diff --git a/ANNOUNCE b/ANNOUNCE
index bca4250..330fd10 100644
--- a/ANNOUNCE
+++ b/ANNOUNCE
@@ -1,4 +1,4 @@
-Libpng 1.6.24 - August 4, 2016
+Libpng 1.6.25 - September 1, 2016
 
 This is a public release of libpng, intended for use in production codes.
 
@@ -7,78 +7,29 @@
 Source files with LF line endings (for Unix/Linux) and with a
 "configure" script
 
-   libpng-1.6.24.tar.xz (LZMA-compressed, recommended)
-   libpng-1.6.24.tar.gz
+   libpng-1.6.25.tar.xz (LZMA-compressed, recommended)
+   libpng-1.6.25.tar.gz
 
 Source files with CRLF line endings (for Windows), without the
 "configure" script
 
-   lpng1624.7z  (LZMA-compressed, recommended)
-   lpng1624.zip
+   lpng1625.7z  (LZMA-compressed, recommended)
+   lpng1625.zip
 
 Other information:
 
-   libpng-1.6.24-README.txt
-   libpng-1.6.24-LICENSE.txt
-   libpng-1.6.24-*.asc (armored detached GPG signatures)
+   libpng-1.6.25-README.txt
+   libpng-1.6.25-LICENSE.txt
+   libpng-1.6.25-*.asc (armored detached GPG signatures)
 
-Changes since the last public release (1.6.23):
-  Avoid potential overflow of the PNG_IMAGE_SIZE macro.  This macro
-    is not used within libpng, but is used in some of the examples.
-  Correct filter heuristic overflow handling. This was broken when the
-    write filter code was moved out-of-line; if there is a single filter and
-    the heuristic sum overflows the calculation of the filtered line is not
-    completed.  In versions prior to 1.6 the code was duplicated in-line
-    and the check not performed, so the filter operation completed; however,
-    in the multi-filter case where the sum is performed the 'none' filter would
-    be selected if all the sums overflowed, even if it wasn't in the filter
-    list.  The fix to the first problem is simply to provide PNG_SIZE_MAX as
-    the current lmins sum value; this means the sum can never exceed it and
-    overflows silently.  A reasonable compiler that does choose to inline
-    the code will simply eliminate the sum check.
-  The fix to the second problem is to use high precision arithmetic (this is
-    implemented in 1.7), however a simple safe fix here is to chose the lowest
-    numbered filter in the list from png_set_filter (this only works if the
-    first problem is also fixed) (John Bowler).
-  Use a more efficient absolute value calculation on SSE2 (Matthieu Darbois).
-  Fixed the case where PNG_IMAGE_BUFFER_SIZE can overflow in the application
-    as a result of the application using an increased 'row_stride'; previously
-    png_image_finish_read only checked for overflow on the base calculation of
-    components.  (I.e. it checked for overflow of a 32-bit number on the total
-    number of pixel components in the output format, not the possibly padded row
-    length and not the number of bytes, which for linear formats is twice the
-    number of components.)
-  MSVC does not like '-(unsigned)', so replaced it with 0U-(unsigned)
-  MSVC does not like (uInt) = -(unsigned) (i.e. as an initializer), unless
-    the conversion is explicitly invoked by a cast.
-  Put the SKIP definition in the correct place. It needs to come after the
-    png.h include (see all the other .c files in contrib/libtests) because it
-    depends on PNG_LIBPNG_VER.
-  Removed the three compile warning options from the individual project
-    files into the zlib.props globals.  It increases the warning level from 4
-    to All and adds a list of the warnings that need to be turned off.  This is
-    semi-documentary; the intent is to tell libpng users which warnings have
-    been examined and judged non-fixable at present.  The warning about
-    structure padding is fixable, but it would be a signficant change (moving
-    structure members around).
-  Optimized absolute value calculation in filter selection, similar to
-    code in the PAETH decoder in pngrutil.c. Build with PNG_USE_ABS to
-    use this.
-  Added pngcp to the build together with a pngcp.dfa configuration test.
-  Added high resolution timing to pngcp.
-  Added "Common linking failures" section to INSTALL.
-  Relocated misplaced #endif in png.c sRGB profile checking.
-  Fixed two Coverity issues in pngcp.c.
-  Avoid filter-selection heuristic sum calculations in cases where only one
-    filter is a candidate for selection. This trades off code size (added
-    private png_setup_*_row_only() functions) for speed.
-  Fixed some indentation to comply with our coding style.
-  Added contrib/tools/reindent.
-  Eliminated unnecessary tests of boolean png_isaligned() vs 0.
-  Conditionally compile SSE2 headers in contrib/intel/intel_sse.patch
-  Conditionally compile png_decompress_chunk().
-  Conditionally compile ARM_NEON headers in pngpriv.h
-  Updated contrib/intel/intel_sse.patch
+Changes since the last public release (1.6.24):
+  Reject oversized iCCP profile immediately.
+  Cleaned up PNG_DEBUG compile of pngtest.c.
+  Conditionally compile png_inflate().
+  Don't install pngcp; it conflicts with pngcp in the pngtools package.
+  Minor editing of INSTALL, (whitespace, added copyright line)
+  Added MIPS support (Mandar Sahastrabuddhe <Mandar.Sahastrabuddhe@imgtec.com>).
+  Rebased contrib/intel/intel_sse.patch after the MIPS implementation.
 
 (subscription required; visit
 https://lists.sourceforge.net/lists/listinfo/png-mng-implement
diff --git a/CHANGES b/CHANGES
index d4f0c8b..923fc50 100644
--- a/CHANGES
+++ b/CHANGES
@@ -5675,6 +5675,34 @@
 Version 1.6.24[August 4, 2016]
   No changes.
 
+Version 1.6.25beta01 [August 12, 2016]
+  Reject oversized iCCP profile immediately.
+  Cleaned up PNG_DEBUG compile of pngtest.c.
+  Conditionally compile png_inflate().
+
+Version 1.6.25beta02 [August 18, 2016]
+  Don't install pngcp; it conflicts with pngcp in the pngtools package.
+  Minor editing of INSTALL, (whitespace, added copyright line)
+
+Version 1.6.25rc01 [August 24, 2016]
+  No changes.
+
+Version 1.6.25rc02 [August 29, 2016]
+  Added MIPS support (Mandar Sahastrabuddhe <Mandar.Sahastrabuddhe@imgtec.com>).
+  Only the UP filter is currently implemented.
+
+Version 1.6.25rc03 [August 29, 2016]
+  Rebased contrib/intel/intel_sse.patch after the MIPS implementation.
+
+Version 1.6.25rc04 [August 30, 2016]
+  Added MIPS support for SUB, AVG, and PAETH filters (Mandar Sahastrabuddhe).
+
+Version 1.6.25rc05 [August 30, 2016]
+  Rebased contrib/intel/intel_sse.patch after the MIPS implementation update..
+
+Version 1.6.25 [September 1, 2016]
+  No changes.
+
 Send comments/corrections/commendations to png-mng-implement at lists.sf.net
 (subscription required; visit
 https://lists.sourceforge.net/lists/listinfo/png-mng-implement
diff --git a/CMakeLists.txt b/CMakeLists.txt
index 186b65a..eb63b2b 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -35,7 +35,7 @@
 
 set(PNGLIB_MAJOR 1)
 set(PNGLIB_MINOR 6)
-set(PNGLIB_RELEASE 24)
+set(PNGLIB_RELEASE 25)
 set(PNGLIB_NAME libpng${PNGLIB_MAJOR}${PNGLIB_MINOR})
 set(PNGLIB_VERSION ${PNGLIB_MAJOR}.${PNGLIB_MINOR}.${PNGLIB_RELEASE})
 
@@ -696,7 +696,7 @@
 # SET UP LINKS
 if(PNG_SHARED)
   set_target_properties(png PROPERTIES
-#   VERSION 16.${PNGLIB_RELEASE}.1.6.24
+#   VERSION 16.${PNGLIB_RELEASE}.1.6.25
     VERSION 16.${PNGLIB_RELEASE}.0
     SOVERSION 16
     CLEAN_DIRECT_OUTPUT 1)
diff --git a/INSTALL b/INSTALL
index eed4dc8..58ec0b6 100644
--- a/INSTALL
+++ b/INSTALL
@@ -1,25 +1,25 @@
 
-Installing libpng
+    Installing libpng
 
 Contents
 
-   I. Simple installation
-  II. Rebuilding the configure scripts
- III. Using scripts/makefile*
-  IV. Using cmake
-   V. Directory structure
-  VI. Building with project files
- VII. Building with makefiles
-VIII. Configuring libpng for 16-bit platforms
-  IX. Configuring for DOS
-   X. Configuring for Medium Model
-  XI. Prepending a prefix to exported symbols
- XII. Configuring for compiler xxx:
-XIII. Removing unwanted object code
- XIV. Changes to the build and configuration of libpng in libpng-1.5.x
-  XV. Setjmp/longjmp issues
- XVI. Common linking failures
-XVII. Other sources of information about libpng
+       I. Simple installation
+      II. Rebuilding the configure scripts
+     III. Using scripts/makefile*
+      IV. Using cmake
+       V. Directory structure
+      VI. Building with project files
+     VII. Building with makefiles
+    VIII. Configuring libpng for 16-bit platforms
+      IX. Configuring for DOS
+       X. Configuring for Medium Model
+      XI. Prepending a prefix to exported symbols
+     XII. Configuring for compiler xxx:
+    XIII. Removing unwanted object code
+     XIV. Changes to the build and configuration of libpng in libpng-1.5.x
+      XV. Setjmp/longjmp issues
+     XVI. Common linking failures
+    XVII. Other sources of information about libpng
 
 I. Simple installation
 
@@ -90,22 +90,24 @@
 and LD_LIBRARY_PATH in your environment before running "make test"
 or "make distcheck":
 
-ZLIBLIB=/path/to/lib export ZLIBLIB
-ZLIBINC=/path/to/include export ZLIBINC
-CPPFLAGS="-I$ZLIBINC" export CPPFLAGS
-LDFLAGS="-L$ZLIBLIB" export LDFLAGS
-LD_LIBRARY_PATH="$ZLIBLIB:$LD_LIBRARY_PATH" export LD_LIBRARY_PATH
+    ZLIBLIB=/path/to/lib export ZLIBLIB
+    ZLIBINC=/path/to/include export ZLIBINC
+    CPPFLAGS="-I$ZLIBINC" export CPPFLAGS
+    LDFLAGS="-L$ZLIBLIB" export LDFLAGS
+    LD_LIBRARY_PATH="$ZLIBLIB:$LD_LIBRARY_PATH" export LD_LIBRARY_PATH
 
 If you are using one of the makefile scripts, put ZLIBLIB and ZLIBINC
-in your environment and type "make ZLIBLIB=$ZLIBLIB ZLIBINC=$ZLIBINC test".
+in your environment and type
+
+    make ZLIBLIB=$ZLIBLIB ZLIBINC=$ZLIBINC test
 
 IV. Using cmake
 
 If you want to use "cmake" (see www.cmake.org), type
 
-   cmake . -DCMAKE_INSTALL_PREFIX=/path
-   make
-   make install
+    cmake . -DCMAKE_INSTALL_PREFIX=/path
+    make
+    make install
 
 As when using the simple configure method described above, "/path" points to
 the installation directory where you want to put the libpng "lib", "include",
@@ -119,7 +121,7 @@
 
 Your directory structure should look like this:
 
-   .. (the parent directory)
+    .. (the parent directory)
       libpng (this directory)
           INSTALL (this file)
           README
@@ -163,10 +165,15 @@
 Copy the file (or files) that you need from the
 scripts directory into this directory, for example
 
-   MSDOS example: copy scripts\makefile.msc makefile
-                  copy scripts\pnglibconf.h.prebuilt pnglibconf.h
-   UNIX example:  cp scripts/makefile.std makefile
-                  cp scripts/pnglibconf.h.prebuilt pnglibconf.h
+MSDOS example:
+
+    copy scripts\makefile.msc makefile
+    copy scripts\pnglibconf.h.prebuilt pnglibconf.h
+
+UNIX example:
+
+    cp scripts/makefile.std makefile
+    cp scripts/pnglibconf.h.prebuilt pnglibconf.h
 
 Read the makefile to see if you need to change any source or
 target directories to match your preferences.
@@ -242,7 +249,7 @@
 never going to use a capability, you can change the #define to #undef
 before recompiling libpng and save yourself code and data space, or
 you can turn off individual capabilities with defines that begin with
-PNG_NO_.
+"PNG_NO_".
 
 In libpng-1.5.0 and later, the #define's are in pnglibconf.h instead.
 
@@ -310,7 +317,7 @@
 approach is documented in pngconf.h
 
 Despite these changes, libpng 1.5.0 only supports the native C function
-calling standard on those platforms tested so far (__cdecl on Microsoft
+calling standard on those platforms tested so far ("__cdecl" on Microsoft
 Windows).  This is because the support requirements for alternative
 calling conventions seem to no longer exist.  Developers who find it
 necessary to set PNG_API_RULE to 1 should advise the mailing list
@@ -374,7 +381,7 @@
 configure libpng with PNG_NO_SETJMP in your pngusr.dfa file, with
 -DPNG_NO_SETJMP on your compile line, or with
 
-  #undef PNG_SETJMP_SUPPORTED
+    #undef PNG_SETJMP_SUPPORTED
 
 in your pnglibconf.h or pngusr.h.
 
@@ -400,3 +407,8 @@
 Further information can be found in the README and libpng-manual.txt
 files, in the individual makefiles, in png.h, and the manual pages
 libpng.3 and png.5.
+
+Copyright (c) 1998-2002,2006-2016 Glenn Randers-Pehrson
+This document is released under the libpng license.
+For conditions of distribution and use, see the disclaimer
+and license in png.h.
diff --git a/LICENSE b/LICENSE
index d20f39e..d48a293 100644
--- a/LICENSE
+++ b/LICENSE
@@ -10,7 +10,7 @@
 
 This code is released under the libpng license.
 
