constants */
  __vector short zero16 = { __8X(0) },
    pw_f130_f054 = { __4X2(F_0_541 + F_0_765, F_0_541) },
    pw_f054_mf130 = { __4X2(F_0_541, F_0_541 - F_1_847) },
    pw_mf078_f117 = { __4X2(F_1_175 - F_1_961, F_1_175) },
    pw_f117_f078 = { __4X2(F_1_175, F_1_175 - F_0_390) },
    pw_mf060_mf089 = { __4X2(F_0_298 - F_0_899, -F_0_899) },
    pw_mf089_f060 = { __4X2(-F_0_899, F_1_501 - F_0_899) },
    pw_mf050_mf256 = { __4X2(F_2_053 - F_2_562, -F_2_562) },
    pw_mf256_f050 = { __4X2(-F_2_562, F_3_072 - F_2_562) };
  __vector unsigned short pass1_bits = { __8X(PASS1_BITS) };
  __vector int zero32 = { __4X(0) },
    pd_descale_p1 = { __4X(1 << (DESCALE_P1 - 1)) },
    pd_descale_p2 = { __4X(1 << (DESCALE_P2 - 1)) };
  __vector unsigned int descale_p1 = { __4X(DESCALE_P1) },
    descale_p2 = { __4X(DESCALE_P2) },
    const_bits = { __4X(CONST_BITS) };
  __vector signed char pb_centerjsamp = { __16X(CENTERJSAMPLE) };
  /* Pass 1: process columns */
  col0 = *(__vector short *)&coef_block[0];
  col1 = *(__vector short *)&coef_block[8];
  col2 = *(__vector short *)&coef_block[16];
  col3 = *(__vector short *)&coef_block[24];
  col4 = *(__vector short *)&coef_block[32];
  col5 = *(__vector short *)&coef_block[40];
  col6 = *(__vector short *)&coef_block[48];
  col7 = *(__vector short *)&coef_block[56];
  tmp1 = vec_or(col1, col2);
  tmp2 = vec_or(col3, col4);
  tmp1 = vec_or(tmp1, tmp2);
  tmp3 = vec_or(col5, col6);
  tmp3 = vec_or(tmp3, col7);
  tmp1 = vec_or(tmp1, tmp3);
  quant0 = *(__vector short *)&dct_table[0];
  col0 = vec_mladd(col0, quant0, zero16);
  if (vec_all_eq(tmp1, zero16)) {
    /* AC terms all zero */
    col0 = vec_sl(col0, pass1_bits);
    row0 = vec_splat(col0, 0);
    row1 = vec_splat(col0, 1);
    row2 = vec_splat(col0, 2);
    row3 = vec_splat(col0, 3);
    row4 = vec_splat(col0, 4);
    row5 = vec_splat(col0, 5);
    row6 = vec_splat(col0, 6);
    row7 = vec_splat(col0, 7);
  } else {
    quant1 = *(__vector short *)&dct_table[8];
    quant2 = *(__vector short *)&dct_table[16];
    quant3 = *(__vector short *)&dct_table[24];
    quant4 = *(__vector short *)&dct_table[32];
    quant5 = *(__vector short *)&dct_table[40];
    quant6 = *(__vector short *)&dct_table[48];
    quant7 = *(__vector short *)&dct_table[56];
    col1 = vec_mladd(col1, quant1, zero16);
    col2 = vec_mladd(col2, quant2, zero16);
    col3 = vec_mladd(col3, quant3, zero16);
    col4 = vec_mladd(col4, quant4, zero16);
    col5 = vec_mladd(col5, quant5, zero16);
    col6 = vec_mladd(col6, quant6, zero16);
    col7 = vec_mladd(col7, quant7, zero16);
    DO_IDCT(col, 1);
    TRANSPOSE(out, row);
  }
  /* Pass 2: process rows */
  DO_IDCT(row, 2);
  TRANSPOSE(out, col);
  outb = vec_packs(col0, col0);
  outb = vec_add(outb, pb_centerjsamp);
  outptr = (int *)(output_buf[0] + output_col);
  vec_ste((__vector int)outb, 0, outptr);
  vec_ste((__vector int)outb, 4, outptr);
  outb = vec_packs(col1, col1);
  outb = vec_add(outb, pb_centerjsamp);
  outptr = (int *)(output_buf[1] + output_col);
  vec_ste((__vector int)outb, 0, outptr);
  vec_ste((__vector int)outb, 4, outptr);
  outb = vec_packs(col2, col2);
  outb = vec_add(outb, pb_centerjsamp);
  outptr = (int *)(output_buf[2] + output_col);
  vec_ste((__vector int)outb, 0, outptr);
  vec_ste((__vector int)outb, 4, outptr);
  outb = vec_packs(col3, col3);
  outb = vec_add(outb, pb_centerjsamp);
  outptr = (int *)(output_buf[3] + output_col);
  vec_ste((__vector int)outb, 0, outptr);
  vec_ste((__vector int)outb, 4, outptr);
  outb = vec_packs(col4, col4);
  outb = vec_add(outb, pb_centerjsamp);
  outptr = (int *)(output_buf[4] + output_col);
  vec_ste((__vector int)outb, 0, outptr);
  vec_ste((__vector int)outb, 4, outptr);
  outb = vec_packs(col5, col5);
  outb = vec_add(outb, pb_centerjsamp);
  outptr = (int *)(output_buf[5] + output_col);
  vec_ste((__vector int)outb, 0, outptr);
  vec_ste((__vector int)outb, 4, outptr);
  outb = vec_packs(col6, col6);
  outb = vec_add(outb, pb_centerjsamp);
  outptr = (int *)(output_buf[6] + output_col);
  vec_ste((__vector int)outb, 0, outptr);
  vec_ste((__vector int)outb, 4, outptr);
  outb = vec_packs(col7, col7);
  outb = vec_add(outb, pb_centerjsamp);
  outptr = (int *)(output_buf[7] + output_col);
  vec_ste((__vector int)outb, 0, outptr);
  vec_ste((__vector int)outb, 4, outptr);
 }
--- a/simd/jquanti-altivec.c
+++ b/simd/jquanti-altivec.c
@@ -0,0 +1,236 @@
 /*
 * AltiVec optimizations for libjpeg-turbo
 *
 * Copyright (C) 2014, D. R. Commander.
 * All rights reserved.
 * This software is provided 'as-is', without any express or implied
 * warranty.  In no event will the authors be held liable for any damages
 * arising from the use of this software.
 *
 * Permission is granted to anyone to use this software for any purpose,
 * including commercial applications, and to alter it and redistribute it
 * freely, subject to the following restrictions:
 *
 * 1. The origin of this software must not be misrepresented; you must not
 *    claim that you wrote the original software. If you use this software
 *    in a product, an acknowledgment in the product documentation would be
 *    appreciated but is not required.
 * 2. Altered source versions must be plainly marked as such, and must not be
 *    misrepresented as being the original software.
 * 3. This notice may not be removed or altered from any source distribution.
 */
 /* INTEGER QUANTIZATION AND SAMPLE CONVERSION */
 #include "jsimd_altivec.h"
 /* NOTE: The address will either be aligned or offset by 8 bytes, so we can
 * always get the data we want by using a single vector load (although we may
 * have to permute the result.)
 */
 #define LOAD_ROW(row) {  \
  elemptr = sample_data[row] + start_col;  \
  in##row = vec_ld(0, elemptr);  \
  if ((size_t)elemptr & 15)  \
    in##row = vec_perm(in##row, in##row, vec_lvsl(0, elemptr));  \
 }
 void
 jsimd_convsamp_altivec (JSAMPARRAY sample_data, JDIMENSION start_col,
                        DCTELEM * workspace)
 {
  JSAMPROW elemptr;
  __vector unsigned char in0, in1, in2, in3, in4, in5, in6, in7;
  __vector short out0, out1, out2, out3, out4, out5, out6, out7;
  /* Constants */
  __vector short pw_centerjsamp = { __8X(CENTERJSAMPLE) };
  __vector unsigned char zero = { __16X(0) };
  LOAD_ROW(0);
  LOAD_ROW(1);
  LOAD_ROW(2);
  LOAD_ROW(3);
  LOAD_ROW(4);
  LOAD_ROW(5);
  LOAD_ROW(6);
  LOAD_ROW(7);
  out0 = (__vector short)vec_mergeh(zero, in0);
  out1 = (__vector short)vec_mergeh(zero, in1);
  out2 = (__vector short)vec_mergeh(zero, in2);
  out3 = (__vector short)vec_mergeh(zero, in3);
  out4 = (__vector short)vec_mergeh(zero, in4);
  out5 = (__vector short)vec_mergeh(zero, in5);
  out6 = (__vector short)vec_mergeh(zero, in6);
  out7 = (__vector short)vec_mergeh(zero, in7);
  out0 = vec_sub(out0, pw_centerjsamp);
  out1 = vec_sub(out1, pw_centerjsamp);
  out2 = vec_sub(out2, pw_centerjsamp);
  out3 = vec_sub(out3, pw_centerjsamp);
  out4 = vec_sub(out4, pw_centerjsamp);
  out5 = vec_sub(out5, pw_centerjsamp);
  out6 = vec_sub(out6, pw_centerjsamp);
  out7 = vec_sub(out7, pw_centerjsamp);
  vec_st(out0, 0, workspace);
  vec_st(out1, 16, workspace);
  vec_st(out2, 32, workspace);
  vec_st(out3, 48, workspace);
  vec_st(out4, 64, workspace);
  vec_st(out5, 80, workspace);
  vec_st(out6, 96, workspace);
  vec_st(out7, 112, workspace);
 }
 #define WORD_BIT 16
 /* There is no AltiVec unsigned multiply instruction, hence this. */
 #define MULTIPLY(vs0, vs1, out) {  \
  tmpe = vec_mule((__vector unsigned short)vs0,  \
                  (__vector unsigned short)vs1);  \
  tmpo = vec_mulo((__vector unsigned short)vs0,  \
                  (__vector unsigned short)vs1);  \
  out = (__vector short)vec_perm((__vector unsigned short)tmpe,  \
                                 (__vector unsigned short)tmpo,  \
                                 shift_pack_index);  \
 }
 void
 jsimd_quantize_altivec (JCOEFPTR coef_block, DCTELEM * divisors,
                        DCTELEM * workspace)
 {
  __vector short row0, row1, row2, row3, row4, row5, row6, row7;
  __vector short row0s, row1s, row2s, row3s, row4s, row5s, row6s, row7s;
  __vector short corr0, corr1, corr2, corr3, corr4, corr5, corr6, corr7;
  __vector short recip0, recip1, recip2, recip3, recip4, recip5, recip6,
    recip7;
  __vector short scale0, scale1, scale2, scale3, scale4, scale5, scale6,
    scale7;
  __vector unsigned int tmpe, tmpo;
  /* Constants */
  __vector unsigned short pw_word_bit_m1 = { __8X(WORD_BIT - 1) };
  __vector unsigned char shift_pack_index =
      { 0, 1, 16, 17, 4, 5, 20, 21, 8, 9, 24, 25, 12, 13, 28, 29};
  row0 = vec_ld(0, workspace);
  row1 = vec_ld(16, workspace);
  row2 = vec_ld(32, workspace);
  row3 = vec_ld(48, workspace);
  row4 = vec_ld(64, workspace);
  row5 = vec_ld(80, workspace);
  row6 = vec_ld(96, workspace);
  row7 = vec_ld(112, workspace);
  /* Branch-less absolute value */
  row0s = vec_sra(row0, pw_word_bit_m1);
  row1s = vec_sra(row1, pw_word_bit_m1);
  row2s = vec_sra(row2, pw_word_bit_m1);
  row3s = vec_sra(row3, pw_word_bit_m1);
  row4s = vec_sra(row4, pw_word_bit_m1);
  row5s = vec_sra(row5, pw_word_bit_m1);
  row6s = vec_sra(row6, pw_word_bit_m1);
  row7s = vec_sra(row7, pw_word_bit_m1);
  row0 = vec_xor(row0, row0s);
  row1 = vec_xor(row1, row1s);
  row2 = vec_xor(row2, row2s);
  row3 = vec_xor(row3, row3s);
  row4 = vec_xor(row4, row4s);
  row5 = vec_xor(row5, row5s);
  row6 = vec_xor(row6, row6s);
  row7 = vec_xor(row7, row7s);
  row0 = vec_sub(row0, row0s);
  row1 = vec_sub(row1, row1s);
  row2 = vec_sub(row2, row2s);
  row3 = vec_sub(row3, row3s);
  row4 = vec_sub(row4, row4s);
  row5 = vec_sub(row5, row5s);
  row6 = vec_sub(row6, row6s);
  row7 = vec_sub(row7, row7s);
  corr0 = vec_ld(DCTSIZE2 * 2, divisors);
  corr1 = vec_ld(DCTSIZE2 * 2 + 16, divisors);
  corr2 = vec_ld(DCTSIZE2 * 2 + 32, divisors);
  corr3 = vec_ld(DCTSIZE2 * 2 + 48, divisors);
  corr4 = vec_ld(DCTSIZE2 * 2 + 64, divisors);
  corr5 = vec_ld(DCTSIZE2 * 2 + 80, divisors);
  corr6 = vec_ld(DCTSIZE2 * 2 + 96, divisors);
  corr7 = vec_ld(DCTSIZE2 * 2 + 112, divisors);
  row0 = vec_add(row0, corr0);
  row1 = vec_add(row1, corr1);
  row2 = vec_add(row2, corr2);
  row3 = vec_add(row3, corr3);
  row4 = vec_add(row4, corr4);
  row5 = vec_add(row5, corr5);
  row6 = vec_add(row6, corr6);
  row7 = vec_add(row7, corr7);
  recip0 = vec_ld(0, divisors);
  recip1 = vec_ld(16, divisors);
  recip2 = vec_ld(32, divisors);
  recip3 = vec_ld(48, divisors);
  recip4 = vec_ld(64, divisors);
  recip5 = vec_ld(80, divisors);
  recip6 = vec_ld(96, divisors);
  recip7 = vec_ld(112, divisors);
  MULTIPLY(row0, recip0, row0);
  MULTIPLY(row1, recip1, row1);
  MULTIPLY(row2, recip2, row2);
  MULTIPLY(row3, recip3, row3);
  MULTIPLY(row4, recip4, row4);
  MULTIPLY(row5, recip5, row5);
  MULTIPLY(row6, recip6, row6);
  MULTIPLY(row7, recip7, row7);
  scale0 = vec_ld(DCTSIZE2 * 4, divisors);
  scale1 = vec_ld(DCTSIZE2 * 4 + 16, divisors);
  scale2 = vec_ld(DCTSIZE2 * 4 + 32, divisors);
  scale3 = vec_ld(DCTSIZE2 * 4 + 48, divisors);
  scale4 = vec_ld(DCTSIZE2 * 4 + 64, divisors);
  scale5 = vec_ld(DCTSIZE2 * 4 + 80, divisors);
  scale6 = vec_ld(DCTSIZE2 * 4 + 96, divisors);
  scale7 = vec_ld(DCTSIZE2 * 4 + 112, divisors);
  MULTIPLY(row0, scale0, row0);
  MULTIPLY(row1, scale1, row1);
  MULTIPLY(row2, scale2, row2);
  MULTIPLY(row3, scale3, row3);
  MULTIPLY(row4, scale4, row4);
  MULTIPLY(row5, scale5, row5);
  MULTIPLY(row6, scale6, row6);
  MULTIPLY(row7, scale7, row7);
  row0 = vec_xor(row0, row0s);
  row1 = vec_xor(row1, row1s);
  row2 = vec_xor(row2, row2s);
  row3 = vec_xor(row3, row3s);
  row4 = vec_xor(row4, row4s);
  row5 = vec_xor(row5, row5s);
  row6 = vec_xor(row6, row6s);
  row7 = vec_xor(row7, row7s);
  row0 = vec_sub(row0, row0s);
  row1 = vec_sub(row1, row1s);
  row2 = vec_sub(row2, row2s);
  row3 = vec_sub(row3, row3s);
  row4 = vec_sub(row4, row4s);
  row5 = vec_sub(row5, row5s);
  row6 = vec_sub(row6, row6s);
  row7 = vec_sub(row7, row7s);
  vec_st(row0, 0, coef_block);
  vec_st(row1, 16, coef_block);
  vec_st(row2, 32, coef_block);
  vec_st(row3, 48, coef_block);
  vec_st(row4, 64, coef_block);
  vec_st(row5, 80, coef_block);
  vec_st(row6, 96, coef_block);
  vec_st(row7, 112, coef_block);
 }
--- a/simd/jsimd.h
+++ b/simd/jsimd.h
@@ -116,6 +116,28 @@ EXTERN(void) jsimd_extxrgb_ycc_convert_mips_dspr2
        (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
         JDIMENSION output_row, int num_rows);
 EXTERN(void) jsimd_rgb_ycc_convert_altivec
        (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
         JDIMENSION output_row, int num_rows);
 EXTERN(void) jsimd_extrgb_ycc_convert_altivec
        (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
         JDIMENSION output_row, int num_rows);
 EXTERN(void) jsimd_extrgbx_ycc_convert_altivec
        (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
         JDIMENSION output_row, int num_rows);
 EXTERN(void) jsimd_extbgr_ycc_convert_altivec
        (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
         JDIMENSION output_row, int num_rows);
 EXTERN(void) jsimd_extbgrx_ycc_convert_altivec
        (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
         JDIMENSION output_row, int num_rows);
 EXTERN(void) jsimd_extxbgr_ycc_convert_altivec
        (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
         JDIMENSION output_row, int num_rows);
 EXTERN(void) jsimd_extxrgb_ycc_convert_altivec
        (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
         JDIMENSION output_row, int num_rows);
 /* RGB & extended RGB --> Grayscale Colorspace Conversion */
 EXTERN(void) jsimd_rgb_gray_convert_mmx
        (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
@@ -184,6 +206,28 @@ EXTERN(void) jsimd_extxrgb_gray_convert_mips_dspr2
        (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
         JDIMENSION output_row, int num_rows);
 EXTERN(void) jsimd_rgb_gray_convert_altivec
        (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
         JDIMENSION output_row, int num_rows);
 EXTERN(void) jsimd_extrgb_gray_convert_altivec
        (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
         JDIMENSION output_row, int num_rows);
 EXTERN(void) jsimd_extrgbx_gray_convert_altivec
        (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
         JDIMENSION output_row, int num_rows);
 EXTERN(void) jsimd_extbgr_gray_convert_altivec
        (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
         JDIMENSION output_row, int num_rows);
 EXTERN(void) jsimd_extbgrx_gray_convert_altivec
        (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
         JDIMENSION output_row, int num_rows);
 EXTERN(void) jsimd_extxbgr_gray_convert_altivec
        (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
         JDIMENSION output_row, int num_rows);
 EXTERN(void) jsimd_extxrgb_gray_convert_altivec
        (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
         JDIMENSION output_row, int num_rows);
 /* YCC --> RGB & extended RGB Colorspace Conversion */
 EXTERN(void) jsimd_ycc_rgb_convert_mmx
        (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row,
@@ -524,6 +568,9 @@ EXTERN(void) jsimd_convsamp_neon
 EXTERN(void) jsimd_convsamp_mips_dspr2
        (JSAMPARRAY sample_data, JDIMENSION start_col, DCTELEM * workspace);
 EXTERN(void) jsimd_convsamp_altivec
        (JSAMPARRAY sample_data, JDIMENSION start_col, DCTELEM * workspace);
 /* Floating Point Sample Conversion */
 EXTERN(void) jsimd_convsamp_float_3dnow
        (JSAMPARRAY sample_data, JDIMENSION start_col, FAST_FLOAT * workspace);
@@ -545,6 +592,8 @@ EXTERN(void) jsimd_fdct_islow_sse2 (DCTELEM * data);
 EXTERN(void) jsimd_fdct_islow_mips_dspr2 (DCTELEM * data);
 EXTERN(void) jsimd_fdct_islow_altivec (DCTELEM * data);
 /* Fast Integer Forward DCT */
 EXTERN(void) jsimd_fdct_ifast_mmx (DCTELEM * data);
@@ -576,6 +625,9 @@ EXTERN(void) jsimd_quantize_neon
 EXTERN(void) jsimd_quantize_mips_dspr2
        (JCOEFPTR coef_block, DCTELEM * divisors, DCTELEM * workspace);
 EXTERN(void) jsimd_quantize_altivec
        (JCOEFPTR coef_block, DCTELEM * divisors, DCTELEM * workspace);
 /* Floating Point Quantization */
 EXTERN(void) jsimd_quantize_float_3dnow
        (JCOEFPTR coef_block, FAST_FLOAT * divisors, FAST_FLOAT * workspace);
@@ -644,6 +696,10 @@ EXTERN(void) jsimd_idct_islow_mips_dspr2
        (void * dct_table, JCOEFPTR coef_block, int * output_buf,
         JSAMPLE * output_col);
 EXTERN(void) jsimd_idct_islow_altivec
        (void * dct_table, JCOEFPTR coef_block, JSAMPARRAY output_buf,
         JDIMENSION output_col);
 /* Fast Integer Inverse DCT */
 EXTERN(void) jsimd_idct_ifast_mmx
        (void * dct_table, JCOEFPTR coef_block, JSAMPARRAY output_buf,
@@ -665,6 +721,10 @@ EXTERN(void) jsimd_idct_ifast_rows_mips_dspr2
        (DCTELEM * wsptr, JSAMPARRAY output_buf, JDIMENSION output_col,
         const int * idct_coefs);
 EXTERN(void) jsimd_idct_ifast_altivec
        (void * dct_table, JCOEFPTR coef_block, JSAMPARRAY output_buf,
         JDIMENSION output_col);
 /* Floating Point Inverse DCT */
 EXTERN(void) jsimd_idct_float_3dnow
        (void * dct_table, JCOEFPTR coef_block, JSAMPARRAY output_buf,
--- a/simd/jsimd_powerpc_altivec.c
+++ b/simd/jsimd_powerpc_altivec.c
@@ -29,6 +29,14 @@
 #include "jsimd.h"
 #include <altivec.h>
 /* Common code */
 #define __4X(a) a, a, a, a
 #define __4X2(a, b) a, b, a, b, a, b, a, b
 #define __8X(a) __4X(a), __4X(a)
 #define __16X(a) __8X(a), __8X(a)
 #define TRANSPOSE(row, col)  \
 {  \
  __vector short row04l, row04h, row15l, row15h,  \
@@ -47,7 +55,7 @@
  row37h = vec_mergel(row##3, row##7); /* row37h=(34 74 35 75 36 76 37 77) */ \
  \
                                       /* transpose coefficients (phase 2) */ \
-  col01e = vec_mergeh(row04l, row26l); /* col01e=(00 20 40 60 01 21 41 61} */ \
+  col01e = vec_mergeh(row04l, row26l); /* col01e=(00 20 40 60 01 21 41 61) */ \
  col23e = vec_mergel(row04l, row26l); /* col23e=(02 22 42 62 03 23 43 63) */ \
  col45e = vec_mergeh(row04h, row26h); /* col45e=(04 24 44 64 05 25 45 65) */ \
  col67e = vec_mergel(row04h, row26h); /* col67e=(06 26 46 66 07 27 47 67) */ \
@@ -58,7 +66,7 @@
  \
                                       /* transpose coefficients (phase 3) */ \
  col##0 = vec_mergeh(col01e, col01o); /* col0=(00 10 20 30 40 50 60 70) */   \
-  col##1 = vec_mergel(col01e, col01o); /* col1=(01 11 21 31 41 51 61 71} */   \
+  col##1 = vec_mergel(col01e, col01o); /* col1=(01 11 21 31 41 51 61 71) */   \
  col##2 = vec_mergeh(col23e, col23o); /* col2=(02 12 22 32 42 52 62 72) */   \