-libpng versions 1.0.7, July 1, 2000 through 1.6.24, August 4, 2016 are
+libpng versions 1.0.7, July 1, 2000 through 1.6.25, September 1, 2016 are
 Copyright (c) 2000-2002, 2004, 2006-2016 Glenn Randers-Pehrson, are
 derived from libpng-1.0.6, and are distributed according to the same
 disclaimer and license as libpng-1.0.6 with the following individuals
@@ -127,4 +127,4 @@
 
 Glenn Randers-Pehrson
 glennrp at users.sourceforge.net
-August 4, 2016
+September 1, 2016
diff --git a/Makefile.am b/Makefile.am
index 39e9bde..fb209ed 100644
--- a/Makefile.am
+++ b/Makefile.am
@@ -1,7 +1,7 @@
 # Makefile.am, the source file for Makefile.in (and hence Makefile), is
 #
 # Copyright (c) 2004-2016 Glenn Randers-Pehrson
-# Last changed in libpng 1.6.22 [(PENDING RELEASE)]
+# Last changed in libpng 1.6.25 [September 1, 2016]
 #
 # This code is released under the libpng license.
 # For conditions of distribution and use, see the disclaimer
@@ -12,13 +12,13 @@
 ACLOCAL_AMFLAGS = -I scripts
 
 # test programs - run on make check, make distcheck
-check_PROGRAMS= pngtest pngunknown pngstest pngvalid pngimage
+check_PROGRAMS= pngtest pngunknown pngstest pngvalid pngimage pngcp
 if HAVE_CLOCK_GETTIME
 check_PROGRAMS += timepng
 endif
 
 # Utilities - installed
-bin_PROGRAMS= pngcp pngfix png-fix-itxt
+bin_PROGRAMS= pngfix png-fix-itxt
 
 # This ensures that pnglibconf.h gets built at the start of 'make all' or
 # 'make check', but it does not add dependencies to the individual programs,
@@ -97,6 +97,11 @@
 	arm/filter_neon.S arm/filter_neon_intrinsics.c
 endif
 
+if PNG_MIPS_MSA
+libpng@PNGLIB_MAJOR@@PNGLIB_MINOR@_la_SOURCES += mips/mips_init.c\
+	mips/filter_msa_intrinsics.c
+endif
+
 nodist_libpng@PNGLIB_MAJOR@@PNGLIB_MINOR@_la_SOURCES = pnglibconf.h
 
 libpng@PNGLIB_MAJOR@@PNGLIB_MINOR@_la_LDFLAGS = -no-undefined -export-dynamic \
diff --git a/README b/README
index a8fba91..17bca9a 100644
--- a/README
+++ b/README
@@ -1,4 +1,4 @@
-README for libpng version 1.6.24 - August 4, 2016 (shared library 16.0)
+README for libpng version 1.6.25 - September 1, 2016 (shared library 16.0)
 See the note about version numbers near the top of png.h
 
 See INSTALL for instructions on how to install libpng.
@@ -180,15 +180,18 @@
       pngwutil.c    =>  Write utility functions
       arm           =>  Contains optimized code for the ARM platform
       contrib       =>  Contributions
+       arm-neon         =>  Optimized code for ARM-NEON platform
        examples         =>  Example programs
        gregbook         =>  source code for PNG reading and writing, from
                             Greg Roelofs' "PNG: The Definitive Guide",
                             O'Reilly, 1999
+       intel            =>  Optimized code for INTEL-SSE2 platform
        libtests         =>  Test programs
        pngminim         =>  Minimal decoder, encoder, and progressive decoder
                             programs demonstrating use of pngusr.dfa
        pngminus         =>  Simple pnm2png and png2pnm programs
        pngsuite         =>  Test images
+       testpngs
        tools            =>  Various tools
        visupng          =>  Contains a MSVC workspace for VisualPng
       projects      =>  Contains project files and workspaces for
diff --git a/TODO b/TODO
index 0b3b651..cdb9e1f 100644
--- a/TODO
+++ b/TODO
@@ -5,7 +5,9 @@
 Better C++ wrapper/full C++ implementation?
 Fix problem with C++ and EXTERN "C".
 cHRM transformation.
-Remove setjmp/longjmp usage in favor of returning error codes.
+Remove setjmp/longjmp usage in favor of returning error codes. As a start on
+  this, minimize the use of png_error(), replacing them with
+  png_warning(); return(0; or similar.
 Palette creation.
 Add "grayscale->palette" transformation and "palette->grayscale" detection.
 Improved dithering.
diff --git a/configure.ac b/configure.ac
index 3a3d208..83c5fc2 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1,7 +1,7 @@
 # configure.ac
 
-# Copyright (c) 2004-2015 Glenn Randers-Pehrson
-# Last changed in libpng 1.6.22 [May 26, 2016]
+# Copyright (c) 2004-2016 Glenn Randers-Pehrson
+# Last changed in libpng 1.6.25 [September 1, 2016]
 
 # This code is released under the libpng license.
 # For conditions of distribution and use, see the disclaimer
@@ -25,7 +25,7 @@
 
 dnl Version number stuff here:
 
-AC_INIT([libpng],[1.6.24],[png-mng-implement@lists.sourceforge.net])
+AC_INIT([libpng],[1.6.25],[png-mng-implement@lists.sourceforge.net])
 AC_CONFIG_MACRO_DIR([scripts])
 
 # libpng does not follow GNU file name conventions (hence 'foreign')
@@ -46,10 +46,10 @@
 dnl AM_PREREQ([1.11.2])
 dnl stop configure from automagically running automake
 
-PNGLIB_VERSION=1.6.24
+PNGLIB_VERSION=1.6.25
 PNGLIB_MAJOR=1
 PNGLIB_MINOR=6
-PNGLIB_RELEASE=24
+PNGLIB_RELEASE=25
 
 dnl End of version number stuff
 
@@ -346,6 +346,51 @@
       *)    test "$enable_arm_neon" != '';;
     esac])
 
+# MIPS
+# ===
+#
+# MIPS MSA (SIMD) support.
+
+AC_ARG_ENABLE([mips-msa],
+   AS_HELP_STRING([[[--enable-mips-msa]]],
+      [Enable MIPS MSA optimizations: =no/off, check, api, yes/on:]
+      [no/off: disable the optimizations; check: use internal checking code]
+      [(deprecated and poorly supported); api: disable by default, enable by]
+      [a call to png_set_option; yes/on: turn on unconditionally.]
+      [If not specified: determined by the compiler.]),
+   [case "$enableval" in
+      no|off)
+         # disable the default enabling on __mips_msa systems:
+         AC_DEFINE([PNG_MIPS_MSA_OPT], [0],
+                   [Disable MIPS MSA optimizations])
+         # Prevent inclusion of the assembler files below:
+         enable_mips_msa=no;;
+      check)
+         AC_DEFINE([PNG_MIPS_MSA_CHECK_SUPPORTED], [],
+                   [Check for MIPS MSA support at run-time]);;
+      api)
+         AC_DEFINE([PNG_MIPS_MSA_API_SUPPORTED], [],
+                   [Turn on MIPS MSA optimizations at run-time]);;
+      yes|on)
+         AC_DEFINE([PNG_MIPS_MSA_OPT], [2],
+                   [Enable MIPS MSA optimizations])
+         AC_MSG_WARN([--enable-mips-msa: please specify 'check' or 'api', if]
+            [you want the optimizations unconditionally pass '-mmsa -mfp64']
+            [to the compiler.]);;
+      *)
+         AC_MSG_ERROR([--enable-mips-msa=${enable_mips_msa}: invalid value])
+   esac])
+
+# Add MIPS specific files to all builds where the host_cpu is mips ('mips*') or
+# where MIPS optimizations were explicitly requested (this allows a fallback if a
+# future host CPU does not match 'mips*')
+
+AM_CONDITIONAL([PNG_MIPS_MSA],
+   [test "$enable_mips_msa" != 'no' &&
+    case "$host_cpu" in
+      mipsel*|mips64el*) :;;
+    esac])
+
 AC_MSG_NOTICE([[Extra options for compiler: $PNG_COPTS]])
 
 # Config files, substituting as above
diff --git a/contrib/intel/intel_sse.patch b/contrib/intel/intel_sse.patch
index c5579e3..a852538 100644
--- a/contrib/intel/intel_sse.patch
+++ b/contrib/intel/intel_sse.patch
@@ -1,13 +1,13 @@
 diff --git a/configure.ac b/configure.ac
---- a/configure.ac	2016-05-25 18:59:10.000000000 -0400
-+++ b/configure.ac	2016-05-25 19:48:10.631751170 -0400
-@@ -341,16 +341,50 @@ AC_ARG_ENABLE([arm-neon],
+--- a/configure.ac	2016-08-29 11:46:27.000000000 -0400
++++ b/configure.ac	2016-08-29 16:57:03.866355018 -0400
+@@ -386,16 +386,51 @@ AC_ARG_ENABLE([mips-msa],
+ # future host CPU does not match 'mips*')
  
- AM_CONDITIONAL([PNG_ARM_NEON],
-    [test "$enable_arm_neon" != 'no' &&
+ AM_CONDITIONAL([PNG_MIPS_MSA],
+    [test "$enable_mips_msa" != 'no' &&
      case "$host_cpu" in
-       arm*|aarch64*) :;;
-       *)    test "$enable_arm_neon" != '';;
+       mipsel*|mips64el*) :;;
      esac])
  
 +# INTEL
@@ -44,6 +44,7 @@
 +      i?86|x86_64) :;;
 +      *)    test "$enable_intel_sse" != '';;
 +    esac])
++
  AC_MSG_NOTICE([[Extra options for compiler: $PNG_COPTS]])
  
  # Config files, substituting as above
@@ -53,21 +54,22 @@
  
  AC_OUTPUT
 diff --git a/Makefile.am b/Makefile.am
---- a/Makefile.am	2016-05-17 18:15:12.000000000 -0400
-+++ b/Makefile.am	2016-05-25 19:48:10.631751170 -0400
-@@ -92,16 +92,20 @@ libpng@PNGLIB_MAJOR@@PNGLIB_MINOR@_la_SO
- 	pngset.c pngtrans.c pngwio.c pngwrite.c pngwtran.c pngwutil.c\
- 	png.h pngconf.h pngdebug.h pnginfo.h pngpriv.h pngstruct.h pngusr.dfa
- 
- if PNG_ARM_NEON
- libpng@PNGLIB_MAJOR@@PNGLIB_MINOR@_la_SOURCES += arm/arm_init.c\
+--- a/Makefile.am	2016-08-29 11:46:27.000000000 -0400
++++ b/Makefile.am	2016-08-29 16:57:45.955528215 -0400
+@@ -97,16 +97,21 @@ libpng@PNGLIB_MAJOR@@PNGLIB_MINOR@_la_SO
  	arm/filter_neon.S arm/filter_neon_intrinsics.c
  endif
  
+ if PNG_MIPS_MSA
+ libpng@PNGLIB_MAJOR@@PNGLIB_MINOR@_la_SOURCES += mips/mips_init.c\
+ 	mips/filter_msa_intrinsics.c
+ endif
+ 
 +if PNG_INTEL_SSE
 +libpng@PNGLIB_MAJOR@@PNGLIB_MINOR@_la_SOURCES += contrib/intel/intel_init.c\
 +    contrib/intel/filter_sse2_intrinsics.c
 +endif
++
  nodist_libpng@PNGLIB_MAJOR@@PNGLIB_MINOR@_la_SOURCES = pnglibconf.h
  
  libpng@PNGLIB_MAJOR@@PNGLIB_MINOR@_la_LDFLAGS = -no-undefined -export-dynamic \
@@ -77,16 +79,16 @@
  #   Versioned symbols and restricted exports
  if HAVE_SOLARIS_LD
 diff --git a/pngpriv.h b/pngpriv.h
---- a/pngpriv.h	2016-08-01 18:13:38.770128810 -0500
-+++ b/pngpriv.h	2016-08-01 18:50:19.130179017 -0500
-@@ -177,16 +177,52 @@
- #  endif /* !PNG_ARM_NEON_IMPLEMENTATION */
- 
- #  ifndef PNG_ARM_NEON_IMPLEMENTATION
-       /* Use the intrinsics code by default. */
- #     define PNG_ARM_NEON_IMPLEMENTATION 1
+--- debug16/pngpriv.h	2016-08-30 10:46:36.000000000 -0400
++++ libpng16/pngpriv.h	2016-08-30 11:57:25.672280202 -0400
+@@ -185,16 +185,52 @@
+ #ifndef PNG_MIPS_MSA_OPT
+ #  if defined(__mips_msa) && (__mips_isa_rev >= 5) && defined(PNG_ALIGNED_MEMORY_SUPPORTED)
+ #     define PNG_MIPS_MSA_OPT 2
+ #  else
+ #     define PNG_MIPS_MSA_OPT 0
  #  endif
- #endif /* PNG_ARM_NEON_OPT > 0 */
+ #endif
  
 +#ifndef PNG_INTEL_SSE_OPT
 +#   ifdef PNG_INTEL_SSE
@@ -124,24 +126,23 @@
 +#   endif
 +#endif
 +
- /* Is this a build of a DLL where compilation of the object modules requires
-  * different preprocessor settings to those required for a simple library?  If
-  * so PNG_BUILD_DLL must be set.
-  *
-  * If libpng is used inside a DLL but that DLL does not export the libpng APIs
-  * PNG_BUILD_DLL must not be set.  To avoid the code below kicking in build a
-  * static library of libpng then link the DLL against that.
-  */
-@@ -1185,16 +1221,31 @@ PNG_INTERNAL_FUNCTION(void,png_read_filt
+ #if PNG_MIPS_MSA_OPT > 0
+ #  define PNG_FILTER_OPTIMIZATIONS png_init_filter_functions_msa
+ #  ifndef PNG_MIPS_MSA_IMPLEMENTATION
+ #     if defined(__mips_msa)
+ #        if defined(__clang__)
+ #        elif defined(__GNUC__)
+ #           if __GNUC__ < 4 || (__GNUC__ == 4 && __GNUC_MINOR__ < 7)
+ #              define PNG_MIPS_MSA_IMPLEMENTATION 2
+@@ -1232,16 +1268,31 @@ PNG_INTERNAL_FUNCTION(void,png_read_filt
+ PNG_INTERNAL_FUNCTION(void,png_read_filter_row_avg4_msa,(png_row_infop
      row_info, png_bytep row, png_const_bytep prev_row),PNG_EMPTY);
- PNG_INTERNAL_FUNCTION(void,png_read_filter_row_avg4_neon,(png_row_infop
+ PNG_INTERNAL_FUNCTION(void,png_read_filter_row_paeth3_msa,(png_row_infop
      row_info, png_bytep row, png_const_bytep prev_row),PNG_EMPTY);
- PNG_INTERNAL_FUNCTION(void,png_read_filter_row_paeth3_neon,(png_row_infop
-     row_info, png_bytep row, png_const_bytep prev_row),PNG_EMPTY);
- PNG_INTERNAL_FUNCTION(void,png_read_filter_row_paeth4_neon,(png_row_infop
+ PNG_INTERNAL_FUNCTION(void,png_read_filter_row_paeth4_msa,(png_row_infop
      row_info, png_bytep row, png_const_bytep prev_row),PNG_EMPTY);
  #endif
-+ 
+ 
 +#if PNG_INTEL_SSE_IMPLEMENTATION > 0
 +PNG_INTERNAL_FUNCTION(void,png_read_filter_row_sub3_sse2,(png_row_infop
 +    row_info, png_bytep row, png_const_bytep prev_row),PNG_EMPTY);
@@ -156,7 +157,7 @@
 +PNG_INTERNAL_FUNCTION(void,png_read_filter_row_paeth4_sse2,(png_row_infop
 +    row_info, png_bytep row, png_const_bytep prev_row),PNG_EMPTY);
 +#endif
- 
++
  /* Choose the best filter to use and filter the row data */
  PNG_INTERNAL_FUNCTION(void,png_write_find_filter,(png_structrp png_ptr,
      png_row_infop row_info),PNG_EMPTY);
@@ -164,15 +165,17 @@
  #ifdef PNG_SEQUENTIAL_READ_SUPPORTED
  PNG_INTERNAL_FUNCTION(void,png_read_IDAT_data,(png_structrp png_ptr,
     png_bytep output, png_alloc_size_t avail_out),PNG_EMPTY);
-@@ -1914,16 +1965,20 @@ PNG_INTERNAL_FUNCTION(void, PNG_FILTER_O
-    /* List *all* the possible optimizations here - this branch is required if
-     * the builder of libpng passes the definition of PNG_FILTER_OPTIMIZATIONS in
-     * CFLAGS in place of CPPFLAGS *and* uses symbol prefixing.
-     */
- #  if PNG_ARM_NEON_OPT > 0
+    /* Read 'avail_out' bytes of data from the IDAT stream.  If the output buffer
+@@ -1967,16 +2018,21 @@ PNG_INTERNAL_FUNCTION(void, PNG_FILTER_O
  PNG_INTERNAL_FUNCTION(void, png_init_filter_functions_neon,
     (png_structp png_ptr, unsigned int bpp), PNG_EMPTY);
- #  endif
+ #endif
+ 
+ #if PNG_MIPS_MSA_OPT > 0
+ PNG_INTERNAL_FUNCTION(void, png_init_filter_functions_msa,
+    (png_structp png_ptr, unsigned int bpp), PNG_EMPTY);
+ #endif
++
 +#  if PNG_INTEL_SSE_IMPLEMENTATION > 0
 +PNG_INTERNAL_FUNCTION(void, png_init_filter_functions_sse2,
 +   (png_structp png_ptr, unsigned int bpp), PNG_EMPTY);
diff --git a/contrib/mips-msa/README b/contrib/mips-msa/README
new file mode 100644
index 0000000..10acc93
--- /dev/null
+++ b/contrib/mips-msa/README
@@ -0,0 +1,83 @@
+OPERATING SYSTEM SPECIFIC MIPS MSA DETECTION
+--------------------------------------------
+
+Detection of the ability to execute MIPS MSA on an MIPS processor requires
+operating system support.  (The information is not available in user mode.)
+
+HOW TO USE THIS
+---------------
+
+This directory contains C code fragments that can be included in mips/mips_init.c
+by setting the macro PNG_MIPS_MSA_FILE to the file name in "" or <> at build
+time.  This setting is not recorded in pnglibconf.h and can be changed simply by
+rebuilding mips/msa_init.o with the required macro definition.
+
+For any of this code to be used the MIPS MSA code must be enabled and run time
+checks must be supported.  I.e.:
+
+#if PNG_MIPS_MSA_OPT > 0
+#ifdef PNG_MIPS_MSA_CHECK_SUPPORTED
+
+This is done in a 'configure' build by passing configure the argument:
+
+   --enable-mips-msa=check
+
+Apart from the basic Linux implementation in contrib/mips-msa/linux.c this code
+is unsupported.  That means that it is not even compiled on a regular basis and
+may be broken in any given minor release.
+
+FILE FORMAT
+-----------
+
+Each file documents its testing status as of the last time it was tested (which
+may have been a long time ago):
+
+STATUS: one of:
+   SUPPORTED: This indicates that the file is included in the regularly
+         performed test builds and bugs are fixed when discovered.
+   COMPILED: This indicates that the code did compile at least once.  See the
+         more detailed description for the extent to which the result was
+         successful.
+   TESTED: This means the code was fully compiled into the libpng test programs
+         and these were run at least once.
+
+BUG REPORTS: an email address to which to send reports of problems
+
+The file is a fragment of C code. It should not define any 'extern' symbols;
+everything should be static.  It must define the function:
+
+static int png_have_msa(png_structp png_ptr);
+
+That function must return 1 if MIPS MSA instructions are supported, 0 if not.
+It must not execute png_error unless it detects a bug.  A png_error will prevent
+the reading of the PNG and in the future, writing too.
+
+BUG REPORTS
+-----------
+
+If you mail a bug report for any file that is not SUPPORTED there may only be
+limited response.  Consider fixing it and sending a patch to fix the problem -
+this is more likely to result in action.
+
+CONTRIBUTIONS
+-------------
+
+You may send contributions of new implementations to
+png-mng-implement@sourceforge.net.  Please write code in strict C90 C where
+possible.  Obviously OS dependencies are to be expected.  If you submit code you
+must have the authors permission and it must have a license that is acceptable
+to the current maintainer; in particular that license must permit modification
+and redistribution.
+
+Please try to make the contribution a single file and give the file a clear and
+unambiguous name that identifies the target OS.  If multiple files really are
+required put them all in a sub-directory.
+
+You must also be prepared to handle bug reports from users of the code, either
+by joining the png-mng-implement mailing list or by providing an email for the
+"BUG REPORTS" entry or both.  Please make sure that the header of the file
+contains the STATUS and BUG REPORTS fields as above.
+
+Please list the OS requirements as precisely as possible.  Ideally you should
+also list the environment in which the code has been tested and certainly list
+any environments where you suspect it might not work.
diff --git a/contrib/mips-msa/linux.c b/contrib/mips-msa/linux.c
new file mode 100644
index 0000000..140215c
--- /dev/null
+++ b/contrib/mips-msa/linux.c
@@ -0,0 +1,64 @@
+/* contrib/mips-msa/linux.c
+ *
+ * Copyright (c) 2016 Glenn Randers-Pehrson
+ * Written by Mandar Sahastrabuddhe, 2016.
+ * Last changed in libpng 1.6.25beta03 [August 29, 2016]
+ *
+ * This code is released under the libpng license.
+ * For conditions of distribution and use, see the disclaimer
+ * and license in png.h
+ *
+ * SEE contrib/mips-msa/README before reporting bugs
+ *
+ * STATUS: SUPPORTED
+ * BUG REPORTS: png-mng-implement@sourceforge.net
+ *
+ * png_have_msa implemented for Linux by reading the widely available
+ * pseudo-file /proc/cpuinfo.
+ *
+ * This code is strict ANSI-C and is probably moderately portable; it does
+ * however use <stdio.h> and it assumes that /proc/cpuinfo is never localized.
+ */
+
+#include <stdio.h>
+#include <string.h>
+#include <stdlib.h>
+
+static int
+png_have_msa(png_structp png_ptr)
+{
+   FILE *f = fopen("/proc/cpuinfo", "rb");
+
+   char *string = "msa";
+   char word[10];
+
+   if (f != NULL)
+   {
+      while(!feof(f))
+      {
+         int ch = fgetc(f);
+         static int i = 0;
+
+         while(!(ch <= 32))
+         {
+            word[i++] = ch;
+            ch = fgetc(f);
+         }
+
+         int val = strcmp(string, word);
+
+         if (val == 0)
+            return 1;
+
+         i = 0;
+         memset(word, 0, 10);
+      }
+
+      fclose(f);
+   }
+#ifdef PNG_WARNINGS_SUPPORTED
+   else
+      png_warning(png_ptr, "/proc/cpuinfo open failed");
+#endif
+   return 0;
+}
diff --git a/libpng-manual.txt b/libpng-manual.txt
index 98ac4b1..d969f96 100644
--- a/libpng-manual.txt
+++ b/libpng-manual.txt
@@ -1,6 +1,6 @@
 libpng-manual.txt - A description on how to use and modify libpng
 
- libpng version 1.6.24 - August 4, 2016
+ libpng version 1.6.25 - September 1, 2016
  Updated and distributed by Glenn Randers-Pehrson
  <glennrp at users.sourceforge.net>
  Copyright (c) 1998-2016 Glenn Randers-Pehrson
@@ -11,7 +11,7 @@
 
  Based on:
 
- libpng versions 0.97, January 1998, through 1.6.24 - August 4, 2016
+ libpng versions 0.97, January 1998, through 1.6.25 - September 1, 2016
  Updated and distributed by Glenn Randers-Pehrson
  Copyright (c) 1998-2016 Glenn Randers-Pehrson
 
@@ -5350,7 +5350,7 @@
 an official declaration.
 
 This is your unofficial assurance that libpng from version 0.71 and
-upward through 1.6.24 are Y2K compliant.  It is my belief that earlier
+upward through 1.6.25 are Y2K compliant.  It is my belief that earlier
 versions were also Y2K compliant.
 
 Libpng only has two year fields.  One is a 2-byte unsigned integer
diff --git a/libpng.3 b/libpng.3
index 9162df6..4893dc9 100644
--- a/libpng.3
+++ b/libpng.3
@@ -1,6 +1,6 @@
-.TH LIBPNG 3 "August 4, 2016"
+.TH LIBPNG 3 "September 1, 2016"
 .SH NAME
-libpng \- Portable Network Graphics (PNG) Reference Library 1.6.24
+libpng \- Portable Network Graphics (PNG) Reference Library 1.6.25
 .SH SYNOPSIS
 \fB
 #include <png.h>\fP
@@ -510,7 +510,7 @@
 .SH LIBPNG.TXT
 libpng-manual.txt - A description on how to use and modify libpng
 
- libpng version 1.6.24 - August 4, 2016
+ libpng version 1.6.25 - September 1, 2016
  Updated and distributed by Glenn Randers-Pehrson
  <glennrp at users.sourceforge.net>
  Copyright (c) 1998-2016 Glenn Randers-Pehrson
@@ -521,7 +521,7 @@
 
  Based on:
 
- libpng versions 0.97, January 1998, through 1.6.24 - August 4, 2016
+ libpng versions 0.97, January 1998, through 1.6.25 - September 1, 2016
  Updated and distributed by Glenn Randers-Pehrson
  Copyright (c) 1998-2016 Glenn Randers-Pehrson
 
@@ -5860,7 +5860,7 @@
 an official declaration.
 
 This is your unofficial assurance that libpng from version 0.71 and
-upward through 1.6.24 are Y2K compliant.  It is my belief that earlier
+upward through 1.6.25 are Y2K compliant.  It is my belief that earlier
 versions were also Y2K compliant.
 
 Libpng only has two year fields.  One is a 2-byte unsigned integer
@@ -5962,7 +5962,7 @@
  ...
  1.5.27                  15    10527  15.so.15.27[.0]
  ...
- 1.6.24                  16    10624  16.so.16.24[.0]
+ 1.6.25                  16    10625  16.so.16.25[.0]
 
 Henceforth the source version will match the shared-library minor
 and patch numbers; the shared-library major version number will be
@@ -6018,7 +6018,7 @@
 
 Thanks to Frank J. T. Wojcik for helping with the documentation.
 
-Libpng version 1.6.24 - August 4, 2016:
+Libpng version 1.6.25 - September 1, 2016:
 Initially created in 1995 by Guy Eric Schalnat, then of Group 42, Inc.
 Currently maintained by Glenn Randers-Pehrson (glennrp at users.sourceforge.net).
 
@@ -6043,7 +6043,7 @@
 
 This code is released under the libpng license.
 