  col##3 = vec_mergel(col23e, col23o); /* col3=(03 13 23 33 43 53 63 73) */   \
  col##4 = vec_mergeh(col45e, col45o); /* col4=(04 14 24 34 44 54 64 74) */   \
@@ -66,125 +74,3 @@
  col##6 = vec_mergeh(col67e, col67o); /* col6=(06 16 26 36 46 56 66 76) */   \
  col##7 = vec_mergel(col67e, col67o); /* col7=(07 17 27 37 47 57 67 77) */   \
 }
 static const __vector short constants __attribute__((aligned(16))) =
 {
  98 << 5,   /* FIX(0.382683433) */
  139 << 5,  /* FIX(0.541196100) */
  181 << 5,  /* FIX(0.707106781) */
  334 << 5   /* FIX(1.306562965) */
 };
 #define DO_DCT()  \
 {  \
  /* Even part */  \
  \
  tmp10 = vec_add(tmp0, tmp3);  \
  tmp13 = vec_sub(tmp0, tmp3);  \
  tmp11 = vec_add(tmp1, tmp2);  \
  tmp12 = vec_sub(tmp1, tmp2);  \
  \
  out0  = vec_add(tmp10, tmp11);  \
  out4  = vec_sub(tmp10, tmp11);  \
  \
  z1 = vec_add(tmp12, tmp13);  \
  z1 = vec_sl(z1, PRE_MULTIPLY_SCALE_BITS);  \
  z1 = vec_madds(z1, PW_0707, zero);  \
  \
  out2 = vec_add(tmp13, z1);  \
  out6 = vec_sub(tmp13, z1);  \
  \
  /* Odd part */  \
  \
  tmp10 = vec_add(tmp4, tmp5);  \
  tmp11 = vec_add(tmp5, tmp6);  \
  tmp12 = vec_add(tmp6, tmp7);  \
  \
  tmp10 = vec_sl(tmp10, PRE_MULTIPLY_SCALE_BITS);  \
  tmp12 = vec_sl(tmp12, PRE_MULTIPLY_SCALE_BITS);  \
  z5 = vec_sub(tmp10, tmp12);  \
  z5 = vec_madds(z5, PW_0382, zero);  \
  \
  z2 = vec_madds(tmp10, PW_0541, zero);  \
  z2 = vec_add(z2, z5);  \
  \
  z4 = vec_madds(tmp12, PW_1306, zero);  \
  z4 = vec_add(z4, z5);  \
  \
  tmp11 = vec_sl(tmp11, PRE_MULTIPLY_SCALE_BITS);  \
  z3 = vec_madds(tmp11, PW_0707, zero);  \
  \
  z11 = vec_add(tmp7, z3);  \
  z13 = vec_sub(tmp7, z3);  \
  \
  out5 = vec_add(z13, z2);  \
  out3 = vec_sub(z13, z2);  \
  out1 = vec_add(z11, z4);  \
  out7 = vec_sub(z11, z4);  \
 }
 void
 jsimd_fdct_ifast_altivec (DCTELEM *data)
 {
  __vector short row0, row1, row2, row3, row4, row5, row6, row7,
    col0, col1, col2, col3, col4, col5, col6, col7,
    tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, tmp10, tmp11, tmp12, tmp13,
    z1, z2, z3, z4, z5, z11, z13,
    out0, out1, out2, out3, out4, out5, out6, out7;
  /* Constants */
  __vector short zero = vec_splat_s16(0),
    PW_0382 = vec_splat(constants, 0),
    PW_0541 = vec_splat(constants, 1),
    PW_0707 = vec_splat(constants, 2),
    PW_1306 = vec_splat(constants, 3);
  __vector unsigned short PRE_MULTIPLY_SCALE_BITS = vec_splat_u16(2);
  /* Pass 1: process rows. */
  row0 = *(__vector short *)&data[0];
  row1 = *(__vector short *)&data[8];
  row2 = *(__vector short *)&data[16];
  row3 = *(__vector short *)&data[24];
  row4 = *(__vector short *)&data[32];
  row5 = *(__vector short *)&data[40];
  row6 = *(__vector short *)&data[48];
  row7 = *(__vector short *)&data[56];
  TRANSPOSE(row, col);
  tmp0 = vec_add(col0, col7);
  tmp7 = vec_sub(col0, col7);
  tmp1 = vec_add(col1, col6);
  tmp6 = vec_sub(col1, col6);
  tmp2 = vec_add(col2, col5);
  tmp5 = vec_sub(col2, col5);
  tmp3 = vec_add(col3, col4);
  tmp4 = vec_sub(col3, col4);
  DO_DCT();
  /* Pass 2: process columns. */
  TRANSPOSE(out, row);
  tmp0 = vec_add(row0, row7);
  tmp7 = vec_sub(row0, row7);
  tmp1 = vec_add(row1, row6);
  tmp6 = vec_sub(row1, row6);
  tmp2 = vec_add(row2, row5);
  tmp5 = vec_sub(row2, row5);
  tmp3 = vec_add(row3, row4);
  tmp4 = vec_sub(row3, row4);
  DO_DCT();
  *(__vector short *)&data[0] = out0;
  *(__vector short *)&data[8] = out1;
  *(__vector short *)&data[16] = out2;
  *(__vector short *)&data[24] = out3;
  *(__vector short *)&data[32] = out4;
  *(__vector short *)&data[40] = out5;
  *(__vector short *)&data[48] = out6;
  *(__vector short *)&data[56] = out7;
 }
--- a/simd/jsimd_arm64_neon.S
+++ b/simd/jsimd_arm64_neon.S
@@ -6,6 +6,7 @@
 * Author: Siarhei Siamashka <siarhei.siamashka@nokia.com>
 * Copyright (C) 2013-2014, Linaro Limited
 * Author: Ragesh Radhakrishnan <ragesh.r@linaro.org>
 * Copyright (C) 2014, D. R. Commander.  All rights reserved.
 *
 * This software is provided 'as-is', without any express or implied
 * warranty.  In no event will the authors be held liable for any damages
@@ -197,21 +198,21 @@ _\fname:
    tmp13 = q1;                                                               \
 }
-#define XFIX_0_899976223                    v0.4h[0]
+#define XFIX_0_899976223                    v0.h[0]
-#define XFIX_0_541196100                    v0.4h[1]
+#define XFIX_0_541196100                    v0.h[1]
-#define XFIX_2_562915447                    v0.4h[2]
+#define XFIX_2_562915447                    v0.h[2]
-#define XFIX_0_298631336_MINUS_0_899976223  v0.4h[3]
+#define XFIX_0_298631336_MINUS_0_899976223  v0.h[3]
-#define XFIX_1_501321110_MINUS_0_899976223  v1.4h[0]
+#define XFIX_1_501321110_MINUS_0_899976223  v1.h[0]
-#define XFIX_2_053119869_MINUS_2_562915447  v1.4h[1]
+#define XFIX_2_053119869_MINUS_2_562915447  v1.h[1]
-#define XFIX_0_541196100_PLUS_0_765366865   v1.4h[2]
+#define XFIX_0_541196100_PLUS_0_765366865   v1.h[2]
-#define XFIX_1_175875602                    v1.4h[3]
+#define XFIX_1_175875602                    v1.h[3]
-#define XFIX_1_175875602_MINUS_0_390180644  v2.4h[0]
+#define XFIX_1_175875602_MINUS_0_390180644  v2.h[0]
-#define XFIX_0_541196100_MINUS_1_847759065  v2.4h[1]
+#define XFIX_0_541196100_MINUS_1_847759065  v2.h[1]
-#define XFIX_3_072711026_MINUS_2_562915447  v2.4h[2]
+#define XFIX_3_072711026_MINUS_2_562915447  v2.h[2]
-#define XFIX_1_175875602_MINUS_1_961570560  v2.4h[3]
+#define XFIX_1_175875602_MINUS_1_961570560  v2.h[3]
 .balign 16
-jsimd_idct_islow_neon_consts:
+Ljsimd_idct_islow_neon_consts:
    .short FIX_0_899976223                    /* d0[0] */
    .short FIX_0_541196100                    /* d0[1] */
    .short FIX_2_562915447                    /* d0[2] */
@@ -256,54 +257,54 @@ asm_function jsimd_idct_islow_neon
    /* Save all NEON registers and x15 (32 NEON registers * 8 bytes + 16) */
    sub             sp, sp, 272
    str             x15, [sp], 16
-    adr             x15, jsimd_idct_islow_neon_consts
+    adr             x15, Ljsimd_idct_islow_neon_consts
-    st1             {v0.8b - v3.8b}, [sp], 32
+    st1             {v0.8b, v1.8b, v2.8b, v3.8b}, [sp], 32
-    st1             {v4.8b - v7.8b}, [sp], 32
+    st1             {v4.8b, v5.8b, v6.8b, v7.8b}, [sp], 32
-    st1             {v8.8b - v11.8b}, [sp], 32
+    st1             {v8.8b, v9.8b, v10.8b, v11.8b}, [sp], 32
-    st1             {v12.8b - v15.8b}, [sp], 32
+    st1             {v12.8b, v13.8b, v14.8b, v15.8b}, [sp], 32
-    st1             {v16.8b - v19.8b}, [sp], 32
+    st1             {v16.8b, v17.8b, v18.8b, v19.8b}, [sp], 32
-    st1             {v20.8b - v23.8b}, [sp], 32
+    st1             {v20.8b, v21.8b, v22.8b, v23.8b}, [sp], 32
-    st1             {v24.8b - v27.8b}, [sp], 32
+    st1             {v24.8b, v25.8b, v26.8b, v27.8b}, [sp], 32
-    st1             {v28.8b - v31.8b}, [sp], 32
+    st1             {v28.8b, v29.8b, v30.8b, v31.8b}, [sp], 32
    ld1             {v16.4h, v17.4h, v18.4h, v19.4h}, [COEF_BLOCK], 32
    ld1             {v0.4h, v1.4h, v2.4h, v3.4h}, [DCT_TABLE], 32
    ld1             {v20.4h, v21.4h, v22.4h, v23.4h}, [COEF_BLOCK], 32
    mul             v16.4h, v16.4h, v0.4h
    mul             v17.4h, v17.4h, v1.4h
-    ins             v16.2d[1], v17.2d[0]  /* 128 bit q8 */
+    ins             v16.d[1], v17.d[0]  /* 128 bit q8 */
    ld1             {v4.4h, v5.4h, v6.4h, v7.4h}, [DCT_TABLE], 32
    mul             v18.4h, v18.4h, v2.4h
    mul             v19.4h, v19.4h, v3.4h
-    ins             v18.2d[1], v19.2d[0]  /* 128 bit q9 */
+    ins             v18.d[1], v19.d[0]  /* 128 bit q9 */
    ld1             {v24.4h, v25.4h, v26.4h, v27.4h}, [COEF_BLOCK], 32
    mul             v20.4h, v20.4h, v4.4h
    mul             v21.4h, v21.4h, v5.4h
-    ins             v20.2d[1], v21.2d[0]  /* 128 bit q10 */
+    ins             v20.d[1], v21.d[0]  /* 128 bit q10 */
    ld1             {v0.4h, v1.4h, v2.4h, v3.4h}, [DCT_TABLE], 32
    mul             v22.4h, v22.4h, v6.4h
    mul             v23.4h, v23.4h, v7.4h
-    ins             v22.2d[1], v23.2d[0]  /* 128 bit q11 */
+    ins             v22.d[1], v23.d[0]  /* 128 bit q11 */
    ld1             {v28.4h, v29.4h, v30.4h, v31.4h}, [COEF_BLOCK]
    mul             v24.4h, v24.4h, v0.4h
    mul             v25.4h, v25.4h, v1.4h
-    ins             v24.2d[1], v25.2d[0]  /* 128 bit q12 */
+    ins             v24.d[1], v25.d[0]  /* 128 bit q12 */
    ld1             {v4.4h, v5.4h, v6.4h, v7.4h}, [DCT_TABLE], 32
    mul             v28.4h, v28.4h, v4.4h
    mul             v29.4h, v29.4h, v5.4h
-    ins             v28.2d[1], v29.2d[0]  /* 128 bit q14 */
+    ins             v28.d[1], v29.d[0]  /* 128 bit q14 */
    mul             v26.4h, v26.4h, v2.4h
    mul             v27.4h, v27.4h, v3.4h
-    ins             v26.2d[1], v27.2d[0]  /* 128 bit q13 */
+    ins             v26.d[1], v27.d[0]  /* 128 bit q13 */
    ld1             {v0.4h, v1.4h, v2.4h, v3.4h}, [x15]  /* load constants */
    add             x15, x15, #16
    mul             v30.4h, v30.4h, v6.4h
    mul             v31.4h, v31.4h, v7.4h
-    ins             v30.2d[1], v31.2d[0]  /* 128 bit q15 */
+    ins             v30.d[1], v31.d[0]  /* 128 bit q15 */
    /* Go to the bottom of the stack */
    sub             sp, sp, 352
    stp             x4, x5, [sp], 16