-libpng versions 1.0.7, July 1, 2000 through 1.6.24, August 4, 2016 are
+libpng versions 1.0.7, July 1, 2000 through 1.6.25, September 1, 2016 are
 Copyright (c) 2000-2002, 2004, 2006-2016 Glenn Randers-Pehrson, are
 derived from libpng-1.0.6, and are distributed according to the same
 disclaimer and license as libpng-1.0.6 with the following individuals
@@ -6168,7 +6168,7 @@
 
 Glenn Randers-Pehrson
 glennrp at users.sourceforge.net
-August 4, 2016
+September 1, 2016
 
 .\" end of man page
 
diff --git a/libpngpf.3 b/libpngpf.3
index 4e8e6ed..3d6eab5 100644
--- a/libpngpf.3
+++ b/libpngpf.3
@@ -1,6 +1,6 @@
-.TH LIBPNGPF 3 "August 4, 2016"
+.TH LIBPNGPF 3 "September 1, 2016"
 .SH NAME
-libpng \- Portable Network Graphics (PNG) Reference Library 1.6.24
+libpng \- Portable Network Graphics (PNG) Reference Library 1.6.25
 (private functions)
 .SH SYNOPSIS
 \fB#include \fI"pngpriv.h"
diff --git a/mips/filter_msa_intrinsics.c b/mips/filter_msa_intrinsics.c
new file mode 100644
index 0000000..943bb3d
--- /dev/null
+++ b/mips/filter_msa_intrinsics.c
@@ -0,0 +1,807 @@
+
+/* filter_msa_intrinsics.c - MSA optimised filter functions
+ *
+ * Copyright (c) 2016 Glenn Randers-Pehrson
+ * Written by Mandar Sahastrabuddhe, August 2016.
+ * Last changed in libpng 1.6.25 [September 1, 2016]
+ *
+ * This code is released under the libpng license.
+ * For conditions of distribution and use, see the disclaimer
+ * and license in png.h
+ */
+#include <stdio.h>
+#include <stdint.h>
+#include "../pngpriv.h"
+
+#ifdef PNG_READ_SUPPORTED
+
+/* This code requires -mfpu=msa on the command line: */
+#if PNG_MIPS_MSA_IMPLEMENTATION == 1 /* intrinsics code from pngpriv.h */
+
+#include <msa.h>
+
+/* libpng row pointers are not necessarily aligned to any particular boundary,
+ * however this code will only work with appropriate alignment. mips/mips_init.c
+ * checks for this (and will not compile unless it is done). This code uses
+ * variants of png_aligncast to avoid compiler warnings.
+ */
+#define png_ptr(type,pointer) png_aligncast(type *,pointer)
+#define png_ptrc(type,pointer) png_aligncastconst(const type *,pointer)
+
+/* The following relies on a variable 'temp_pointer' being declared with type
+ * 'type'.  This is written this way just to hide the GCC strict aliasing
+ * warning; note that the code is safe because there never is an alias between
+ * the input and output pointers.
+ */
+#define png_ldr(type,pointer)\
+   (temp_pointer = png_ptr(type,pointer), *temp_pointer)
+
+#if PNG_MIPS_MSA_OPT > 0
+
+#ifdef CLANG_BUILD
+   #define MSA_SRLI_B(a, b)   __msa_srli_b((v16i8) a, b)
+
+   #define LW(psrc)                              \
+   ( {                                           \
+       uint8_t *psrc_lw_m = (uint8_t *) (psrc);  \
+       uint32_t val_m;                           \
+                                                 \
+       asm volatile (                            \
+           "lw  %[val_m],  %[psrc_lw_m]  \n\t"   \
+                                                 \
+           : [val_m] "=r" (val_m)                \
+           : [psrc_lw_m] "m" (*psrc_lw_m)        \
+       );                                        \
+                                                 \
+       val_m;                                    \
+   } )
+
+   #define SH(val, pdst)                         \
+   {                                             \
+       uint8_t *pdst_sh_m = (uint8_t *) (pdst);  \
+       uint16_t val_m = (val);                   \
+                                                 \
+       asm volatile (                            \
+           "sh  %[val_m],  %[pdst_sh_m]  \n\t"   \
+                                                 \
+           : [pdst_sh_m] "=m" (*pdst_sh_m)       \
+           : [val_m] "r" (val_m)                 \
+       );                                        \
+   }
+
+   #define SW(val, pdst)                         \
+   {                                             \
+       uint8_t *pdst_sw_m = (uint8_t *) (pdst);  \
+       uint32_t val_m = (val);                   \
+                                                 \
+       asm volatile (                            \
+           "sw  %[val_m],  %[pdst_sw_m]  \n\t"   \
+                                                 \
+           : [pdst_sw_m] "=m" (*pdst_sw_m)       \
+           : [val_m] "r" (val_m)                 \
+       );                                        \
+   }
+
+       #if (__mips == 64)
+        #define SD(val, pdst)                         \
+        {                                             \
+            uint8_t *pdst_sd_m = (uint8_t *) (pdst);  \
+            uint64_t val_m = (val);                   \
+                                                      \
+            asm volatile (                            \
+                "sd  %[val_m],  %[pdst_sd_m]  \n\t"   \
+                                                      \
+                : [pdst_sd_m] "=m" (*pdst_sd_m)       \
+                : [val_m] "r" (val_m)                 \
+            );                                        \
+        }
+    #else
+        #define SD(val, pdst)                                          \
+        {                                                              \
+            uint8_t *pdst_sd_m = (uint8_t *) (pdst);                   \
+            uint32_t val0_m, val1_m;                                   \
+                                                                       \
+            val0_m = (uint32_t) ((val) & 0x00000000FFFFFFFF);          \
+            val1_m = (uint32_t) (((val) >> 32) & 0x00000000FFFFFFFF);  \
+                                                                       \
+            SW(val0_m, pdst_sd_m);                                     \
+            SW(val1_m, pdst_sd_m + 4);                                 \
+        }
+    #endif
+#else
+   #define MSA_SRLI_B(a, b)   (a >> b)
+
+#if (__mips_isa_rev >= 6)
+   #define LW(psrc)                              \
+   ( {                                           \
+       uint8_t *psrc_lw_m = (uint8_t *) (psrc);  \
+       uint32_t val_m;                           \
+                                                 \
+       asm volatile (                            \
+           "lw  %[val_m],  %[psrc_lw_m]  \n\t"   \
+                                                 \
+           : [val_m] "=r" (val_m)                \
+           : [psrc_lw_m] "m" (*psrc_lw_m)        \
+       );                                        \
+                                                 \
+       val_m;                                    \
+   } )
+
+   #define SH(val, pdst)                         \
+   {                                             \
+       uint8_t *pdst_sh_m = (uint8_t *) (pdst);  \
+       uint16_t val_m = (val);                   \
+                                                 \
+       asm volatile (                            \
+           "sh  %[val_m],  %[pdst_sh_m]  \n\t"   \
+                                                 \
+           : [pdst_sh_m] "=m" (*pdst_sh_m)       \
+           : [val_m] "r" (val_m)                 \
+       );                                        \
+   }
+
+   #define SW(val, pdst)                         \
+   {                                             \
+       uint8_t *pdst_sw_m = (uint8_t *) (pdst);  \
+       uint32_t val_m = (val);                   \
+                                                 \
+       asm volatile (                            \
+           "sw  %[val_m],  %[pdst_sw_m]  \n\t"   \
+                                                 \
+           : [pdst_sw_m] "=m" (*pdst_sw_m)       \
+           : [val_m] "r" (val_m)                 \
+       );                                        \
+   }
+
+   #if (__mips == 64)
+        #define SD(val, pdst)                         \
+        {                                             \
+            uint8_t *pdst_sd_m = (uint8_t *) (pdst);  \
+            uint64_t val_m = (val);                   \
+                                                      \
+            asm volatile (                            \
+                "sd  %[val_m],  %[pdst_sd_m]  \n\t"   \
+                                                      \
+                : [pdst_sd_m] "=m" (*pdst_sd_m)       \
+                : [val_m] "r" (val_m)                 \
+            );                                        \
+        }
+    #else
+        #define SD(val, pdst)                                          \
+        {                                                              \
+            uint8_t *pdst_sd_m = (uint8_t *) (pdst);                   \
+            uint32_t val0_m, val1_m;                                   \
+                                                                       \
+            val0_m = (uint32_t) ((val) & 0x00000000FFFFFFFF);          \
+            val1_m = (uint32_t) (((val) >> 32) & 0x00000000FFFFFFFF);  \
+                                                                       \
+            SW(val0_m, pdst_sd_m);                                     \
+            SW(val1_m, pdst_sd_m + 4);                                 \
+        }
+    #endif
+#else  // !(__mips_isa_rev >= 6)
+   #define LW(psrc)                              \
+   ( {                                           \
+       uint8_t *psrc_lw_m = (uint8_t *) (psrc);  \
+       uint32_t val_m;                           \
+                                                 \
+       asm volatile (                            \
+           "ulw  %[val_m],  %[psrc_lw_m]  \n\t"  \
+                                                 \
+           : [val_m] "=r" (val_m)                \
+           : [psrc_lw_m] "m" (*psrc_lw_m)        \
+       );                                        \
+                                                 \
+       val_m;                                    \
+   } )
+
+   #define SH(val, pdst)                         \
+   {                                             \
+       uint8_t *pdst_sh_m = (uint8_t *) (pdst);  \
+       uint16_t val_m = (val);                   \
+                                                 \
+       asm volatile (                            \
+           "ush  %[val_m],  %[pdst_sh_m]  \n\t"  \
+                                                 \
+           : [pdst_sh_m] "=m" (*pdst_sh_m)       \
+           : [val_m] "r" (val_m)                 \
+       );                                        \
+   }
+
+   #define SW(val, pdst)                         \
+   {                                             \
+       uint8_t *pdst_sw_m = (uint8_t *) (pdst);  \
+       uint32_t val_m = (val);                   \
+                                                 \
+       asm volatile (                            \
+           "usw  %[val_m],  %[pdst_sw_m]  \n\t"  \
+                                                 \
+           : [pdst_sw_m] "=m" (*pdst_sw_m)       \
+           : [val_m] "r" (val_m)                 \
+       );                                        \
+   }
+
+   #define SD(val, pdst)                                          \
+    {                                                              \
+        uint8_t *pdst_sd_m = (uint8_t *) (pdst);                   \
+        uint32_t val0_m, val1_m;                                   \
+                                                                   \
+        val0_m = (uint32_t) ((val) & 0x00000000FFFFFFFF);          \
+        val1_m = (uint32_t) (((val) >> 32) & 0x00000000FFFFFFFF);  \
+                                                                   \
+        SW(val0_m, pdst_sd_m);                                     \
+        SW(val1_m, pdst_sd_m + 4);                                 \
+    }
+
+    #define SW_ZERO(pdst)                      \
+    {                                          \
+        uint8_t *pdst_m = (uint8_t *) (pdst);  \
+                                               \
+        asm volatile (                         \
+            "usw  $0,  %[pdst_m]  \n\t"        \
+                                               \
+            : [pdst_m] "=m" (*pdst_m)          \
+            :                                  \
+        );                                     \
+    }
+#endif  // (__mips_isa_rev >= 6)
+#endif
+
+#define LD_B(RTYPE, psrc) *((RTYPE *) (psrc))
+#define LD_UB(...) LD_B(v16u8, __VA_ARGS__)
+#define LD_B2(RTYPE, psrc, stride, out0, out1)  \
+{                                               \
+    out0 = LD_B(RTYPE, (psrc));                 \
+    out1 = LD_B(RTYPE, (psrc) + stride);        \
+}
+#define LD_UB2(...) LD_B2(v16u8, __VA_ARGS__)
+#define LD_B4(RTYPE, psrc, stride, out0, out1, out2, out3)   \
+{                                                            \
+    LD_B2(RTYPE, (psrc), stride, out0, out1);                \
+    LD_B2(RTYPE, (psrc) + 2 * stride , stride, out2, out3);  \
+}
+#define LD_UB4(...) LD_B4(v16u8, __VA_ARGS__)
+
+#define ST_B(RTYPE, in, pdst) *((RTYPE *) (pdst)) = (in)
+#define ST_UB(...) ST_B(v16u8, __VA_ARGS__)
+#define ST_B2(RTYPE, in0, in1, pdst, stride)  \
+{                                             \
+    ST_B(RTYPE, in0, (pdst));                 \
+    ST_B(RTYPE, in1, (pdst) + stride);        \
+}
+#define ST_UB2(...) ST_B2(v16u8, __VA_ARGS__)
+#define ST_B4(RTYPE, in0, in1, in2, in3, pdst, stride)    \
+{                                                         \
+    ST_B2(RTYPE, in0, in1, (pdst), stride);               \
+    ST_B2(RTYPE, in2, in3, (pdst) + 2 * stride, stride);  \
+}
+#define ST_UB4(...) ST_B4(v16u8, __VA_ARGS__)
+
+#define ADD2(in0, in1, in2, in3, out0, out1)  \
+{                                             \
+    out0 = in0 + in1;                         \
+    out1 = in2 + in3;                         \
+}
+#define ADD3(in0, in1, in2, in3, in4, in5,  \
+             out0, out1, out2)              \
+{                                           \
+    ADD2(in0, in1, in2, in3, out0, out1);   \
+    out2 = in4 + in5;                       \
+}
+#define ADD4(in0, in1, in2, in3, in4, in5, in6, in7,  \
+             out0, out1, out2, out3)                  \
+{                                                     \
+    ADD2(in0, in1, in2, in3, out0, out1);             \
+    ADD2(in4, in5, in6, in7, out2, out3);             \
+}
+
+#define ILVR_B2(RTYPE, in0, in1, in2, in3, out0, out1)      \
+{                                                           \
+    out0 = (RTYPE) __msa_ilvr_b((v16i8) in0, (v16i8) in1);  \
+    out1 = (RTYPE) __msa_ilvr_b((v16i8) in2, (v16i8) in3);  \
+}
+#define ILVR_B2_SH(...) ILVR_B2(v8i16, __VA_ARGS__)
+
+#define HSUB_UB2(RTYPE, in0, in1, out0, out1)                 \
+{                                                             \
+    out0 = (RTYPE) __msa_hsub_u_h((v16u8) in0, (v16u8) in0);  \
+    out1 = (RTYPE) __msa_hsub_u_h((v16u8) in1, (v16u8) in1);  \
+}
+#define HSUB_UB2_SH(...) HSUB_UB2(v8i16, __VA_ARGS__)
+
+#define SLDI_B2_0(RTYPE, in0, in1, out0, out1, slide_val)                 \
+{                                                                         \
+    v16i8 zero_m = { 0 };                                                 \
+    out0 = (RTYPE) __msa_sldi_b((v16i8) zero_m, (v16i8) in0, slide_val);  \
+    out1 = (RTYPE) __msa_sldi_b((v16i8) zero_m, (v16i8) in1, slide_val);  \
+}
+#define SLDI_B2_0_UB(...) SLDI_B2_0(v16u8, __VA_ARGS__)
+
+#define SLDI_B3_0(RTYPE, in0, in1, in2, out0, out1, out2,  slide_val)     \
+{                                                                         \
+    v16i8 zero_m = { 0 };                                                 \
+    SLDI_B2_0(RTYPE, in0, in1, out0, out1, slide_val);                    \
+    out2 = (RTYPE) __msa_sldi_b((v16i8) zero_m, (v16i8) in2, slide_val);  \
+}
+#define SLDI_B3_0_UB(...) SLDI_B3_0(v16u8, __VA_ARGS__)
+
+#define ILVEV_W2(RTYPE, in0, in1, in2, in3, out0, out1)      \
+{                                                            \
+    out0 = (RTYPE) __msa_ilvev_w((v4i32) in1, (v4i32) in0);  \
+    out1 = (RTYPE) __msa_ilvev_w((v4i32) in3, (v4i32) in2);  \
+}
+#define ILVEV_W2_UB(...) ILVEV_W2(v16u8, __VA_ARGS__)
+
+#define ADD_ABS_H3(RTYPE, in0, in1, in2, out0, out1, out2)  \
+{                                                           \
+    RTYPE zero = {0};                                       \
+                                                            \
+    out0 = __msa_add_a_h((v8i16) zero, in0);                \
+    out1 = __msa_add_a_h((v8i16) zero, in1);                \
+    out2 = __msa_add_a_h((v8i16) zero, in2);                \
+}
+#define ADD_ABS_H3_SH(...) ADD_ABS_H3(v8i16, __VA_ARGS__)
+
+#define VSHF_B2(RTYPE, in0, in1, in2, in3, mask0, mask1, out0, out1)       \
+{                                                                          \
+    out0 = (RTYPE) __msa_vshf_b((v16i8) mask0, (v16i8) in1, (v16i8) in0);  \
+    out1 = (RTYPE) __msa_vshf_b((v16i8) mask1, (v16i8) in3, (v16i8) in2);  \
+}
+#define VSHF_B2_UB(...) VSHF_B2(v16u8, __VA_ARGS__)
+
+#define CMP_AND_SELECT(inp0, inp1, inp2, inp3, inp4, inp5, out0)              \
+{                                                                             \
+   v8i16 _sel_h0, _sel_h1;                                                    \
+   v16u8 _sel_b0, _sel_b1;                                                    \
+   _sel_h0 = (v8i16) __msa_clt_u_h((v8u16) inp1, (v8u16) inp0);               \
+   _sel_b0 = (v16u8) __msa_pckev_b((v16i8) _sel_h0, (v16i8) _sel_h0);         \
+   inp0 = (v8i16) __msa_bmnz_v((v16u8) inp0, (v16u8) inp1, (v16u8) _sel_h0);  \
+   inp4 = (v16u8) __msa_bmnz_v(inp3, inp4, _sel_b0);                          \
+   _sel_h1 = (v8i16) __msa_clt_u_h((v8u16) inp2, (v8u16) inp0);               \
+   _sel_b1 = (v16u8) __msa_pckev_b((v16i8) _sel_h1, (v16i8) _sel_h1);         \
+   inp4 = (v16u8) __msa_bmnz_v(inp4, inp5, _sel_b1);                          \
+   out0 += inp4;                                                              \
+}
+
+void png_read_filter_row_up_msa(png_row_infop row_info, png_bytep row,
+                                png_const_bytep prev_row)
+{
+   png_size_t i, cnt, cnt16, cnt32;
+   png_size_t istop = row_info->rowbytes;
+   png_bytep rp = row;
+   png_const_bytep pp = prev_row;
+   v16u8 src0, src1, src2, src3, src4, src5, src6, src7;
+
+   for (i = 0; i < (istop >> 6); i++)
+   {
+      LD_UB4(rp, 16, src0, src1, src2, src3);
+      LD_UB4(pp, 16, src4, src5, src6, src7);
+      pp += 64;
+
+	  ADD4(src0, src4, src1, src5, src2, src6, src3, src7,
+	       src0, src1, src2, src3);
+
+      ST_UB4(src0, src1, src2, src3, rp, 16);
+      rp += 64;
+   }
+
+   if (istop & 0x3F)
+   {
+      cnt32 = istop & 0x20;
+      cnt16 = istop & 0x10;
+      cnt = istop & 0xF;
+
+      if(cnt32)
+      {
+         if (cnt16 && cnt)
+         {
+            LD_UB4(rp, 16, src0, src1, src2, src3);
+            LD_UB4(pp, 16, src4, src5, src6, src7);
+
+            ADD4(src0, src4, src1, src5, src2, src6, src3, src7,
+	             src0, src1, src2, src3);
+
+            ST_UB4(src0, src1, src2, src3, rp, 16);
+            rp += 64;
+         }
+         else if (cnt16 || cnt)
+         {
+            LD_UB2(rp, 16, src0, src1);
+            LD_UB2(pp, 16, src4, src5);
+            pp += 32;
+            src2 = LD_UB(rp + 32);
+            src6 = LD_UB(pp);
+
+            ADD3(src0, src4, src1, src5, src2, src6, src0, src1, src2);
+
+            ST_UB2(src0, src1, rp, 16);
+            rp += 32;
+            ST_UB(src2, rp);
+            rp += 16;
+         }
+         else
+         {
+            LD_UB2(rp, 16, src0, src1);
+            LD_UB2(pp, 16, src4, src5);
+
+			ADD2(src0, src4, src1, src5, src0, src1);
+
+            ST_UB2(src0, src1, rp, 16);
+            rp += 32;
+         }
+      }
+      else if (cnt16 && cnt)
+      {
+         LD_UB2(rp, 16, src0, src1);
+         LD_UB2(pp, 16, src4, src5);
+
+         ADD2(src0, src4, src1, src5, src0, src1);
+
+         ST_UB2(src0, src1, rp, 16);
+         rp += 32;
+      }
+      else if (cnt16 || cnt)
+      {
+         src0 = LD_UB(rp);
+         src4 = LD_UB(pp);
+         pp += 16;
+
+         src0 += src4;
+
+         ST_UB(src0, rp);
+         rp += 16;
+      }
+   }
+}
+
+void png_read_filter_row_sub4_msa(png_row_infop row_info, png_bytep row,
+                                  png_const_bytep prev_row)
+{
+   png_size_t count;
+   png_size_t istop = row_info->rowbytes;
+   png_bytep src = row;
+   png_bytep nxt = row + 4;
+   int32_t inp0;
+   v16u8 src0, src1, src2, src3, src4;
+   v16u8 dst0, dst1;
+   v16u8 zero = { 0 };
+
+   istop -= 4;
+
+   inp0 = LW(src);
+   src += 4;
+   src0 = (v16u8) __msa_insert_w((v4i32) zero, 0, inp0);
+
+   for (count = 0; count < istop; count += 16)
+   {
+      src1 = LD_UB(src);
+      src += 16;
+
+      src2 = (v16u8) __msa_sldi_b((v16i8) zero, (v16i8) src1, 4);
+      src3 = (v16u8) __msa_sldi_b((v16i8) zero, (v16i8) src1, 8);
+      src4 = (v16u8) __msa_sldi_b((v16i8) zero, (v16i8) src1, 12);
+      src1 += src0;
+      src2 += src1;
+      src3 += src2;
+      src4 += src3;
+      src0 = src4;
+      ILVEV_W2_UB(src1, src2, src3, src4, dst0, dst1);
+      dst0 = (v16u8) __msa_pckev_d((v2i64) dst1, (v2i64) dst0);
+
+      ST_UB(dst0, nxt);
+      nxt += 16;
+   }
+}
+
+void png_read_filter_row_sub3_msa(png_row_infop row_info, png_bytep row,
+                                  png_const_bytep prev_row)
+{
+   png_size_t count;
+   png_size_t istop = row_info->rowbytes;
+   png_bytep src = row;
+   png_bytep nxt = row + 3;
+   int64_t out0;
+   int32_t inp0, out1;
+   v16u8 src0, src1, src2, src3, src4, dst0, dst1;
+   v16u8 zero = { 0 };
+   v16i8 mask0 = { 0, 1, 2, 16, 17, 18, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 };
+   v16i8 mask1 = { 0, 1, 2, 3, 4, 5, 16, 17, 18, 19, 20, 21, 0, 0, 0, 0 };
+
+   istop -= 3;
+
+   inp0 = LW(src);
+   src += 3;
+   src0 = (v16u8) __msa_insert_w((v4i32) zero, 0, inp0);
+
+   for (count = 0; count < istop; count += 12)
+   {
+      src1 = LD_UB(src);
+      src += 12;
+
+      src2 = (v16u8) __msa_sldi_b((v16i8) zero, (v16i8) src1, 3);
+      src3 = (v16u8) __msa_sldi_b((v16i8) zero, (v16i8) src1, 6);
+      src4 = (v16u8) __msa_sldi_b((v16i8) zero, (v16i8) src1, 9);
+      src1 += src0;
+      src2 += src1;
+      src3 += src2;
+      src4 += src3;
+      src0 = src4;
+      VSHF_B2_UB(src1, src2, src3, src4, mask0, mask0, dst0, dst1);
+      dst0 = (v16u8) __msa_vshf_b(mask1, (v16i8) dst1, (v16i8) dst0);
+      out0 = __msa_copy_s_d((v2i64) dst0, 0);
+      out1 = __msa_copy_s_w((v4i32) dst0, 2);
+
+      SD(out0, nxt);
+      nxt += 8;
+      SW(out1, nxt);
+      nxt += 4;
+   }
+}
+
+void png_read_filter_row_avg4_msa(png_row_infop row_info, png_bytep row,
+                                  png_const_bytep prev_row)
+{
+   png_size_t i;
+   png_bytep src = row;
+   png_bytep nxt = row;
+   png_const_bytep pp = prev_row;
+   png_size_t istop = row_info->rowbytes - 4;
+   int32_t inp0, inp1, out0;
+   v16u8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, dst0, dst1;
+   v16u8 zero = { 0 };
+
+   inp0 = LW(pp);
+   pp += 4;
+   inp1 = LW(src);
+   src += 4;
+   src0 = (v16u8) __msa_insert_w((v4i32) zero, 0, inp0);
+   src1 = (v16u8) __msa_insert_w((v4i32) zero, 0, inp1);
+   src0 = (v16u8) MSA_SRLI_B(src0, 1);
+   src1 += src0;
+   out0 = __msa_copy_s_w((v4i32) src1, 0);
+   SW(out0, nxt);
+   nxt += 4;
+
+   for (i = 0; i < istop; i += 16)
+   {
+      src2 = LD_UB(pp);
+      pp += 16;
+      src6 = LD_UB(src);
+      src += 16;
+
+      SLDI_B2_0_UB(src2, src6, src3, src7, 4);
+      SLDI_B2_0_UB(src2, src6, src4, src8, 8);
+      SLDI_B2_0_UB(src2, src6, src5, src9, 12);
+      src2 = __msa_ave_u_b(src2, src1);
+      src6 += src2;
+      src3 = __msa_ave_u_b(src3, src6);
+      src7 += src3;
+      src4 = __msa_ave_u_b(src4, src7);
+      src8 += src4;
+      src5 = __msa_ave_u_b(src5, src8);
+      src9 += src5;
+      src1 = src9;
+      ILVEV_W2_UB(src6, src7, src8, src9, dst0, dst1);
+      dst0 = (v16u8) __msa_pckev_d((v2i64) dst1, (v2i64) dst0);
+
+      ST_UB(dst0, nxt);
+      nxt += 16;
+   }
+}
+