-    st1             {v8.4h - v11.4h}, [sp], 32  /* save NEON registers */
+    st1             {v8.4h, v9.4h, v10.4h, v11.4h}, [sp], 32  /* save NEON registers */
-    st1             {v12.4h - v15.4h}, [sp], 32
+    st1             {v12.4h, v13.4h, v14.4h, v15.4h}, [sp], 32
    /* 1-D IDCT, pass 1, left 4x8 half */
    add             v4.4h,    ROW7L.4h, ROW3L.4h
    add             v5.4h,    ROW5L.4h, ROW1L.4h
@@ -378,7 +379,7 @@ asm_function jsimd_idct_islow_neon
    rshrn           ROW0L.4h, v12.4s,   #11
    rshrn           ROW4L.4h, v6.4s,    #11
-      beq             3f /* Go to do some special handling for the sparse right 4x8 half */
+      b.eq          3f /* Go to do some special handling for the sparse right 4x8 half */
    /* 1-D IDCT, pass 1, right 4x8 half */
    ld1             {v2.4h},  [x15]    /* reload constants */
@@ -553,33 +554,33 @@ asm_function jsimd_idct_islow_neon
    shrn            ROW4R.4h, v6.4s,    #16
 2:  /* Descale to 8-bit and range limit */
-    ins             v16.2d[1], v17.2d[0]
+    ins             v16.d[1], v17.d[0]
-    ins             v18.2d[1], v19.2d[0]
+    ins             v18.d[1], v19.d[0]
-    ins             v20.2d[1], v21.2d[0]
+    ins             v20.d[1], v21.d[0]
-    ins             v22.2d[1], v23.2d[0]
+    ins             v22.d[1], v23.d[0]
    sqrshrn         v16.8b,   v16.8h,   #2
    sqrshrn2        v16.16b,  v18.8h,   #2
    sqrshrn         v18.8b,   v20.8h,   #2
    sqrshrn2        v18.16b,  v22.8h,   #2
    /* vpop            {v8.4h - d15.4h} */ /* restore NEON registers */
-    ld1             {v8.4h - v11.4h}, [sp], 32
+    ld1             {v8.4h, v9.4h, v10.4h, v11.4h}, [sp], 32
-    ld1             {v12.4h - v15.4h}, [sp], 32
+    ld1             {v12.4h, v13.4h, v14.4h, v15.4h}, [sp], 32
-    ins             v24.2d[1], v25.2d[0]
+    ins             v24.d[1], v25.d[0]
    sqrshrn         v20.8b,   v24.8h,   #2
      /* Transpose the final 8-bit samples and do signed->unsigned conversion */
    /* trn1            v16.8h,    v16.8h,  v18.8h */
    transpose       v16, v18, v3, .16b, .8h
-    ins             v26.2d[1], v27.2d[0]
+    ins             v26.d[1], v27.d[0]
-    ins             v28.2d[1], v29.2d[0]
+    ins             v28.d[1], v29.d[0]
-    ins             v30.2d[1], v31.2d[0]
+    ins             v30.d[1], v31.d[0]
    sqrshrn2        v20.16b,  v26.8h,   #2
    sqrshrn         v22.8b,   v28.8h,   #2
    movi            v0.16b,   #(CENTERJSAMPLE)
    sqrshrn2        v22.16b,  v30.8h,   #2
-    transpose_single v16, v17, v3, .2d, .8b
+    transpose_single v16, v17, v3, .d, .8b
-    transpose_single v18, v19, v3, .2d, .8b
+    transpose_single v18, v19, v3, .d, .8b
    add             v16.8b,   v16.8b,   v0.8b
    add             v17.8b,   v17.8b,   v0.8b
    add             v18.8b,   v18.8b,   v0.8b
@@ -590,7 +591,7 @@ asm_function jsimd_idct_islow_neon
    add             TMP1,     TMP1,     OUTPUT_COL
    add             TMP2,     TMP2,     OUTPUT_COL
    st1             {v16.8b}, [TMP1]
-    transpose_single v20, v21, v3, .2d, .8b
+    transpose_single v20, v21, v3, .d, .8b
    st1             {v17.8b}, [TMP2]
    ldp             TMP1,     TMP2,     [OUTPUT_BUF], 16
    add             TMP1,     TMP1,     OUTPUT_COL
@@ -605,7 +606,7 @@ asm_function jsimd_idct_islow_neon
    add             TMP2,     TMP2,     OUTPUT_COL
    add             TMP3,     TMP3,     OUTPUT_COL
    add             TMP4,     TMP4,     OUTPUT_COL
-    transpose_single v22, v23, v3, .2d, .8b
+    transpose_single v22, v23, v3, .d, .8b
    st1             {v20.8b}, [TMP1]
    add             v22.8b,   v22.8b,   v0.8b
    add             v23.8b,   v23.8b,   v0.8b
@@ -613,14 +614,14 @@ asm_function jsimd_idct_islow_neon
    st1             {v22.8b}, [TMP3]
    st1             {v23.8b}, [TMP4]
    ldr             x15, [sp], 16
-    ld1             {v0.8b - v3.8b}, [sp], 32
+    ld1             {v0.8b, v1.8b, v2.8b, v3.8b}, [sp], 32
-    ld1             {v4.8b - v7.8b}, [sp], 32
+    ld1             {v4.8b, v5.8b, v6.8b, v7.8b}, [sp], 32
-    ld1             {v8.8b - v11.8b}, [sp], 32
+    ld1             {v8.8b, v9.8b, v10.8b, v11.8b}, [sp], 32
-    ld1             {v12.8b - v15.8b}, [sp], 32
+    ld1             {v12.8b, v13.8b, v14.8b, v15.8b}, [sp], 32
-    ld1             {v16.8b - v19.8b}, [sp], 32
+    ld1             {v16.8b, v17.8b, v18.8b, v19.8b}, [sp], 32
-    ld1             {v20.8b - v23.8b}, [sp], 32
+    ld1             {v20.8b, v21.8b, v22.8b, v23.8b}, [sp], 32
-    ld1             {v24.8b - v27.8b}, [sp], 32
+    ld1             {v24.8b, v25.8b, v26.8b, v27.8b}, [sp], 32
-    ld1             {v28.8b - v31.8b}, [sp], 32
+    ld1             {v28.8b, v29.8b, v30.8b, v31.8b}, [sp], 32
    blr             x30
 3:  /* Left 4x8 half is done, right 4x8 half contains mostly zeros */
@@ -636,17 +637,17 @@ asm_function jsimd_idct_islow_neon
    transpose       ROW0L, ROW2L, v3, .16b, .2s
    transpose       ROW5L, ROW7L, v3, .16b, .2s
    cmp             x0, #0
-    beq             4f /* Right 4x8 half has all zeros, go to 'sparse' second pass */
+    b.eq            4f /* Right 4x8 half has all zeros, go to 'sparse' second pass */
    /* Only row 0 is non-zero for the right 4x8 half  */
-    dup             ROW1R.4h, ROW0R.4h[1]
+    dup             ROW1R.4h, ROW0R.h[1]
-    dup             ROW2R.4h, ROW0R.4h[2]
+    dup             ROW2R.4h, ROW0R.h[2]
-    dup             ROW3R.4h, ROW0R.4h[3]
+    dup             ROW3R.4h, ROW0R.h[3]
-    dup             ROW4R.4h, ROW0R.4h[0]
+    dup             ROW4R.4h, ROW0R.h[0]
-    dup             ROW5R.4h, ROW0R.4h[1]
+    dup             ROW5R.4h, ROW0R.h[1]
-    dup             ROW6R.4h, ROW0R.4h[2]
+    dup             ROW6R.4h, ROW0R.h[2]
-    dup             ROW7R.4h, ROW0R.4h[3]
+    dup             ROW7R.4h, ROW0R.h[3]
-    dup             ROW0R.4h, ROW0R.4h[0]
+    dup             ROW0R.4h, ROW0R.h[0]
    b               1b /* Go to 'normal' second pass */
 4:  /* 1-D IDCT, pass 2 (sparse variant with zero rows 4-7), left 4x8 half */
@@ -770,13 +771,13 @@ asm_function jsimd_idct_islow_neon
 * per 1-D IDCT pass, totalling to 5 VQDMULH and 35 VADD/VSUB instructions.
 */
-#define XFIX_1_082392200 v0.4h[0]
+#define XFIX_1_082392200 v0.h[0]
-#define XFIX_1_414213562 v0.4h[1]
+#define XFIX_1_414213562 v0.h[1]
-#define XFIX_1_847759065 v0.4h[2]
+#define XFIX_1_847759065 v0.h[2]
-#define XFIX_2_613125930 v0.4h[3]
+#define XFIX_2_613125930 v0.h[3]
 .balign 16
-jsimd_idct_ifast_neon_consts:
+Ljsimd_idct_ifast_neon_consts:
    .short (277 * 128 - 256 * 128) /* XFIX_1_082392200 */
    .short (362 * 128 - 256 * 128) /* XFIX_1_414213562 */
    .short (473 * 128 - 256 * 128) /* XFIX_1_847759065 */
@@ -810,12 +811,12 @@ asm_function jsimd_idct_ifast_neon
    /* Save NEON registers used in fast IDCT */
    sub             sp, sp, #176
    stp             x22, x23, [sp], 16
-    adr             x23, jsimd_idct_ifast_neon_consts
+    adr             x23, Ljsimd_idct_ifast_neon_consts
-    st1             {v0.8b - v3.8b}, [sp], 32
+    st1             {v0.8b, v1.8b, v2.8b, v3.8b}, [sp], 32
-    st1             {v4.8b - v7.8b}, [sp], 32
+    st1             {v4.8b, v5.8b, v6.8b, v7.8b}, [sp], 32
-    st1             {v8.8b - v11.8b}, [sp], 32
+    st1             {v8.8b, v9.8b, v10.8b, v11.8b}, [sp], 32
-    st1             {v12.8b - v15.8b}, [sp], 32
+    st1             {v12.8b, v13.8b, v14.8b, v15.8b}, [sp], 32
-    st1             {v16.8b - v19.8b}, [sp], 32
+    st1             {v16.8b, v17.8b, v18.8b, v19.8b}, [sp], 32
    ld1             {v8.8h, v9.8h}, [COEF_BLOCK], 32
    ld1             {v0.8h, v1.8h}, [DCT_TABLE], 32
    ld1             {v10.8h, v11.8h}, [COEF_BLOCK], 32
@@ -909,24 +910,24 @@ asm_function jsimd_idct_ifast_neon
    trn2            v15.4s,   v18.4s,   v15.4s
    /* vswp            v14.4h,   v10-MSB.4h */
    umov            x22, v14.d[0]
-    ins             v14.2d[0], v10.2d[1]
+    ins             v14.d[0], v10.d[1]
-    ins             v10.2d[1], x22
+    ins             v10.d[1], x22
    /* vswp            v13.4h,   v9MSB.4h */
    umov            x22, v13.d[0]
-    ins             v13.2d[0], v9.2d[1]
+    ins             v13.d[0], v9.d[1]
-    ins             v9.2d[1], x22
+    ins             v9.d[1], x22
    /* 1-D IDCT, pass 2 */
    sub             v2.8h,    v10.8h,   v14.8h
    /* vswp            v15.4h,   v11MSB.4h */
    umov            x22, v15.d[0]
-    ins             v15.2d[0], v11.2d[1]
+    ins             v15.d[0], v11.d[1]
-    ins             v11.2d[1], x22
+    ins             v11.d[1], x22
    add             v14.8h,   v10.8h,   v14.8h
    /* vswp            v12.4h,   v8-MSB.4h */
    umov            x22, v12.d[0]
-    ins             v12.2d[0], v8.2d[1]
+    ins             v12.d[0], v8.d[1]
-    ins             v8.2d[1], x22
+    ins             v8.d[1],  x22
    sub             v1.8h,    v11.8h,   v13.8h
    add             v13.8h,   v11.8h,   v13.8h
    sub             v5.8h,    v9.8h,    v15.8h
@@ -997,13 +998,13 @@ asm_function jsimd_idct_ifast_neon
    trn1            v9.4s,    v9.4s,    v11.4s
    trn2            v11.4s,   v18.4s,   v11.4s
    /* make copy */
-    ins             v17.2d[0], v8.2d[1]
+    ins             v17.d[0], v8.d[1]
    /* Transpose  d16-d17-msb */
    mov             v18.16b,  v8.16b
    trn1            v8.8b,    v8.8b,    v17.8b
    trn2            v17.8b,   v18.8b,   v17.8b
    /* make copy */
-    ins             v19.2d[0], v9.2d[1]
+    ins             v19.d[0], v9.d[1]
    mov             v18.16b,  v9.16b
    trn1            v9.8b,    v9.8b,    v19.8b
    trn2            v19.8b,   v18.8b,   v19.8b
@@ -1018,7 +1019,7 @@ asm_function jsimd_idct_ifast_neon
    add             TMP2,     TMP2,     OUTPUT_COL
    st1             {v9.8b},  [TMP1]
    /* make copy */
-    ins             v7.2d[0], v10.2d[1]
+    ins             v7.d[0],  v10.d[1]
    mov             v18.16b,  v10.16b
    trn1            v10.8b,   v10.8b,   v7.8b
    trn2            v7.8b,    v18.8b,   v7.8b
@@ -1031,7 +1032,7 @@ asm_function jsimd_idct_ifast_neon
    add             TMP5,     TMP5,     OUTPUT_COL
    st1             {v10.8b}, [TMP1]
    /* make copy */
-    ins             v16.2d[0], v11.2d[1]
+    ins             v16.d[0], v11.d[1]
    mov             v18.16b,  v11.16b
    trn1            v11.8b,   v11.8b,   v16.8b
    trn2            v16.8b,   v18.8b,   v16.8b
@@ -1040,11 +1041,11 @@ asm_function jsimd_idct_ifast_neon
    st1             {v16.8b}, [TMP5]
    sub             sp, sp, #176
    ldp             x22, x23, [sp], 16
-    ld1             {v0.8b - v3.8b}, [sp], 32
+    ld1             {v0.8b, v1.8b, v2.8b, v3.8b}, [sp], 32
-    ld1             {v4.8b - v7.8b}, [sp], 32
+    ld1             {v4.8b, v5.8b, v6.8b, v7.8b}, [sp], 32
-    ld1             {v8.8b - v11.8b}, [sp], 32
+    ld1             {v8.8b, v9.8b, v10.8b, v11.8b}, [sp], 32
-    ld1             {v12.8b - v15.8b}, [sp], 32
+    ld1             {v12.8b, v13.8b, v14.8b, v15.8b}, [sp], 32
-    ld1             {v16.8b - v19.8b}, [sp], 32
+    ld1             {v16.8b, v17.8b, v18.8b, v19.8b}, [sp], 32
    blr             x30
    .unreq          DCT_TABLE
@@ -1095,38 +1096,38 @@ asm_function jsimd_idct_ifast_neon
 #define FIX_3_624509785  (29692) /* FIX(3.624509785) */
 .balign 16
-jsimd_idct_4x4_neon_consts:
+Ljsimd_idct_4x4_neon_consts:
-    .short     FIX_1_847759065     /* v0.4h[0] */
+    .short     FIX_1_847759065     /* v0.h[0] */
-    .short     -FIX_0_765366865    /* v0.4h[1] */
+    .short     -FIX_0_765366865    /* v0.h[1] */
-    .short     -FIX_0_211164243    /* v0.4h[2] */
+    .short     -FIX_0_211164243    /* v0.h[2] */
-    .short     FIX_1_451774981     /* v0.4h[3] */
+    .short     FIX_1_451774981     /* v0.h[3] */
    .short     -FIX_2_172734803    /* d1[0] */
    .short     FIX_1_061594337     /* d1[1] */
    .short     -FIX_0_509795579    /* d1[2] */
    .short     -FIX_0_601344887    /* d1[3] */
-    .short     FIX_0_899976223     /* v2.4h[0] */
+    .short     FIX_0_899976223     /* v2.h[0] */
-    .short     FIX_2_562915447     /* v2.4h[1] */
+    .short     FIX_2_562915447     /* v2.h[1] */
-    .short     1 << (CONST_BITS+1) /* v2.4h[2] */
+    .short     1 << (CONST_BITS+1) /* v2.h[2] */
-    .short     0                   /* v2.4h[3] */
+    .short     0                   /* v2.h[3] */
 .macro idct_helper x4, x6, x8, x10, x12, x14, x16, shift, y26, y27, y28, y29
-    smull           v28.4s, \x4,    v2.4h[2]
+    smull           v28.4s, \x4,    v2.h[2]
-    smlal           v28.4s, \x8,    v0.4h[0]
+    smlal           v28.4s, \x8,    v0.h[0]
-    smlal           v28.4s, \x14,   v0.4h[1]
+    smlal           v28.4s, \x14,   v0.h[1]
-    smull           v26.4s, \x16,   v1.4h[2]
+    smull           v26.4s, \x16,   v1.h[2]
-    smlal           v26.4s, \x12,   v1.4h[3]
+    smlal           v26.4s, \x12,   v1.h[3]
-    smlal           v26.4s, \x10,   v2.4h[0]
+    smlal           v26.4s, \x10,   v2.h[0]
-    smlal           v26.4s, \x6,    v2.4h[1]
+    smlal           v26.4s, \x6,    v2.h[1]
-    smull           v30.4s, \x4,    v2.4h[2]
+    smull           v30.4s, \x4,    v2.h[2]
-    smlsl           v30.4s, \x8,    v0.4h[0]
+    smlsl           v30.4s, \x8,    v0.h[0]
-    smlsl           v30.4s, \x14,   v0.4h[1]
+    smlsl           v30.4s, \x14,   v0.h[1]
-    smull           v24.4s, \x16,   v0.4h[2]
+    smull           v24.4s, \x16,   v0.h[2]
-    smlal           v24.4s, \x12,   v0.4h[3]
+    smlal           v24.4s, \x12,   v0.h[3]
-    smlal           v24.4s, \x10,   v1.4h[0]
+    smlal           v24.4s, \x10,   v1.h[0]
-    smlal           v24.4s, \x6,    v1.4h[1]
+    smlal           v24.4s, \x6,    v1.h[1]
    add             v20.4s, v28.4s, v26.4s
    sub             v28.4s, v28.4s, v26.4s
@@ -1171,15 +1172,15 @@ asm_function jsimd_idct_4x4_neon
    sub             sp, sp, 272
    str             x15, [sp], 16
    /* Load constants (v3.4h is just used for padding) */
-    adr             TMP4, jsimd_idct_4x4_neon_consts
+    adr             TMP4, Ljsimd_idct_4x4_neon_consts
-    st1             {v0.8b - v3.8b}, [sp], 32
+    st1             {v0.8b, v1.8b, v2.8b, v3.8b}, [sp], 32
-    st1             {v4.8b - v7.8b}, [sp], 32
+    st1             {v4.8b, v5.8b, v6.8b, v7.8b}, [sp], 32
-    st1             {v8.8b - v11.8b}, [sp], 32
+    st1             {v8.8b, v9.8b, v10.8b, v11.8b}, [sp], 32
-    st1             {v12.8b - v15.8b}, [sp], 32
+    st1             {v12.8b, v13.8b, v14.8b, v15.8b}, [sp], 32
-    st1             {v16.8b - v19.8b}, [sp], 32
+    st1             {v16.8b, v17.8b, v18.8b, v19.8b}, [sp], 32
-    st1             {v20.8b - v23.8b}, [sp], 32
+    st1             {v20.8b, v21.8b, v22.8b, v23.8b}, [sp], 32
-    st1             {v24.8b - v27.8b}, [sp], 32
+    st1             {v24.8b, v25.8b, v26.8b, v27.8b}, [sp], 32
-    st1             {v28.8b - v31.8b}, [sp], 32
+    st1             {v28.8b, v29.8b, v30.8b, v31.8b}, [sp], 32
    ld1             {v0.4h, v1.4h, v2.4h, v3.4h}, [TMP4]
    /* Load all COEF_BLOCK into NEON registers with the following allocation:
@@ -1203,45 +1204,45 @@ asm_function jsimd_idct_4x4_neon
    ld1             {v18.4h, v19.4h, v20.4h, v21.4h}, [DCT_TABLE], 32
    mul             v4.4h, v4.4h, v18.4h
    mul             v5.4h, v5.4h, v19.4h
-    ins             v4.2d[1], v5.2d[0]    /* 128 bit q4 */
+    ins             v4.d[1], v5.d[0]    /* 128 bit q4 */
    ld1             {v22.4h, v23.4h, v24.4h, v25.4h}, [DCT_TABLE], 32
    mul             v6.4h, v6.4h, v20.4h
    mul             v7.4h, v7.4h, v21.4h
-    ins             v6.2d[1], v7.2d[0]    /* 128 bit q6 */
+    ins             v6.d[1], v7.d[0]    /* 128 bit q6 */
    mul             v8.4h, v8.4h, v22.4h
    mul             v9.4h, v9.4h, v23.4h
-    ins             v8.2d[1], v9.2d[0]    /* 128 bit q8 */
+    ins             v8.d[1], v9.d[0]    /* 128 bit q8 */
    add             DCT_TABLE, DCT_TABLE, #16
    ld1             {v26.4h, v27.4h, v28.4h, v29.4h}, [DCT_TABLE], 32
    mul             v10.4h, v10.4h, v24.4h
    mul             v11.4h, v11.4h, v25.4h
-    ins             v10.2d[1], v11.2d[0]  /* 128 bit q10 */
+    ins             v10.d[1], v11.d[0]  /* 128 bit q10 */
    mul             v12.4h, v12.4h, v26.4h
    mul             v13.4h, v13.4h, v27.4h
-    ins             v12.2d[1], v13.2d[0]  /* 128 bit q12 */
+    ins             v12.d[1], v13.d[0]  /* 128 bit q12 */
    ld1             {v30.4h, v31.4h}, [DCT_TABLE], 16
    mul             v14.4h, v14.4h, v28.4h
    mul             v15.4h, v15.4h, v29.4h
-    ins             v14.2d[1], v15.2d[0]  /* 128 bit q14 */
+    ins             v14.d[1], v15.d[0]  /* 128 bit q14 */
    mul             v16.4h, v16.4h, v30.4h
    mul             v17.4h, v17.4h, v31.4h
-    ins             v16.2d[1], v17.2d[0]  /* 128 bit q16 */
+    ins             v16.d[1], v17.d[0]  /* 128 bit q16 */
    /* Pass 1 */
    idct_helper     v4.4h, v6.4h, v8.4h, v10.4h, v12.4h, v14.4h, v16.4h, 12, v4.4h, v6.4h, v8.4h, v10.4h
    transpose_4x4   v4, v6, v8, v10, v3
-    ins             v10.2d[1], v11.2d[0]
+    ins             v10.d[1], v11.d[0]
    idct_helper     v5.4h, v7.4h, v9.4h, v11.4h, v13.4h, v15.4h, v17.4h, 12, v5.4h, v7.4h, v9.4h, v11.4h
    transpose_4x4   v5, v7, v9, v11, v3
-    ins             v10.2d[1], v11.2d[0]
+    ins             v10.d[1], v11.d[0]
    /* Pass 2 */
    idct_helper     v4.4h, v6.4h, v8.4h, v10.4h, v7.4h, v9.4h, v11.4h, 19, v26.4h, v27.4h, v28.4h, v29.4h
    transpose_4x4   v26, v27, v28, v29, v3
    /* Range limit */
    movi            v30.8h, #0x80
-    ins             v26.2d[1], v27.2d[0]
+    ins             v26.d[1], v27.d[0]
-    ins             v28.2d[1], v29.2d[0]
+    ins             v28.d[1], v29.d[0]
    add             v26.8h, v26.8h, v30.8h
    add             v28.8h, v28.8h, v30.8h
    sqxtun          v26.8b, v26.8h
@@ -1286,14 +1287,14 @@ asm_function jsimd_idct_4x4_neon
    /* vpop            {v8.4h - v15.4h}    ;not available */
    sub             sp, sp, #272
    ldr             x15, [sp], 16
-    ld1             {v0.8b - v3.8b}, [sp], 32
+    ld1             {v0.8b, v1.8b, v2.8b, v3.8b}, [sp], 32
-    ld1             {v4.8b - v7.8b}, [sp], 32
+    ld1             {v4.8b, v5.8b, v6.8b, v7.8b}, [sp], 32
-    ld1             {v8.8b - v11.8b}, [sp], 32
+    ld1             {v8.8b, v9.8b, v10.8b, v11.8b}, [sp], 32
-    ld1             {v12.8b - v15.8b}, [sp], 32
+    ld1             {v12.8b, v13.8b, v14.8b, v15.8b}, [sp], 32
-    ld1             {v16.8b - v19.8b}, [sp], 32
+    ld1             {v16.8b, v17.8b, v18.8b, v19.8b}, [sp], 32
-    ld1             {v20.8b - v23.8b}, [sp], 32