+void png_read_filter_row_avg3_msa(png_row_infop row_info, png_bytep row,
+                                  png_const_bytep prev_row)
+{
+   png_size_t i;
+   png_bytep src = row;
+   png_bytep nxt = row;
+   png_const_bytep pp = prev_row;
+   png_size_t istop = row_info->rowbytes - 3;
+   int64_t out0;
+   int32_t inp0, inp1, out1;
+   int16_t out2;
+   v16u8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, dst0, dst1;
+   v16u8 zero = { 0 };
+   v16i8 mask0 = { 0, 1, 2, 16, 17, 18, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 };
+   v16i8 mask1 = { 0, 1, 2, 3, 4, 5, 16, 17, 18, 19, 20, 21, 0, 0, 0, 0 };
+
+   inp0 = LW(pp);
+   pp += 3;
+   inp1 = LW(src);
+   src += 3;
+   src0 = (v16u8) __msa_insert_w((v4i32) zero, 0, inp0);
+   src1 = (v16u8) __msa_insert_w((v4i32) zero, 0, inp1);
+   src0 = (v16u8) MSA_SRLI_B(src0, 1);
+   src1 += src0;
+   out2 = __msa_copy_s_h((v8i16) src1, 0);
+   SH(out2, nxt);
+   nxt += 2;
+   nxt[0] = src1[2];
+   nxt++;
+
+   for (i = 0; i < istop; i += 12)
+   {
+      src2 = LD_UB(pp);
+      pp += 12;
+      src6 = LD_UB(src);
+      src += 12;
+
+      SLDI_B2_0_UB(src2, src6, src3, src7, 3);
+      SLDI_B2_0_UB(src2, src6, src4, src8, 6);
+      SLDI_B2_0_UB(src2, src6, src5, src9, 9);
+      src2 = __msa_ave_u_b(src2, src1);
+      src6 += src2;
+      src3 = __msa_ave_u_b(src3, src6);
+      src7 += src3;
+      src4 = __msa_ave_u_b(src4, src7);
+      src8 += src4;
+      src5 = __msa_ave_u_b(src5, src8);
+      src9 += src5;
+      src1 = src9;
+      VSHF_B2_UB(src6, src7, src8, src9, mask0, mask0, dst0, dst1);
+      dst0 = (v16u8) __msa_vshf_b(mask1, (v16i8) dst1, (v16i8) dst0);
+      out0 = __msa_copy_s_d((v2i64) dst0, 0);
+      out1 = __msa_copy_s_w((v4i32) dst0, 2);
+
+      SD(out0, nxt);
+      nxt += 8;
+      SW(out1, nxt);
+      nxt += 4;
+   }
+}
+
+void png_read_filter_row_paeth4_msa(png_row_infop row_info,
+                                    png_bytep row,
+                                    png_const_bytep prev_row)
+{
+   int32_t count, rp_end;
+   png_bytep nxt;
+   png_const_bytep prev_nxt;
+   int32_t inp0, inp1, res0;
+   v16u8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9;
+   v16u8 src10, src11, src12, src13, dst0, dst1;
+   v8i16 vec0, vec1, vec2;
+   v16u8 zero = { 0 };
+
+   nxt = row;
+   prev_nxt = prev_row;
+
+   inp0 = LW(nxt);
+   inp1 = LW(prev_nxt);
+   prev_nxt += 4;
+   src0 = (v16u8) __msa_insert_w((v4i32) zero, 0, inp0);
+   src1 = (v16u8) __msa_insert_w((v4i32) zero, 0, inp1);
+
+   src1 += src0;
+   res0 = __msa_copy_s_w((v4i32) src1, 0);
+
+   SW(res0, nxt);
+   nxt += 4;
+
+   /* Remainder */
+   rp_end = row_info->rowbytes - 4;
+
+   for (count = 0; count < rp_end; count += 16)
+   {
+      src2 = LD_UB(prev_nxt);
+      prev_nxt += 16;
+      src6 = LD_UB(prev_row);
+      prev_row += 16;
+      src10 = LD_UB(nxt);
+
+      SLDI_B3_0_UB(src2, src6, src10, src3, src7, src11, 4);
+      SLDI_B3_0_UB(src2, src6, src10, src4, src8, src12, 8);
+      SLDI_B3_0_UB(src2, src6, src10, src5, src9, src13, 12);
+      ILVR_B2_SH(src2, src6, src1, src6, vec0, vec1);
+      HSUB_UB2_SH(vec0, vec1, vec0, vec1);
+      vec2 = vec0 + vec1;
+      ADD_ABS_H3_SH(vec0, vec1, vec2, vec0, vec1, vec2);
+      CMP_AND_SELECT(vec0, vec1, vec2, src1, src2, src6, src10);
+      ILVR_B2_SH(src3, src7, src10, src7, vec0, vec1);
+      HSUB_UB2_SH(vec0, vec1, vec0, vec1);
+      vec2 = vec0 + vec1;
+      ADD_ABS_H3_SH(vec0, vec1, vec2, vec0, vec1, vec2);
+      CMP_AND_SELECT(vec0, vec1, vec2, src10, src3, src7, src11);
+      ILVR_B2_SH(src4, src8, src11, src8, vec0, vec1);
+      HSUB_UB2_SH(vec0, vec1, vec0, vec1);
+      vec2 = vec0 + vec1;
+      ADD_ABS_H3_SH(vec0, vec1, vec2, vec0, vec1, vec2);
+      CMP_AND_SELECT(vec0, vec1, vec2, src11, src4, src8, src12);
+      ILVR_B2_SH(src5, src9, src12, src9, vec0, vec1);
+      HSUB_UB2_SH(vec0, vec1, vec0, vec1);
+      vec2 = vec0 + vec1;
+      ADD_ABS_H3_SH(vec0, vec1, vec2, vec0, vec1, vec2);
+      CMP_AND_SELECT(vec0, vec1, vec2, src12, src5, src9, src13);
+      src1 = src13;
+      ILVEV_W2_UB(src10, src11, src12, src1, dst0, dst1);
+      dst0 = (v16u8) __msa_pckev_d((v2i64) dst1, (v2i64) dst0);
+
+      ST_UB(dst0, nxt);
+      nxt += 16;
+   }
+}
+
+void png_read_filter_row_paeth3_msa(png_row_infop row_info,
+                                    png_bytep row,
+                                    png_const_bytep prev_row)
+{
+   int32_t count, rp_end;
+   png_bytep nxt;
+   png_const_bytep prev_nxt;
+   int64_t out0;
+   int32_t inp0, inp1, out1;
+   int16_t out2;
+   v16u8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, dst0, dst1;
+   v16u8 src10, src11, src12, src13;
+   v8i16 vec0, vec1, vec2;
+   v16u8 zero = { 0 };
+   v16i8 mask0 = { 0, 1, 2, 16, 17, 18, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 };
+   v16i8 mask1 = { 0, 1, 2, 3, 4, 5, 16, 17, 18, 19, 20, 21, 0, 0, 0, 0 };
+
+   nxt = row;
+   prev_nxt = prev_row;
+
+   inp0 = LW(nxt);
+   inp1 = LW(prev_nxt);
+   prev_nxt += 3;
+   src0 = (v16u8) __msa_insert_w((v4i32) zero, 0, inp0);
+   src1 = (v16u8) __msa_insert_w((v4i32) zero, 0, inp1);
+
+   src1 += src0;
+   out2 = __msa_copy_s_h((v8i16) src1, 0);
+
+   SH(out2, nxt);
+   nxt += 2;
+   nxt[0] = src1[2];
+   nxt++;
+
+   /* Remainder */
+   rp_end = row_info->rowbytes - 3;
+
+   for (count = 0; count < rp_end; count += 12)
+   {
+      src2 = LD_UB(prev_nxt);
+      prev_nxt += 12;
+      src6 = LD_UB(prev_row);
+      prev_row += 12;
+      src10 = LD_UB(nxt);
+
+      SLDI_B3_0_UB(src2, src6, src10, src3, src7, src11, 3);
+      SLDI_B3_0_UB(src2, src6, src10, src4, src8, src12, 6);
+      SLDI_B3_0_UB(src2, src6, src10, src5, src9, src13, 9);
+      ILVR_B2_SH(src2, src6, src1, src6, vec0, vec1);
+      HSUB_UB2_SH(vec0, vec1, vec0, vec1);
+      vec2 = vec0 + vec1;
+      ADD_ABS_H3_SH(vec0, vec1, vec2, vec0, vec1, vec2);
+      CMP_AND_SELECT(vec0, vec1, vec2, src1, src2, src6, src10);
+      ILVR_B2_SH(src3, src7, src10, src7, vec0, vec1);
+      HSUB_UB2_SH(vec0, vec1, vec0, vec1);
+      vec2 = vec0 + vec1;
+      ADD_ABS_H3_SH(vec0, vec1, vec2, vec0, vec1, vec2);
+      CMP_AND_SELECT(vec0, vec1, vec2, src10, src3, src7, src11);
+      ILVR_B2_SH(src4, src8, src11, src8, vec0, vec1);
+      HSUB_UB2_SH(vec0, vec1, vec0, vec1);
+      vec2 = vec0 + vec1;
+      ADD_ABS_H3_SH(vec0, vec1, vec2, vec0, vec1, vec2);
+      CMP_AND_SELECT(vec0, vec1, vec2, src11, src4, src8, src12);
+      ILVR_B2_SH(src5, src9, src12, src9, vec0, vec1);
+      HSUB_UB2_SH(vec0, vec1, vec0, vec1);
+      vec2 = vec0 + vec1;
+      ADD_ABS_H3_SH(vec0, vec1, vec2, vec0, vec1, vec2);
+      CMP_AND_SELECT(vec0, vec1, vec2, src12, src5, src9, src13);
+      src1 = src13;
+      VSHF_B2_UB(src10, src11, src12, src13, mask0, mask0, dst0, dst1);
+      dst0 = (v16u8) __msa_vshf_b(mask1, (v16i8) dst1, (v16i8) dst0);
+      out0 = __msa_copy_s_d((v2i64) dst0, 0);
+      out1 = __msa_copy_s_w((v4i32) dst0, 2);
+
+      SD(out0, nxt);
+      nxt += 8;
+      SW(out1, nxt);
+      nxt += 4;
+   }
+}
+
+#endif /* PNG_MIPS_MSA_OPT > 0 */
+#endif /* PNG_MIPS_MSA_IMPLEMENTATION == 1 (intrinsics) */
+#endif /* READ */
diff --git a/mips/mips_init.c b/mips/mips_init.c
new file mode 100644
index 0000000..0bfb7a3
--- /dev/null
+++ b/mips/mips_init.c
@@ -0,0 +1,129 @@
+
+/* mips_init.c - MSA optimised filter functions
+ *
+ * Copyright (c) 2016 Glenn Randers-Pehrson
+ * Written by Mandar Sahastrabuddhe, 2016.
+ * Last changed in libpng 1.6.25 [September 1, 2016]
+ *
+ * This code is released under the libpng license.
+ * For conditions of distribution and use, see the disclaimer
+ * and license in png.h
+ */
+/* Below, after checking __linux__, various non-C90 POSIX 1003.1 functions are
+ * called.
+ */
+#define _POSIX_SOURCE 1
+
+#include <stdio.h>
+#include "../pngpriv.h"
+
+#ifdef PNG_READ_SUPPORTED
+
+#if PNG_MIPS_MSA_OPT > 0
+#ifdef PNG_MIPS_MSA_CHECK_SUPPORTED /* Do run-time checks */
+/* WARNING: it is strongly recommended that you do not build libpng with
+ * run-time checks for CPU features if at all possible.  In the case of the MIPS
+ * MSA instructions there is no processor-specific way of detecting the
+ * presence of the required support, therefore run-time detection is extremely
+ * OS specific.
+ *
+ * You may set the macro PNG_MIPS_MSA_FILE to the file name of file containing
+ * a fragment of C source code which defines the png_have_msa function.  There
+ * are a number of implementations in contrib/mips-msa, but the only one that
+ * has partial support is contrib/mips-msa/linux.c - a generic Linux
+ * implementation which reads /proc/cpufino.
+ */
+#ifndef PNG_MIPS_MSA_FILE
+#  ifdef __linux__
+#     define PNG_MIPS_MSA_FILE "contrib/mips-msa/linux.c"
+#  endif
+#endif
+
+#ifdef PNG_MIPS_MSA_FILE
+
+#include <signal.h> /* for sig_atomic_t */
+static int png_have_msa(png_structp png_ptr);
+#include PNG_MIPS_MSA_FILE
+
+#else  /* PNG_MIPS_MSA_FILE */
+#  error "PNG_MIPS_MSA_FILE undefined: no support for run-time MIPS MSA checks"
+#endif /* PNG_MIPS_MSA_FILE */
+#endif /* PNG_MIPS_MSA_CHECK_SUPPORTED */
+
+#ifndef PNG_ALIGNED_MEMORY_SUPPORTED
+#  error "ALIGNED_MEMORY is required; set: -DPNG_ALIGNED_MEMORY_SUPPORTED"
+#endif
+
+void
+png_init_filter_functions_msa(png_structp pp, unsigned int bpp)
+{
+   /* The switch statement is compiled in for MIPS_MSA_API, the call to
+    * png_have_msa is compiled in for MIPS_MSA_CHECK. If both are defined
+    * the check is only performed if the API has not set the MSA option on
+    * or off explicitly. In this case the check controls what happens.
+    */
+
+#ifdef PNG_MIPS_MSA_API_SUPPORTED
+   switch ((pp->options >> PNG_MIPS_MSA) & 3)
+   {
+      case PNG_OPTION_UNSET:
+         /* Allow the run-time check to execute if it has been enabled -
+          * thus both API and CHECK can be turned on.  If it isn't supported
+          * this case will fall through to the 'default' below, which just
+          * returns.
+          */
+#endif /* PNG_MIPS_MSA_API_SUPPORTED */
+#ifdef PNG_MIPS_MSA_CHECK_SUPPORTED
+         {
+            static volatile sig_atomic_t no_msa = -1; /* not checked */
+
+            if (no_msa < 0)
+               no_msa = !png_have_msa(pp);
+
+            if (no_msa)
+               return;
+         }
+#ifdef PNG_MIPS_MSA_API_SUPPORTED
+         break;
+#endif
+#endif /* PNG_MIPS_MSA_CHECK_SUPPORTED */
+
+#ifdef PNG_MIPS_MSA_API_SUPPORTED
+      default: /* OFF or INVALID */
+         return;
+
+      case PNG_OPTION_ON:
+         /* Option turned on */
+         break;
+   }
+#endif
+
+   /* IMPORTANT: any new external functions used here must be declared using
+    * PNG_INTERNAL_FUNCTION in ../pngpriv.h.  This is required so that the
+    * 'prefix' option to configure works:
+    *
+    *    ./configure --with-libpng-prefix=foobar_
+    *
+    * Verify you have got this right by running the above command, doing a build
+    * and examining pngprefix.h; it must contain a #define for every external
+    * function you add.  (Notice that this happens automatically for the
+    * initialization function.)
+    */
+   pp->read_filter[PNG_FILTER_VALUE_UP-1] = png_read_filter_row_up_msa;
+
+   if (bpp == 3)
+   {
+      pp->read_filter[PNG_FILTER_VALUE_SUB-1] = png_read_filter_row_sub3_msa;
+      pp->read_filter[PNG_FILTER_VALUE_AVG-1] = png_read_filter_row_avg3_msa;
+      pp->read_filter[PNG_FILTER_VALUE_PAETH-1] = png_read_filter_row_paeth3_msa;
+   }
+
+   else if (bpp == 4)
+   {
+      pp->read_filter[PNG_FILTER_VALUE_SUB-1] = png_read_filter_row_sub4_msa;
+      pp->read_filter[PNG_FILTER_VALUE_AVG-1] = png_read_filter_row_avg4_msa;
+      pp->read_filter[PNG_FILTER_VALUE_PAETH-1] = png_read_filter_row_paeth4_msa;
+   }
+}
+#endif /* PNG_MIPS_MSA_OPT > 0 */
+#endif /* READ */
diff --git a/png.5 b/png.5
index 0df6f9a..16d213c 100644
--- a/png.5
+++ b/png.5
@@ -1,4 +1,4 @@
-.TH PNG 5 "August 4, 2016"
+.TH PNG 5 "September 1, 2016"
 .SH NAME
 png \- Portable Network Graphics (PNG) format
 .SH DESCRIPTION
diff --git a/png.c b/png.c
index 0843ebb..33e8c8f 100644
--- a/png.c
+++ b/png.c
@@ -1,8 +1,8 @@
 