+    ld1             {v20.8b, v21.8b, v22.8b, v23.8b}, [sp], 32
-    ld1             {v24.8b - v27.8b}, [sp], 32
+    ld1             {v24.8b, v25.8b, v26.8b, v27.8b}, [sp], 32
-    ld1             {v28.8b - v31.8b}, [sp], 32
+    ld1             {v28.8b, v29.8b, v30.8b, v31.8b}, [sp], 32
    blr             x30
    .unreq          DCT_TABLE
@@ -1325,7 +1326,7 @@ asm_function jsimd_idct_4x4_neon
 */
 .balign 8
-jsimd_idct_2x2_neon_consts:
+Ljsimd_idct_2x2_neon_consts:
    .short     -FIX_0_720959822    /* v14[0] */
    .short     FIX_0_850430095     /* v14[1] */
    .short     -FIX_1_272758580    /* v14[2] */
@@ -1333,10 +1334,10 @@ jsimd_idct_2x2_neon_consts:
 .macro idct_helper x4, x6, x10, x12, x16, shift, y26, y27
    sshll      v15.4s, \x4,    #15
-    smull      v26.4s, \x6,    v14.4h[3]
+    smull      v26.4s, \x6,    v14.h[3]
-    smlal      v26.4s, \x10,   v14.4h[2]
+    smlal      v26.4s, \x10,   v14.h[2]
-    smlal      v26.4s, \x12,   v14.4h[1]
+    smlal      v26.4s, \x12,   v14.h[1]
-    smlal      v26.4s, \x16,   v14.4h[0]
+    smlal      v26.4s, \x16,   v14.h[0]
    add        v20.4s, v15.4s, v26.4s
    sub        v15.4s, v15.4s, v26.4s
@@ -1367,14 +1368,14 @@ asm_function jsimd_idct_2x2_neon
    str             x15, [sp], 16
    /* Load constants */
-    adr             TMP2, jsimd_idct_2x2_neon_consts
+    adr             TMP2, Ljsimd_idct_2x2_neon_consts
-    st1             {v4.8b - v7.8b}, [sp], 32
+    st1             {v4.8b, v5.8b, v6.8b, v7.8b}, [sp], 32
-    st1             {v8.8b - v11.8b}, [sp], 32
+    st1             {v8.8b, v9.8b, v10.8b, v11.8b}, [sp], 32
-    st1             {v12.8b - v15.8b}, [sp], 32
+    st1             {v12.8b, v13.8b, v14.8b, v15.8b}, [sp], 32
-    st1             {v16.8b - v19.8b}, [sp], 32
+    st1             {v16.8b, v17.8b, v18.8b, v19.8b}, [sp], 32
-    st1             {v21.8b - v22.8b}, [sp], 16
+    st1             {v21.8b, v22.8b}, [sp], 16
-    st1             {v24.8b - v27.8b}, [sp], 32
+    st1             {v24.8b, v25.8b, v26.8b, v27.8b}, [sp], 32
-    st1             {v30.8b - v31.8b}, [sp], 16
+    st1             {v30.8b, v31.8b}, [sp], 16
    ld1             {v14.4h}, [TMP2]
    /* Load all COEF_BLOCK into NEON registers with the following allocation:
@@ -1400,25 +1401,25 @@ asm_function jsimd_idct_2x2_neon
    ld1             {v18.4h, v19.4h, v20.4h, v21.4h}, [DCT_TABLE], 32
    mul             v4.4h, v4.4h, v18.4h
    mul             v5.4h, v5.4h, v19.4h
-    ins             v4.2d[1], v5.2d[0]
+    ins             v4.d[1], v5.d[0]
    mul             v6.4h, v6.4h, v20.4h
    mul             v7.4h, v7.4h, v21.4h
-    ins             v6.2d[1], v7.2d[0]
+    ins             v6.d[1], v7.d[0]
    add             DCT_TABLE, DCT_TABLE, #16
    ld1             {v24.4h, v25.4h}, [DCT_TABLE], 16
    mul             v10.4h, v10.4h, v24.4h
    mul             v11.4h, v11.4h, v25.4h
-    ins             v10.2d[1], v11.2d[0]
+    ins             v10.d[1], v11.d[0]
    add             DCT_TABLE, DCT_TABLE, #16
    ld1             {v26.4h, v27.4h}, [DCT_TABLE], 16
    mul             v12.4h, v12.4h, v26.4h
    mul             v13.4h, v13.4h, v27.4h
-    ins             v12.2d[1], v13.2d[0]
+    ins             v12.d[1], v13.d[0]
    add             DCT_TABLE, DCT_TABLE, #16
    ld1             {v30.4h, v31.4h}, [DCT_TABLE], 16
    mul             v16.4h, v16.4h, v30.4h
    mul             v17.4h, v17.4h, v31.4h
-    ins             v16.2d[1], v17.2d[0]
+    ins             v16.d[1], v17.d[0]
    /* Pass 1 */
 #if 0
@@ -1427,14 +1428,14 @@ asm_function jsimd_idct_2x2_neon
    idct_helper     v5.4h, v7.4h, v11.4h, v13.4h, v17.4h, 13, v5.4h, v7.4h
    transpose_4x4   v5.4h, v7.4h, v9.4h,  v11.4h
 #else
-    smull           v26.4s, v6.4h,  v14.4h[3]
+    smull           v26.4s, v6.4h,  v14.h[3]
-    smlal           v26.4s, v10.4h, v14.4h[2]
+    smlal           v26.4s, v10.4h, v14.h[2]
-    smlal           v26.4s, v12.4h, v14.4h[1]
+    smlal           v26.4s, v12.4h, v14.h[1]
-    smlal           v26.4s, v16.4h, v14.4h[0]
+    smlal           v26.4s, v16.4h, v14.h[0]
-    smull           v24.4s, v7.4h,  v14.4h[3]
+    smull           v24.4s, v7.4h,  v14.h[3]
-    smlal           v24.4s, v11.4h, v14.4h[2]
+    smlal           v24.4s, v11.4h, v14.h[2]
-    smlal           v24.4s, v13.4h, v14.4h[1]
+    smlal           v24.4s, v13.4h, v14.h[1]
-    smlal           v24.4s, v17.4h, v14.4h[0]
+    smlal           v24.4s, v17.4h, v14.h[0]
    sshll           v15.4s, v4.4h,  #15
    sshll           v30.4s, v5.4h,  #15
    add             v20.4s, v15.4s, v26.4s
@@ -1445,12 +1446,12 @@ asm_function jsimd_idct_2x2_neon
    sub             v15.4s, v30.4s, v24.4s
    rshrn           v5.4h,  v20.4s, #13
    rshrn           v7.4h,  v15.4s, #13
-    ins             v4.2d[1], v5.2d[0]
+    ins             v4.d[1], v5.d[0]
-    ins             v6.2d[1], v7.2d[0]
+    ins             v6.d[1], v7.d[0]
    transpose       v4, v6, v3, .16b, .8h
    transpose       v6, v10, v3, .16b, .4s
-    ins             v11.2d[0], v10.2d[1]
+    ins             v11.d[0], v10.d[1]
-    ins             v7.2d[0], v6.2d[1]
+    ins             v7.d[0], v6.d[1]
 #endif
    /* Pass 2 */
@@ -1458,10 +1459,10 @@ asm_function jsimd_idct_2x2_neon
    /* Range limit */
    movi            v30.8h, #0x80
-    ins             v26.2d[1], v27.2d[0]
+    ins             v26.d[1], v27.d[0]
    add             v26.8h, v26.8h, v30.8h
    sqxtun          v30.8b, v26.8h
-    ins             v26.2d[0], v30.2d[0]
+    ins             v26.d[0], v30.d[0]
    sqxtun          v27.8b, v26.8h
    /* Store results to the output buffer */
@@ -1476,13 +1477,13 @@ asm_function jsimd_idct_2x2_neon
    sub             sp, sp, #208
    ldr             x15, [sp], 16
-    ld1             {v4.8b - v7.8b}, [sp], 32
+    ld1             {v4.8b, v5.8b, v6.8b, v7.8b}, [sp], 32
-    ld1             {v8.8b - v11.8b}, [sp], 32
+    ld1             {v8.8b, v9.8b, v10.8b, v11.8b}, [sp], 32
-    ld1             {v12.8b - v15.8b}, [sp], 32
+    ld1             {v12.8b, v13.8b, v14.8b, v15.8b}, [sp], 32
-    ld1             {v16.8b - v19.8b}, [sp], 32
+    ld1             {v16.8b, v17.8b, v18.8b, v19.8b}, [sp], 32
-    ld1             {v21.8b - v22.8b}, [sp], 16
+    ld1             {v21.8b, v22.8b}, [sp], 16
-    ld1             {v24.8b - v27.8b}, [sp], 32
+    ld1             {v24.8b, v25.8b, v26.8b, v27.8b}, [sp], 32
-    ld1             {v30.8b - v31.8b}, [sp], 16
+    ld1             {v30.8b, v31.8b}, [sp], 16
    blr             x30
    .unreq          DCT_TABLE
@@ -1514,9 +1515,9 @@ asm_function jsimd_idct_2x2_neon
        ld1  {v4.8b}, [U], 8
        ld1  {v5.8b}, [V], 8
        ld1  {v0.8b}, [Y], 8
-        prfm PLDL1KEEP, [U, #64]
+        prfm pldl1keep, [U, #64]
-        prfm PLDL1KEEP, [V, #64]
+        prfm pldl1keep, [V, #64]
-        prfm PLDL1KEEP, [Y, #64]
+        prfm pldl1keep, [Y, #64]
    .elseif \size == 4
        ld1  {v4.b}[0], [U], 1
        ld1  {v4.b}[1], [U], 1
@@ -1606,14 +1607,14 @@ asm_function jsimd_idct_2x2_neon
 .macro do_yuv_to_rgb_stage1
    uaddw        v6.8h, v2.8h, v4.8b     /* q3 = u - 128 */
    uaddw        v8.8h, v2.8h, v5.8b     /* q2 = v - 128 */
-    smull        v20.4s, v6.4h, v1.4h[1] /* multiply by -11277 */
+    smull        v20.4s, v6.4h, v1.h[1]  /* multiply by -11277 */
-    smlal        v20.4s, v8.4h, v1.4h[2] /* multiply by -23401 */
+    smlal        v20.4s, v8.4h, v1.h[2]  /* multiply by -23401 */
-    smull2       v22.4s, v6.8h, v1.4h[1] /* multiply by -11277 */
+    smull2       v22.4s, v6.8h, v1.h[1]  /* multiply by -11277 */
-    smlal2       v22.4s, v8.8h, v1.4h[2] /* multiply by -23401 */
+    smlal2       v22.4s, v8.8h, v1.h[2]  /* multiply by -23401 */
-    smull        v24.4s, v8.4h, v1.4h[0] /* multiply by 22971 */
+    smull        v24.4s, v8.4h, v1.h[0]  /* multiply by 22971 */
-    smull2       v26.4s, v8.8h, v1.4h[0] /* multiply by 22971 */
+    smull2       v26.4s, v8.8h, v1.h[0]  /* multiply by 22971 */
-    smull        v28.4s, v6.4h, v1.4h[3] /* multiply by 29033 */
+    smull        v28.4s, v6.4h, v1.h[3]  /* multiply by 29033 */
-    smull2       v30.4s, v6.8h, v1.4h[3] /* multiply by 29033 */
+    smull2       v30.4s, v6.8h, v1.h[3]  /* multiply by 29033 */
 .endm
 .macro do_yuv_to_rgb_stage2
@@ -1656,18 +1657,18 @@ asm_function jsimd_idct_2x2_neon
    sqxtun       v1\g_offs\defsize, v20.8h
    ld1          {v0.8b}, [Y], 8
    sqxtun       v1\r_offs\defsize, v24.8h
-    prfm         PLDL1KEEP, [U, #64]
+    prfm         pldl1keep, [U, #64]
-    prfm         PLDL1KEEP, [V, #64]
+    prfm         pldl1keep, [V, #64]
-    prfm         PLDL1KEEP, [Y, #64]
+    prfm         pldl1keep, [Y, #64]
    sqxtun       v1\b_offs\defsize, v28.8h
    uaddw        v6.8h, v2.8h, v4.8b     /* v6.16b = u - 128 */
    uaddw        v8.8h, v2.8h, v5.8b     /* q2 = v - 128 */
-    smull        v20.4s, v6.4h, v1.4h[1] /* multiply by -11277 */
+    smull        v20.4s, v6.4h, v1.h[1]  /* multiply by -11277 */
-    smlal        v20.4s, v8.4h, v1.4h[2] /* multiply by -23401 */
+    smlal        v20.4s, v8.4h, v1.h[2]  /* multiply by -23401 */
-    smull2       v22.4s, v6.8h, v1.4h[1] /* multiply by -11277 */
+    smull2       v22.4s, v6.8h, v1.h[1]  /* multiply by -11277 */
-    smlal2       v22.4s, v8.8h, v1.4h[2] /* multiply by -23401 */