 /* png.c - location for general purpose libpng functions
  *
- * Last changed in libpng 1.6.24 [August 4, 2016]
- * Copyright (c) 1998-2002,2004,2006-2015 Glenn Randers-Pehrson
+ * Last changed in libpng 1.6.25 [September 1, 2016]
+ * Copyright (c) 1998-2002,2004,2006-2016 Glenn Randers-Pehrson
  * (Version 0.96 Copyright (c) 1996, 1997 Andreas Dilger)
  * (Version 0.88 Copyright (c) 1995, 1996 Guy Eric Schalnat, Group 42, Inc.)
  *
@@ -14,7 +14,7 @@
 #include "pngpriv.h"
 
 /* Generate a compiler error if there is an old png.h in the search path. */
-typedef png_libpng_version_1_6_24 Your_png_h_is_not_version_1_6_24;
+typedef png_libpng_version_1_6_25 Your_png_h_is_not_version_1_6_25;
 
 /* Tells libpng that we have already handled the first "num_bytes" bytes
  * of the PNG file signature.  If the PNG data is embedded into another
@@ -775,14 +775,14 @@
 #else
 #  ifdef __STDC__
    return PNG_STRING_NEWLINE \
-      "libpng version 1.6.24 - August 4, 2016" PNG_STRING_NEWLINE \
+      "libpng version 1.6.25 - September 1, 2016" PNG_STRING_NEWLINE \
       "Copyright (c) 1998-2002,2004,2006-2016 Glenn Randers-Pehrson" \
       PNG_STRING_NEWLINE \
       "Copyright (c) 1996-1997 Andreas Dilger" PNG_STRING_NEWLINE \
       "Copyright (c) 1995-1996 Guy Eric Schalnat, Group 42, Inc." \
       PNG_STRING_NEWLINE;
 #  else
-   return "libpng version 1.6.24 - August 4, 2016\
+   return "libpng version 1.6.25 - September 1, 2016\
       Copyright (c) 1998-2002,2004,2006-2016 Glenn Randers-Pehrson\
       Copyright (c) 1996-1997 Andreas Dilger\
       Copyright (c) 1995-1996 Guy Eric Schalnat, Group 42, Inc.";
@@ -1931,8 +1931,8 @@
 static const png_byte D50_nCIEXYZ[12] =
    { 0x00, 0x00, 0xf6, 0xd6, 0x00, 0x01, 0x00, 0x00, 0x00, 0x00, 0xd3, 0x2d };
 
-int /* PRIVATE */
-png_icc_check_length(png_const_structrp png_ptr, png_colorspacerp colorspace,
+static int /* bool */
+icc_check_length(png_const_structrp png_ptr, png_colorspacerp colorspace,
     png_const_charp name, png_uint_32 profile_length)
 {
    if (profile_length < 132)
@@ -1942,6 +1942,40 @@
    return 1;
 }
 
+#ifdef PNG_READ_iCCP_SUPPORTED
+int /* PRIVATE */
+png_icc_check_length(png_const_structrp png_ptr, png_colorspacerp colorspace,
+    png_const_charp name, png_uint_32 profile_length)
+{
+   if (!icc_check_length(png_ptr, colorspace, name, profile_length))
+      return 0;
+
+   /* This needs to be here because the 'normal' check is in
+    * png_decompress_chunk, yet this happens after the attempt to
+    * png_malloc_base the required data.  We only need this on read; on write
+    * the caller supplies the profile buffer so libpng doesn't allocate it.  See
+    * the call to icc_check_length below (the write case).
+    */
+#  ifdef PNG_SET_USER_LIMITS_SUPPORTED
+      else if (png_ptr->user_chunk_malloc_max > 0 &&
+               png_ptr->user_chunk_malloc_max < profile_length)
+         return png_icc_profile_error(png_ptr, colorspace, name, profile_length,
+             "exceeds application limits");
+#  elif PNG_USER_CHUNK_MALLOC_MAX > 0
+      else if (PNG_USER_CHUNK_MALLOC_MAX < profile_length)
+         return png_icc_profile_error(png_ptr, colorspace, name, profile_length,
+             "exceeds libpng limits");
+#  else /* !SET_USER_LIMITS */
+      /* This will get compiled out on all 32-bit and better systems. */
+      else if (PNG_SIZE_MAX < profile_length)
+         return png_icc_profile_error(png_ptr, colorspace, name, profile_length,
+             "exceeds system limits");
+#  endif /* !SET_USER_LIMITS */
+
+   return 1;
+}
+#endif /* READ_iCCP */
+
 int /* PRIVATE */
 png_icc_check_header(png_const_structrp png_ptr, png_colorspacerp colorspace,
     png_const_charp name, png_uint_32 profile_length,
@@ -2377,7 +2411,7 @@
    if ((colorspace->flags & PNG_COLORSPACE_INVALID) != 0)
       return 0;
 
-   if (png_icc_check_length(png_ptr, colorspace, name, profile_length) != 0 &&
+   if (icc_check_length(png_ptr, colorspace, name, profile_length) != 0 &&
        png_icc_check_header(png_ptr, colorspace, name, profile_length, profile,
            color_type) != 0 &&
        png_icc_check_tag_table(png_ptr, colorspace, name, profile_length,
diff --git a/png.h b/png.h
index ebfb3dd..e1f59c3 100644
--- a/png.h
+++ b/png.h
@@ -1,7 +1,7 @@
 
 /* png.h - header file for PNG reference library
  *
- * libpng version 1.6.24, August 4, 2016
+ * libpng version 1.6.25, September 1, 2016
  *
  * Copyright (c) 1998-2002,2004,2006-2016 Glenn Randers-Pehrson
  * (Version 0.96 Copyright (c) 1996, 1997 Andreas Dilger)
@@ -12,7 +12,7 @@
  * Authors and maintainers:
  *   libpng versions 0.71, May 1995, through 0.88, January 1996: Guy Schalnat
  *   libpng versions 0.89, June 1996, through 0.96, May 1997: Andreas Dilger
- *   libpng versions 0.97, January 1998, through 1.6.24, August 4, 2016:
+ *   libpng versions 0.97, January 1998, through 1.6.25, September 1, 2016:
  *     Glenn Randers-Pehrson.
  *   See also "Contributing Authors", below.
  */
@@ -29,7 +29,7 @@
  * files that are distributed with libpng have other copyright owners and
  * are released under other open source licenses.
  *
- * libpng versions 1.0.7, July 1, 2000 through 1.6.24, August 4, 2016 are
+ * libpng versions 1.0.7, July 1, 2000 through 1.6.25, September 1, 2016 are
  * Copyright (c) 2000-2002, 2004, 2006-2016 Glenn Randers-Pehrson, are
  * derived from libpng-1.0.6, and are distributed according to the same
  * disclaimer and license as libpng-1.0.6 with the following individuals
@@ -41,6 +41,7 @@
  *    Cosmin Truta
  *    Gilles Vollant
  *    James Yu
+ *    Mandar Sahastrabuddhe
  *
  * and with the following additions to the disclaimer:
  *
@@ -217,7 +218,7 @@
  *    ...
  *    1.5.27                  15    10527  15.so.15.27[.0]
  *    ...
- *    1.6.24                  16    10624  16.so.16.24[.0]
+ *    1.6.25                  16    10625  16.so.16.25[.0]
  *
  *    Henceforth the source version will match the shared-library major
  *    and minor numbers; the shared-library major version number will be
@@ -245,13 +246,13 @@
  * Y2K compliance in libpng:
  * =========================
  *
- *    August 4, 2016
+ *    September 1, 2016
  *
  *    Since the PNG Development group is an ad-hoc body, we can't make
  *    an official declaration.
  *
  *    This is your unofficial assurance that libpng from version 0.71 and
- *    upward through 1.6.24 are Y2K compliant.  It is my belief that
+ *    upward through 1.6.25 are Y2K compliant.  It is my belief that
  *    earlier versions were also Y2K compliant.
  *
  *    Libpng only has two year fields.  One is a 2-byte unsigned integer
@@ -313,8 +314,8 @@
  */
 
 /* Version information for png.h - this should match the version in png.c */
-#define PNG_LIBPNG_VER_STRING "1.6.24"
-#define PNG_HEADER_VERSION_STRING " libpng version 1.6.24 - August 4, 2016\n"
+#define PNG_LIBPNG_VER_STRING "1.6.25"
+#define PNG_HEADER_VERSION_STRING " libpng version 1.6.25 - September 1, 2016\n"
 
 #define PNG_LIBPNG_VER_SONUM   16
 #define PNG_LIBPNG_VER_DLLNUM  16
@@ -322,7 +323,7 @@
 /* These should match the first 3 components of PNG_LIBPNG_VER_STRING: */
 #define PNG_LIBPNG_VER_MAJOR   1
 #define PNG_LIBPNG_VER_MINOR   6
-#define PNG_LIBPNG_VER_RELEASE 24
+#define PNG_LIBPNG_VER_RELEASE 25
 
 /* This should match the numeric part of the final component of
  * PNG_LIBPNG_VER_STRING, omitting any leading zero:
@@ -353,7 +354,7 @@
  * version 1.0.0 was mis-numbered 100 instead of 10000).  From
  * version 1.0.1 it's    xxyyzz, where x=major, y=minor, z=release
  */
-#define PNG_LIBPNG_VER 10624 /* 1.6.24 */
+#define PNG_LIBPNG_VER 10625 /* 1.6.25 */
 
 /* Library configuration: these options cannot be changed after
  * the library has been built.
@@ -463,7 +464,7 @@
 /* This triggers a compiler error in png.c, if png.c and png.h
  * do not agree upon the version number.
  */
-typedef char* png_libpng_version_1_6_24;
+typedef char* png_libpng_version_1_6_25;
 
 /* Basic control structions.  Read libpng-manual.txt or libpng.3 for more info.
  *
@@ -3226,7 +3227,10 @@
 #endif
 #define PNG_MAXIMUM_INFLATE_WINDOW 2 /* SOFTWARE: force maximum window */
 #define PNG_SKIP_sRGB_CHECK_PROFILE 4 /* SOFTWARE: Check ICC profile for sRGB */
-#define PNG_OPTION_NEXT  6 /* Next option - numbers must be even */
+#ifdef PNG_MIPS_MSA_API_SUPPORTED
+#  define PNG_MIPS_MSA   6 /* HARDWARE: MIPS Msa SIMD instructions supported */
+#endif
+#define PNG_OPTION_NEXT  8 /* Next option - numbers must be even */
 
 /* Return values: NOTE: there are four values and 'off' is *not* zero */
 #define PNG_OPTION_UNSET   0 /* Unset - defaults to off */
diff --git a/pngconf.h b/pngconf.h
index 9da691d..08b4239 100644
--- a/pngconf.h
+++ b/pngconf.h
@@ -1,7 +1,7 @@
 
 /* pngconf.h - machine configurable file for libpng
  *
- * libpng version 1.6.24, August 4, 2016
+ * libpng version 1.6.25, September 1, 2016
  *
  * Copyright (c) 1998-2002,2004,2006-2016 Glenn Randers-Pehrson
  * (Version 0.96 Copyright (c) 1996, 1997 Andreas Dilger)
diff --git a/pngmem.c b/pngmem.c
index 6033bf2..7053ec9 100644
--- a/pngmem.c
+++ b/pngmem.c
@@ -1,7 +1,7 @@
 
 /* pngmem.c - stub functions for memory allocation
  *
- * Last changed in libpng 1.6.24 [August 4, 2016]
+ * Last changed in libpng 1.6.24 [August 4, 2016%]
  * Copyright (c) 1998-2002,2004,2006-2014,2016 Glenn Randers-Pehrson
  * (Version 0.96 Copyright (c) 1996, 1997 Andreas Dilger)
  * (Version 0.88 Copyright (c) 1995, 1996 Guy Eric Schalnat, Group 42, Inc.)
diff --git a/pngpriv.h b/pngpriv.h
index fe3355d..bed1cab 100644
--- a/pngpriv.h
+++ b/pngpriv.h
@@ -1,7 +1,7 @@
 
 /* pngpriv.h - private declarations for use inside libpng
  *
- * Last changed in libpng 1.6.24 [August 4, 2016]
+ * Last changed in libpng 1.6.25 [September 1, 2016]
  * Copyright (c) 1998-2002,2004,2006-2016 Glenn Randers-Pehrson
  * (Version 0.96 Copyright (c) 1996, 1997 Andreas Dilger)
  * (Version 0.88 Copyright (c) 1995, 1996 Guy Eric Schalnat, Group 42, Inc.)
@@ -182,6 +182,35 @@
 #  endif
 #endif /* PNG_ARM_NEON_OPT > 0 */
 
+#ifndef PNG_MIPS_MSA_OPT
+#  if defined(__mips_msa) && (__mips_isa_rev >= 5) && defined(PNG_ALIGNED_MEMORY_SUPPORTED)
+#     define PNG_MIPS_MSA_OPT 2
+#  else
+#     define PNG_MIPS_MSA_OPT 0
+#  endif
+#endif
+
+#if PNG_MIPS_MSA_OPT > 0
+#  define PNG_FILTER_OPTIMIZATIONS png_init_filter_functions_msa
+#  ifndef PNG_MIPS_MSA_IMPLEMENTATION
+#     if defined(__mips_msa)
+#        if defined(__clang__)
+#        elif defined(__GNUC__)
+#           if __GNUC__ < 4 || (__GNUC__ == 4 && __GNUC_MINOR__ < 7)
+#              define PNG_MIPS_MSA_IMPLEMENTATION 2
+#           endif /* no GNUC support */
+#        endif /* __GNUC__ */
+#     else /* !defined __mips_msa */
+#        define PNG_MIPS_MSA_IMPLEMENTATION 2
+#     endif /* __mips_msa */
+#  endif /* !PNG_MIPS_MSA_IMPLEMENTATION */
+
+#  ifndef PNG_MIPS_MSA_IMPLEMENTATION
+#     define PNG_MIPS_MSA_IMPLEMENTATION 1
+#  endif
+#endif /* PNG_MIPS_MSA_OPT > 0 */
+
+
 /* Is this a build of a DLL where compilation of the object modules requires
  * different preprocessor settings to those required for a simple library?  If
  * so PNG_BUILD_DLL must be set.
@@ -1191,6 +1220,23 @@
     row_info, png_bytep row, png_const_bytep prev_row),PNG_EMPTY);
 #endif
 
+#if PNG_MIPS_MSA_OPT > 0
+PNG_INTERNAL_FUNCTION(void,png_read_filter_row_up_msa,(png_row_infop row_info,
+    png_bytep row, png_const_bytep prev_row),PNG_EMPTY);
+PNG_INTERNAL_FUNCTION(void,png_read_filter_row_sub3_msa,(png_row_infop
+    row_info, png_bytep row, png_const_bytep prev_row),PNG_EMPTY);
+PNG_INTERNAL_FUNCTION(void,png_read_filter_row_sub4_msa,(png_row_infop
+    row_info, png_bytep row, png_const_bytep prev_row),PNG_EMPTY);
+PNG_INTERNAL_FUNCTION(void,png_read_filter_row_avg3_msa,(png_row_infop
+    row_info, png_bytep row, png_const_bytep prev_row),PNG_EMPTY);
+PNG_INTERNAL_FUNCTION(void,png_read_filter_row_avg4_msa,(png_row_infop
+    row_info, png_bytep row, png_const_bytep prev_row),PNG_EMPTY);
+PNG_INTERNAL_FUNCTION(void,png_read_filter_row_paeth3_msa,(png_row_infop
+    row_info, png_bytep row, png_const_bytep prev_row),PNG_EMPTY);
+PNG_INTERNAL_FUNCTION(void,png_read_filter_row_paeth4_msa,(png_row_infop
+    row_info, png_bytep row, png_const_bytep prev_row),PNG_EMPTY);
+#endif
+
 /* Choose the best filter to use and filter the row data */
 PNG_INTERNAL_FUNCTION(void,png_write_find_filter,(png_structrp png_ptr,
     png_row_infop row_info),PNG_EMPTY);
@@ -1494,9 +1540,11 @@
    /* The 'name' is used for information only */
 