+    smlal2       v22.4s, v8.8h, v1.h[2]  /* multiply by -23401 */
-    smull        v24.4s, v8.4h, v1.4h[0] /* multiply by 22971 */
+    smull        v24.4s, v8.4h, v1.h[0]  /* multiply by 22971 */
-    smull2       v26.4s, v8.8h, v1.4h[0] /* multiply by 22971 */
+    smull2       v26.4s, v8.8h, v1.h[0]  /* multiply by 22971 */
 .else /**************************** rgb565 ***********************************/
    sqshlu       v21.8h, v20.8h, #8
    sqshlu       v25.8h, v24.8h, #8
@@ -1675,21 +1676,21 @@ asm_function jsimd_idct_2x2_neon
    uaddw        v6.8h, v2.8h, v4.8b     /* v6.16b = u - 128 */
    uaddw        v8.8h, v2.8h, v5.8b     /* q2 = v - 128 */
    ld1          {v0.8b}, [Y], 8
-    smull        v20.4s, v6.4h, v1.4h[1] /* multiply by -11277 */
+    smull        v20.4s, v6.4h, v1.h[1]  /* multiply by -11277 */
-    smlal        v20.4s, v8.4h, v1.4h[2] /* multiply by -23401 */
+    smlal        v20.4s, v8.4h, v1.h[2]  /* multiply by -23401 */
-    smull2       v22.4s, v6.8h, v1.4h[1] /* multiply by -11277 */
+    smull2       v22.4s, v6.8h, v1.h[1]  /* multiply by -11277 */
-    smlal2       v22.4s, v8.8h, v1.4h[2] /* multiply by -23401 */
+    smlal2       v22.4s, v8.8h, v1.h[2]  /* multiply by -23401 */
    sri          v25.8h, v21.8h, #5
-    smull        v24.4s, v8.4h, v1.4h[0] /* multiply by 22971 */
+    smull        v24.4s, v8.4h, v1.h[0]  /* multiply by 22971 */
-    smull2       v26.4s, v8.8h, v1.4h[0] /* multiply by 22971 */
+    smull2       v26.4s, v8.8h, v1.h[0]  /* multiply by 22971 */
-    prfm         PLDL1KEEP, [U, #64]
+    prfm         pldl1keep, [U, #64]
-    prfm         PLDL1KEEP, [V, #64]
+    prfm         pldl1keep, [V, #64]
-    prfm         PLDL1KEEP, [Y, #64]
+    prfm         pldl1keep, [Y, #64]
    sri          v25.8h, v29.8h, #11
 .endif
    do_store     \bpp, 8
-    smull        v28.4s, v6.4h, v1.4h[3] /* multiply by 29033 */
+    smull        v28.4s, v6.4h, v1.h[3]  /* multiply by 29033 */
-    smull2       v30.4s, v6.8h, v1.4h[3] /* multiply by 29033 */
+    smull2       v30.4s, v6.8h, v1.h[3]  /* multiply by 29033 */
 .endm
 .macro do_yuv_to_rgb
@@ -1702,7 +1703,7 @@ asm_function jsimd_idct_2x2_neon
 */
 .balign 16
-jsimd_ycc_\colorid\()_neon_consts:
+Ljsimd_ycc_\colorid\()_neon_consts:
    .short          0,      0,     0,      0
    .short          22971, -11277, -23401, 29033
    .short          -128,  -128,   -128,   -128
@@ -1717,7 +1718,7 @@ asm_function jsimd_ycc_\colorid\()_convert_neon
    INPUT_BUF0      .req x5
    INPUT_BUF1      .req x6
-    INPUT_BUF2      .req INPUT_BUF
+    INPUT_BUF2      .req x1
    RGB             .req x7
    Y               .req x8
@@ -1728,16 +1729,16 @@ asm_function jsimd_ycc_\colorid\()_convert_neon
    sub             sp, sp, 336
    str             x15, [sp], 16
    /* Load constants to d1, d2, d3 (v0.4h is just used for padding) */
-    adr             x15, jsimd_ycc_\colorid\()_neon_consts
+    adr             x15, Ljsimd_ycc_\colorid\()_neon_consts
    /* Save NEON registers */
-    st1             {v0.8b - v3.8b}, [sp], 32
+    st1             {v0.8b, v1.8b, v2.8b, v3.8b}, [sp], 32
-    st1             {v4.8b - v7.8b}, [sp], 32
+    st1             {v4.8b, v5.8b, v6.8b, v7.8b}, [sp], 32
-    st1             {v8.8b - v11.8b}, [sp], 32
+    st1             {v8.8b, v9.8b, v10.8b, v11.8b}, [sp], 32
-    st1             {v12.8b - v15.8b}, [sp], 32
+    st1             {v12.8b, v13.8b, v14.8b, v15.8b}, [sp], 32
-    st1             {v16.8b - v19.8b}, [sp], 32
+    st1             {v16.8b, v17.8b, v18.8b, v19.8b}, [sp], 32
-    st1             {v20.8b - v23.8b}, [sp], 32
+    st1             {v20.8b, v21.8b, v22.8b, v23.8b}, [sp], 32
-    st1             {v24.8b - v27.8b}, [sp], 32
+    st1             {v24.8b, v25.8b, v26.8b, v27.8b}, [sp], 32
-    st1             {v28.8b - v31.8b}, [sp], 32
+    st1             {v28.8b, v29.8b, v30.8b, v31.8b}, [sp], 32
    ld1             {v0.4h, v1.4h}, [x15], 16
    ld1             {v2.8h}, [x15]
@@ -1748,8 +1749,8 @@ asm_function jsimd_ycc_\colorid\()_convert_neon
    stp             x8, x9, [sp], 16
    stp             x10, x30, [sp], 16
    ldr             INPUT_BUF0, [INPUT_BUF]
-    ldr             INPUT_BUF1, [INPUT_BUF, 8]
+    ldr             INPUT_BUF1, [INPUT_BUF, #8]
-    ldr             INPUT_BUF2, [INPUT_BUF, 16]
+    ldr             INPUT_BUF2, [INPUT_BUF, #16]
    .unreq          INPUT_BUF
    /* Initially set v10, v11.4h, v12.8b, d13 to 0xFF */
@@ -1758,7 +1759,7 @@ asm_function jsimd_ycc_\colorid\()_convert_neon
    /* Outer loop over scanlines */
    cmp             NUM_ROWS, #1
-    blt             9f
+    b.lt            9f
 0:
    lsl             x16, INPUT_ROW, #3
    ldr             Y, [INPUT_BUF0, x16]
@@ -1770,60 +1771,60 @@ asm_function jsimd_ycc_\colorid\()_convert_neon
    /* Inner loop over pixels */
    subs            N, N, #8
-    blt             3f
+    b.lt            3f
    do_load         8
    do_yuv_to_rgb_stage1
    subs            N, N, #8
-    blt             2f
+    b.lt            2f
 1:
    do_yuv_to_rgb_stage2_store_load_stage1
    subs            N, N, #8
-    bge             1b
+    b.ge            1b
 2:
    do_yuv_to_rgb_stage2
    do_store        \bpp, 8
    tst             N, #7
-    beq             8f
+    b.eq            8f
 3:
    tst             N, #4
-    beq             3f
+    b.eq            3f
    do_load         4
 3:
    tst             N, #2
-    beq             4f
+    b.eq            4f
    do_load         2
 4:
    tst             N, #1
-    beq             5f
+    b.eq            5f
    do_load         1
 5:
    do_yuv_to_rgb
    tst             N, #4
-    beq             6f
+    b.eq            6f
    do_store        \bpp, 4
 6:
    tst             N, #2
-    beq             7f
+    b.eq            7f
    do_store        \bpp, 2
 7:
    tst             N, #1
-    beq             8f
+    b.eq            8f
    do_store        \bpp, 1
 8:
    subs            NUM_ROWS, NUM_ROWS, #1
-    bgt             0b
+    b.gt            0b
 9:
    /* Restore all registers and return */
    sub             sp, sp, #336
    ldr             x15, [sp], 16
-    ld1             {v0.8b - v3.8b}, [sp], 32
+    ld1             {v0.8b, v1.8b, v2.8b, v3.8b}, [sp], 32
-    ld1             {v4.8b - v7.8b}, [sp], 32
+    ld1             {v4.8b, v5.8b, v6.8b, v7.8b}, [sp], 32
-    ld1             {v8.8b - v11.8b}, [sp], 32
+    ld1             {v8.8b, v9.8b, v10.8b, v11.8b}, [sp], 32
-    ld1             {v12.8b - v15.8b}, [sp], 32
+    ld1             {v12.8b, v13.8b, v14.8b, v15.8b}, [sp], 32
-    ld1             {v16.8b - v19.8b}, [sp], 32
+    ld1             {v16.8b, v17.8b, v18.8b, v19.8b}, [sp], 32
-    ld1             {v20.8b - v23.8b}, [sp], 32
+    ld1             {v20.8b, v21.8b, v22.8b, v23.8b}, [sp], 32
-    ld1             {v24.8b - v27.8b}, [sp], 32
+    ld1             {v24.8b, v25.8b, v26.8b, v27.8b}, [sp], 32
-    ld1             {v28.8b - v31.8b}, [sp], 32
+    ld1             {v28.8b, v29.8b, v30.8b, v31.8b}, [sp], 32
    /* pop             {r4, r5, r6, r7, r8, r9, r10, pc} */
    ldp             x4, x5, [sp], 16
    ldp             x6, x7, [sp], 16
--- a/simd/jsimd_powerpc.c
+++ b/simd/jsimd_powerpc.c
@@ -1,5 +1,5 @@
 /*
- * jsimd_powerpc64.c
+ * jsimd_powerpc.c
 *
 * Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
 * Copyright 2009-2011, 2014 D. R. Commander
@@ -42,12 +42,38 @@ init_simd (void)
 GLOBAL(int)
 jsimd_can_rgb_ycc (void)
 {
  init_simd();
  /* The code is optimised for these values only */
  if (BITS_IN_JSAMPLE != 8)
    return 0;
  if (sizeof(JDIMENSION) != 4)
    return 0;
  if ((RGB_PIXELSIZE != 3) && (RGB_PIXELSIZE != 4))
    return 0;
  if (simd_support & JSIMD_ALTIVEC)
    return 1;
  return 0;
 }
 GLOBAL(int)
 jsimd_can_rgb_gray (void)
 {
  init_simd();
  /* The code is optimised for these values only */
  if (BITS_IN_JSAMPLE != 8)
    return 0;
  if (sizeof(JDIMENSION) != 4)
    return 0;
  if ((RGB_PIXELSIZE != 3) && (RGB_PIXELSIZE != 4))
    return 0;
  if (simd_support & JSIMD_ALTIVEC)
    return 1;
  return 0;
 }
@@ -68,6 +94,37 @@ jsimd_rgb_ycc_convert (j_compress_ptr cinfo,
                       JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
                       JDIMENSION output_row, int num_rows)
 {
  void (*altivecfct)(JDIMENSION, JSAMPARRAY, JSAMPIMAGE, JDIMENSION, int);
  switch(cinfo->in_color_space) {
    case JCS_EXT_RGB:
      altivecfct=jsimd_extrgb_ycc_convert_altivec;
      break;
    case JCS_EXT_RGBX:
    case JCS_EXT_RGBA:
      altivecfct=jsimd_extrgbx_ycc_convert_altivec;
      break;
    case JCS_EXT_BGR:
      altivecfct=jsimd_extbgr_ycc_convert_altivec;
      break;
    case JCS_EXT_BGRX:
    case JCS_EXT_BGRA:
      altivecfct=jsimd_extbgrx_ycc_convert_altivec;
      break;
    case JCS_EXT_XBGR:
    case JCS_EXT_ABGR:
      altivecfct=jsimd_extxbgr_ycc_convert_altivec;
      break;
    case JCS_EXT_XRGB:
    case JCS_EXT_ARGB:
      altivecfct=jsimd_extxrgb_ycc_convert_altivec;
      break;
    default:
      altivecfct=jsimd_rgb_ycc_convert_altivec;
      break;
  }
  altivecfct(cinfo->image_width, input_buf, output_buf, output_row, num_rows);
 }
 GLOBAL(void)
@@ -75,6 +132,37 @@ jsimd_rgb_gray_convert (j_compress_ptr cinfo,
                        JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
                        JDIMENSION output_row, int num_rows)
 {
  void (*altivecfct)(JDIMENSION, JSAMPARRAY, JSAMPIMAGE, JDIMENSION, int);
  switch(cinfo->in_color_space) {
    case JCS_EXT_RGB:
      altivecfct=jsimd_extrgb_gray_convert_altivec;
      break;
    case JCS_EXT_RGBX:
    case JCS_EXT_RGBA:
      altivecfct=jsimd_extrgbx_gray_convert_altivec;
      break;
    case JCS_EXT_BGR:
      altivecfct=jsimd_extbgr_gray_convert_altivec;
      break;
    case JCS_EXT_BGRX:
    case JCS_EXT_BGRA:
      altivecfct=jsimd_extbgrx_gray_convert_altivec;
      break;
    case JCS_EXT_XBGR:
    case JCS_EXT_ABGR:
      altivecfct=jsimd_extxbgr_gray_convert_altivec;
      break;
    case JCS_EXT_XRGB:
    case JCS_EXT_ARGB:
      altivecfct=jsimd_extxrgb_gray_convert_altivec;
      break;
    default:
      altivecfct=jsimd_rgb_gray_convert_altivec;
      break;
  }
  altivecfct(cinfo->image_width, input_buf, output_buf, output_row, num_rows);
 }
 GLOBAL(void)
@@ -202,6 +290,21 @@ jsimd_h2v1_merged_upsample (j_decompress_ptr cinfo,
 GLOBAL(int)
 jsimd_can_convsamp (void)
 {
  init_simd();
  /* The code is optimised for these values only */
  if (DCTSIZE != 8)
    return 0;
  if (BITS_IN_JSAMPLE != 8)
    return 0;
  if (sizeof(JDIMENSION) != 4)
    return 0;
  if (sizeof(DCTELEM) != 2)
    return 0;
  if (simd_support & JSIMD_ALTIVEC)
    return 1;
  return 0;
 }
@@ -215,6 +318,7 @@ GLOBAL(void)
 jsimd_convsamp (JSAMPARRAY sample_data, JDIMENSION start_col,
                DCTELEM * workspace)
 {
  jsimd_convsamp_altivec(sample_data, start_col, workspace);
 }
 GLOBAL(void)
@@ -226,6 +330,17 @@ jsimd_convsamp_float (JSAMPARRAY sample_data, JDIMENSION start_col,
 GLOBAL(int)
 jsimd_can_fdct_islow (void)
 {
  init_simd();
  /* The code is optimised for these values only */
  if (DCTSIZE != 8)
    return 0;
  if (sizeof(DCTELEM) != 2)
    return 0;
  if (simd_support & JSIMD_ALTIVEC)
    return 1;
  return 0;
 }
@@ -255,6 +370,7 @@ jsimd_can_fdct_float (void)
 GLOBAL(void)
 jsimd_fdct_islow (DCTELEM * data)
 {
  jsimd_fdct_islow_altivec(data);
 }
 GLOBAL(void)
@@ -271,6 +387,19 @@ jsimd_fdct_float (FAST_FLOAT * data)
 GLOBAL(int)
 jsimd_can_quantize (void)
 {
  init_simd();
  /* The code is optimised for these values only */
  if (DCTSIZE != 8)
    return 0;
  if (sizeof(JCOEF) != 2)
    return 0;
  if (sizeof(DCTELEM) != 2)
    return 0;
  if (simd_support & JSIMD_ALTIVEC)
    return 1;
  return 0;
 }
@@ -284,6 +413,7 @@ GLOBAL(void)
 jsimd_quantize (JCOEFPTR coef_block, DCTELEM * divisors,
                DCTELEM * workspace)
 {
  jsimd_quantize_altivec(coef_block, divisors, workspace);
 }
 GLOBAL(void)
@@ -321,12 +451,34 @@ jsimd_idct_4x4 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
 GLOBAL(int)
 jsimd_can_idct_islow (void)
 {
  init_simd();
  /* The code is optimised for these values only */
  if (DCTSIZE != 8)
    return 0;
  if (sizeof(JCOEF) != 2)
    return 0;
  if (simd_support & JSIMD_ALTIVEC)
    return 1;
  return 0;
 }
 GLOBAL(int)
 jsimd_can_idct_ifast (void)
 {
  init_simd();
  /* The code is optimised for these values only */
  if (DCTSIZE != 8)
    return 0;
  if (sizeof(JCOEF) != 2)
    return 0;
  if (simd_support & JSIMD_ALTIVEC)
    return 1;
  return 0;
 }
@@ -341,6 +493,8 @@ jsimd_idct_islow (j_decompress_ptr cinfo, jpeg_component_info * compptr,
                  JCOEFPTR coef_block, JSAMPARRAY output_buf,
                  JDIMENSION output_col)
 {
  jsimd_idct_islow_altivec(compptr->dct_table, coef_block, output_buf,
                           output_col);
 }
 GLOBAL(void)
@@ -348,6 +502,8 @@ jsimd_idct_ifast (j_decompress_ptr cinfo, jpeg_component_info * compptr,
                  JCOEFPTR coef_block, JSAMPARRAY output_buf,
                  JDIMENSION output_col)
 {
  jsimd_idct_ifast_altivec(compptr->dct_table, coef_block, output_buf,
                           output_col);
 }
 GLOBAL(void)
--- a/tjbench.c
+++ b/tjbench.c
@@ -273,82 +273,6 @@ int decomp(unsigned char *srcbuf, unsigned char **jpegbuf,
 }
 int dotestyuv(unsigned char *srcbuf, int w, int h, int subsamp,
 	char *filename)
 {
 	char tempstr[1024], tempstr2[80];
 	FILE *file=NULL;  tjhandle handle=NULL;
 	unsigned char *dstbuf=NULL;
 	double start, elapsed;
 	int i, retval=0, ps=tjPixelSize[pf];
 	int yuvsize=0;
 	yuvsize=tjBufSizeYUV(w, h, subsamp);
 	if((dstbuf=(unsigned char *)malloc(yuvsize)) == NULL)
 		_throwunix("allocating image buffer");
 	if(!quiet)
 		printf(">>>>>  %s (%s) <--> YUV %s  <<<<<\n", pixFormatStr[pf],
 			(flags&TJFLAG_BOTTOMUP)? "Bottom-up":"Top-down", subNameLong[subsamp]);
 	if(quiet==1)
 		printf("%s\t%s\t%s\tN/A\t", pixFormatStr[pf],
 			(flags&TJFLAG_BOTTOMUP)? "BU":"TD", subNameLong[subsamp]);
 	if((handle=tjInitCompress())==NULL)
 		_throwtj("executing tjInitCompress()");
 	/* Execute once to preload cache */
 	if(tjEncodeYUV2(handle, srcbuf, w, 0, h, pf, dstbuf, subsamp, flags)==-1)
 		_throwtj("executing tjEncodeYUV2()");
 	/* Benchmark */
 	for(i=0, start=gettime(); (elapsed=gettime()-start)<benchtime; i++)
 	{
 		if(tjEncodeYUV2(handle, srcbuf, w, 0, h, pf, dstbuf, subsamp, flags)==-1)
 			_throwtj("executing tjEncodeYUV2()");
 	}
 	if(tjDestroy(handle)==-1) _throwtj("executing tjDestroy()");
 	handle=NULL;
 	if(quiet==1) printf("%-4d  %-4d\t", w, h);
 	if(quiet)
 	{
 		printf("%s%c%s%c",
 			sigfig((double)(w*h)/1000000.*(double)i/elapsed, 4, tempstr, 1024),
 			quiet==2? '\n':'\t',
 			sigfig((double)(w*h*ps)/(double)yuvsize, 4, tempstr2, 80),
 			quiet==2? '\n':'\t');
 	}
 	else
 	{
 		printf("\n%s size: %d x %d\n", "Image", w, h);
 		printf("C--> Frame rate:           %f fps\n", (double)i/elapsed);
 		printf("     Output image size:    %d bytes\n", yuvsize);
 		printf("     Compression ratio:    %f:1\n",
 			(double)(w*h*ps)/(double)yuvsize);
 		printf("     Source throughput:    %f Megapixels/sec\n",
 			(double)(w*h)/1000000.*(double)i/elapsed);
 		printf("     Output bit stream:    %f Megabits/sec\n",
 			(double)yuvsize*8./1000000.*(double)i/elapsed);
 	}
 	snprintf(tempstr, 1024, "%s_%s.yuv", filename, subName[subsamp]);
 	if((file=fopen(tempstr, "wb"))==NULL)
 		_throwunix("opening reference image");
 	if(fwrite(dstbuf, yuvsize, 1, file)!=1)
 		_throwunix("writing reference image");
 	fclose(file);  file=NULL;
 	if(!quiet) printf("Reference image written to %s\n", tempstr);
 	bailout:
 	if(file) {fclose(file);  file=NULL;}
 	if(dstbuf) {free(dstbuf);  dstbuf=NULL;}
 	if(handle) {tjDestroy(handle);  handle=NULL;}
 	return retval;
 }
 int fullTest(unsigned char *srcbuf, int w, int h, int subsamp, int jpegqual,
 	char *filename)
 {
--- a/3
+++ b/3
@@ -50,4 +50,7 @@ TURBOJPEG_1.4
 		tjDecompressToYUVPlanes;
 		tjEncodeYUV3;
 		tjEncodeYUVPlanes;
 		tjPlaneHeight;
 		tjPlaneSizeYUV;
 		tjPlaneWidth;
 } TURBOJPEG_1.2;
--- a/turbojpeg-mapfile.jni
+++ b/turbojpeg-mapfile.jni
@@ -76,6 +76,9 @@ TURBOJPEG_1.4
 		tjDecompressToYUVPlanes;
 		tjEncodeYUV3;
 		tjEncodeYUVPlanes;
 		tjPlaneHeight;
 		tjPlaneSizeYUV;
 		tjPlaneWidth;
 		Java_org_libjpegturbo_turbojpeg_TJ_bufSizeYUV__IIII;
 		Java_org_libjpegturbo_turbojpeg_TJCompressor_compressFromYUV___3_3B_3II_3III_3BII;
 		Java_org_libjpegturbo_turbojpeg_TJCompressor_encodeYUV___3BIIIIII_3_3B_3I_3III;
--- a/turbojpeg.c
+++ b/turbojpeg.c
@@ -206,7 +206,8 @@ static int setCompDefaults(struct jpeg_compress_struct *cinfo,
 	}
 	cinfo->input_components=tjPixelSize[pixelFormat];
-	cinfo->master->use_moz_defaults = TRUE;
+	if((env=getenv("TJ_REVERT"))!=NULL && strlen(env)>0 && !strcmp(env, "1"))
 		cinfo->master->compress_profile=JCP_FASTEST;
 	jpeg_set_defaults(cinfo);
 	if((env=getenv("TJ_OPTIMIZE"))!=NULL && strlen(env)>0 && !strcmp(env, "1"))
@@ -241,7 +242,7 @@ static int setCompDefaults(struct jpeg_compress_struct *cinfo,
 	else jpeg_set_colorspace(cinfo, JCS_YCbCr);
 	/* Set scan pattern again as colorspace might have changed */
-	if (cinfo->master->use_moz_defaults)
+	if(cinfo->master->compress_profile == JCP_MAX_COMPRESSION)
 		jpeg_simple_progression(cinfo);
 	cinfo->comp_info[0].h_samp_factor=tjMCUWidth[subsamp]/8;
@@ -319,6 +320,14 @@ static int setDecompDefaults(struct jpeg_decompress_struct *dinfo,
 static int getSubsamp(j_decompress_ptr dinfo)
 {
 	int retval=-1, i, k;
 	/* The sampling factors actually have no meaning with grayscale JPEG files,
 	   and in fact it's possible to generate grayscale JPEGs with sampling
 	   factors > 1 (even though those sampling factors are ignored by the
 	   decompressor.)  Thus, we need to treat grayscale as a special case. */
 	if(dinfo->num_components==1 && dinfo->jpeg_color_space==JCS_GRAYSCALE)
 		return TJSAMP_GRAY;
 	for(i=0; i<NUMSUBOPT; i++)
 	{
 		if(dinfo->num_components==pixelsize[i]
--- a/yuvjpeg.c
+++ b/yuvjpeg.c
@@ -196,6 +196,7 @@ int main(int argc, char *argv[]) {
  image_buffer =
   malloc(frame_width*frame_height + 2*(frame_width/2)*(frame_height/2));
  if (!image_buffer) {
    free(yuv_buffer);
    fprintf(stderr, "Memory allocation failure!\n");
    return 1;
  }
@@ -217,9 +218,6 @@ int main(int argc, char *argv[]) {
  jpeg_stdio_dest(&cinfo, jpg_fd);
  if (jpeg_c_bool_param_supported(&cinfo, JBOOLEAN_USE_MOZ_DEFAULTS))
    jpeg_c_set_bool_param(&cinfo, JBOOLEAN_USE_MOZ_DEFAULTS, TRUE);
  cinfo.image_width = luma_width;
  cinfo.image_height = luma_height;
  cinfo.input_components = 3;