 /* Routines for checking parts of an ICC profile. */
+#ifdef PNG_READ_iCCP_SUPPORTED
 PNG_INTERNAL_FUNCTION(int,png_icc_check_length,(png_const_structrp png_ptr,
    png_colorspacerp colorspace, png_const_charp name,
    png_uint_32 profile_length), PNG_EMPTY);
+#endif /* READ_iCCP */
 PNG_INTERNAL_FUNCTION(int,png_icc_check_header,(png_const_structrp png_ptr,
    png_colorspacerp colorspace, png_const_charp name,
    png_uint_32 profile_length,
@@ -1918,7 +1966,12 @@
 #  if PNG_ARM_NEON_OPT > 0
 PNG_INTERNAL_FUNCTION(void, png_init_filter_functions_neon,
    (png_structp png_ptr, unsigned int bpp), PNG_EMPTY);
-#  endif
+#endif
+
+#if PNG_MIPS_MSA_OPT > 0
+PNG_INTERNAL_FUNCTION(void, png_init_filter_functions_msa,
+   (png_structp png_ptr, unsigned int bpp), PNG_EMPTY);
+#endif
 #endif
 
 PNG_INTERNAL_FUNCTION(png_uint_32, png_check_keyword, (png_structrp png_ptr,
diff --git a/pngrutil.c b/pngrutil.c
index 1c8a179..3eaa635 100644
--- a/pngrutil.c
+++ b/pngrutil.c
@@ -1,7 +1,7 @@
 
 /* pngrutil.c - utilities to read a PNG file
  *
- * Last changed in libpng 1.6.24 [August 4, 2016]
+ * Last changed in libpng 1.6.25 [September 1, 2016]
  * Copyright (c) 1998-2002,2004,2006-2016 Glenn Randers-Pehrson
  * (Version 0.96 Copyright (c) 1996, 1997 Andreas Dilger)
  * (Version 0.88 Copyright (c) 1995, 1996 Guy Eric Schalnat, Group 42, Inc.)
@@ -461,6 +461,7 @@
 #endif /* Zlib >= 1.2.4 */
 
 #ifdef PNG_READ_COMPRESSED_TEXT_SUPPORTED
+#if defined(PNG_READ_zTXt_SUPPORTED) || defined (PNG_READ_iTXt_SUPPORTED)
 /* png_inflate now returns zlib error codes including Z_OK and Z_STREAM_END to
  * allow the caller to do multiple calls if required.  If the 'finish' flag is
  * set Z_FINISH will be passed to the final inflate() call and Z_STREAM_END must
@@ -590,7 +591,6 @@
    }
 }
 
-#if defined(PNG_READ_zTXt_SUPPORTED) || defined (PNG_READ_iTXt_SUPPORTED)
 /*
  * Decompress trailing data in a chunk.  The assumption is that read_buffer
  * points at an allocated area holding the contents of a chunk with a
diff --git a/pngtest.c b/pngtest.c
index dd222a6..9034d16 100644
--- a/pngtest.c
+++ b/pngtest.c
@@ -1,7 +1,7 @@
 
 /* pngtest.c - a simple test program to test libpng
  *
- * Last changed in libpng 1.6.24 [August 4, 2016]
+ * Last changed in libpng 1.6.25 [September 1, 2016]
  * Copyright (c) 1998-2002,2004,2006-2016 Glenn Randers-Pehrson
  * (Version 0.96 Copyright (c) 1996, 1997 Andreas Dilger)
  * (Version 0.88 Copyright (c) 1995, 1996 Guy Eric Schalnat, Group 42, Inc.)
@@ -514,10 +514,10 @@
 typedef memory_information *memory_infop;
 
 static memory_infop pinformation = NULL;
-static int current_allocation = 0;
-static int maximum_allocation = 0;
-static int total_allocation = 0;
-static int num_allocations = 0;
+static png_alloc_size_t current_allocation = 0;
+static png_alloc_size_t maximum_allocation = 0;
+static png_alloc_size_t total_allocation = 0;
+static png_alloc_size_t num_allocations = 0;
 
 png_voidp PNGCBAPI png_debug_malloc PNGARG((png_structp png_ptr,
     png_alloc_size_t size));
@@ -604,9 +604,10 @@
          if (pinfo->pointer == ptr)
          {
             *ppinfo = pinfo->next;
-            current_allocation -= pinfo->size;
-            if (current_allocation < 0)
+            if (current_allocation < pinfo->size)
                fprintf(STDERR, "Duplicate free of memory\n");
+            else
+               current_allocation -= pinfo->size;
             /* We must free the list element too, but first kill
                the memory that is to be freed. */
             memset(ptr, 0x55, pinfo->size);
@@ -938,6 +939,12 @@
        read_user_chunk_callback);
 #endif
 
+#ifdef PNG_SET_USER_LIMITS_SUPPORTED
+#  ifdef CHUNK_LIMIT /* from the build, for testing */
+      png_set_chunk_malloc_max(read_ptr, CHUNK_LIMIT);
+#  endif /* CHUNK_LIMIT */
+#endif
+
 #ifdef PNG_SETJMP_SUPPORTED
    pngtest_debug("Setting jmpbuf for read struct");
    if (setjmp(png_jmpbuf(read_ptr)))
@@ -1876,7 +1883,7 @@
    {
       int i;
 #if defined(PNG_USER_MEM_SUPPORTED) && PNG_DEBUG
-      int allocation_now = current_allocation;
+      png_alloc_size_t allocation_now = current_allocation;
 #endif
       for (i=2; i<argc; ++i)
       {
@@ -1909,15 +1916,15 @@
          }
 #if defined(PNG_USER_MEM_SUPPORTED) && PNG_DEBUG
          if (allocation_now != current_allocation)
-            fprintf(STDERR, "MEMORY ERROR: %d bytes lost\n",
-                current_allocation - allocation_now);
+            fprintf(STDERR, "MEMORY ERROR: %lu bytes lost\n",
+                (unsigned long)(current_allocation - allocation_now));
 
          if (current_allocation != 0)
          {
             memory_infop pinfo = pinformation;
 
-            fprintf(STDERR, "MEMORY ERROR: %d bytes still allocated\n",
-                current_allocation);
+            fprintf(STDERR, "MEMORY ERROR: %lu bytes still allocated\n",
+                (unsigned long)current_allocation);
 
             while (pinfo != NULL)
             {
@@ -1930,14 +1937,14 @@
 #endif
       }
 #if defined(PNG_USER_MEM_SUPPORTED) && PNG_DEBUG
-         fprintf(STDERR, " Current memory allocation: %10d bytes\n",
-             current_allocation);
-         fprintf(STDERR, " Maximum memory allocation: %10d bytes\n",
-             maximum_allocation);
-         fprintf(STDERR, " Total   memory allocation: %10d bytes\n",
-             total_allocation);
-         fprintf(STDERR, "     Number of allocations: %10d\n",
-             num_allocations);
+         fprintf(STDERR, " Current memory allocation: %20lu bytes\n",
+             (unsigned long)current_allocation);
+         fprintf(STDERR, " Maximum memory allocation: %20lu bytes\n",
+             (unsigned long) maximum_allocation);
+         fprintf(STDERR, " Total   memory allocation: %20lu bytes\n",
+             (unsigned long)total_allocation);
+         fprintf(STDERR, "     Number of allocations: %20lu\n",
+             (unsigned long)num_allocations);
 #endif
    }
 
@@ -1948,7 +1955,7 @@
       {
          int kerror;
 #if defined(PNG_USER_MEM_SUPPORTED) && PNG_DEBUG
-         int allocation_now = current_allocation;
+         png_alloc_size_t allocation_now = current_allocation;
 #endif
          if (i == 1)
             status_dots_requested = 1;
@@ -1998,15 +2005,15 @@
          }
 #if defined(PNG_USER_MEM_SUPPORTED) && PNG_DEBUG
          if (allocation_now != current_allocation)
-             fprintf(STDERR, "MEMORY ERROR: %d bytes lost\n",
-                 current_allocation - allocation_now);
+             fprintf(STDERR, "MEMORY ERROR: %lu bytes lost\n",
+                 (unsigned long)(current_allocation - allocation_now));
 
          if (current_allocation != 0)
          {
              memory_infop pinfo = pinformation;
 
-             fprintf(STDERR, "MEMORY ERROR: %d bytes still allocated\n",
-                 current_allocation);
+             fprintf(STDERR, "MEMORY ERROR: %lu bytes still allocated\n",
+                 (unsigned long)current_allocation);
 
              while (pinfo != NULL)
              {
@@ -2018,14 +2025,14 @@
 #endif
        }
 #if defined(PNG_USER_MEM_SUPPORTED) && PNG_DEBUG
-       fprintf(STDERR, " Current memory allocation: %10d bytes\n",
-           current_allocation);
-       fprintf(STDERR, " Maximum memory allocation: %10d bytes\n",
-           maximum_allocation);
-       fprintf(STDERR, " Total   memory allocation: %10d bytes\n",
-           total_allocation);
-       fprintf(STDERR, "     Number of allocations: %10d\n",
-           num_allocations);
+       fprintf(STDERR, " Current memory allocation: %20lu bytes\n",
+           (unsigned long)current_allocation);
+       fprintf(STDERR, " Maximum memory allocation: %20lu bytes\n",
+           (unsigned long)maximum_allocation);
+       fprintf(STDERR, " Total   memory allocation: %20lu bytes\n",
+           (unsigned long)total_allocation);
+       fprintf(STDERR, "     Number of allocations: %20lu\n",
+           (unsigned long)num_allocations);
 #endif
    }
 
@@ -2081,4 +2088,4 @@
 #endif
 
 /* Generate a compiler error if there is an old png.h in the search path. */
-typedef png_libpng_version_1_6_24 Your_png_h_is_not_version_1_6_24;
+typedef png_libpng_version_1_6_25 Your_png_h_is_not_version_1_6_25;
diff --git a/projects/vstudio/README.txt b/projects/vstudio/README.txt
index 0aabdb7..0457c2d 100644
--- a/projects/vstudio/README.txt
+++ b/projects/vstudio/README.txt
@@ -1,7 +1,7 @@
 
 VisualStudio instructions
 
-libpng version 1.6.24 - August 4, 2016
+libpng version 1.6.25 - September 1, 2016
 
 Copyright (c) 2010,2013,2015 Glenn Randers-Pehrson
 
diff --git a/projects/vstudio/zlib.props b/projects/vstudio/zlib.props
index 8d370b2..8adbd1e 100644
--- a/projects/vstudio/zlib.props
+++ b/projects/vstudio/zlib.props
@@ -2,7 +2,7 @@
 <!--
  * zlib.props - location of zlib source
  *
- * libpng version 1.6.24 - August 4, 2016
+ * libpng version 1.6.25 - September 1, 2016
  *
  * Copyright (c) 1998-2011 Glenn Randers-Pehrson
  *
diff --git a/scripts/README.txt b/scripts/README.txt
index 56933f2..242da14 100644
--- a/scripts/README.txt
+++ b/scripts/README.txt
@@ -1,9 +1,9 @@
 
-Makefiles for  libpng version 1.6.24 - August 4, 2016
+Makefiles for  libpng version 1.6.25 - September 1, 2016
 
 pnglibconf.h.prebuilt       =>  Stores configuration settings
  makefile.linux    =>  Linux/ELF makefile
-                       (gcc, creates libpng16.so.16.1.6.24)
+                       (gcc, creates libpng16.so.16.1.6.25)
  makefile.gcc      =>  Generic makefile (gcc, creates static libpng.a)
  makefile.knr      =>  Archaic UNIX Makefile that converts files with
                        ansi2knr (Requires ansi2knr.c from
@@ -33,12 +33,12 @@
  makefile.os2      =>  OS/2 Makefile (gcc and emx, requires libpng.def)
  makefile.sco      =>  For SCO OSr5  ELF and Unixware 7 with Native cc
  makefile.sggcc    =>  Silicon Graphics (gcc,
-                       creates libpng16.so.16.1.6.24)
+                       creates libpng16.so.16.1.6.25)
  makefile.sgi      =>  Silicon Graphics IRIX makefile (cc, creates static lib)
  makefile.solaris  =>  Solaris 2.X makefile (gcc,
-                       creates libpng16.so.16.1.6.24)
+                       creates libpng16.so.16.1.6.25)
  makefile.so9      =>  Solaris 9 makefile (gcc,
-                       creates libpng16.so.16.1.6.24)
+                       creates libpng16.so.16.1.6.25)
  makefile.std      =>  Generic UNIX makefile (cc, creates static libpng.a)
  makefile.sunos    =>  Sun makefile
  makefile.32sunu   =>  Sun Ultra 32-bit makefile
diff --git a/scripts/def.c b/scripts/def.c
index 7c04875..9aaa56e 100644
--- a/scripts/def.c
+++ b/scripts/def.c
@@ -21,7 +21,7 @@
 PNG_DFN "OS2 CODE PRELOAD MOVEABLE DISCARDABLE"
 PNG_DFN ""
 PNG_DFN "EXPORTS"
-PNG_DFN ";Version 1.6.24"
+PNG_DFN ";Version 1.6.25"
 
 #define PNG_EXPORTA(ordinal, type, name, args, attributes)\
         PNG_DFN "@" SYMBOL_PREFIX "@@" name "@"
diff --git a/scripts/libpng-config-head.in b/scripts/libpng-config-head.in
index b0c6e38..d889518 100644
--- a/scripts/libpng-config-head.in
+++ b/scripts/libpng-config-head.in
@@ -11,7 +11,7 @@
 
 # Modeled after libxml-config.
 
-version=1.6.24
+version=1.6.25
 prefix=""
 libdir=""
 libs=""
diff --git a/scripts/libpng.pc.in b/scripts/libpng.pc.in
index bed81d3..8263f8c 100644
--- a/scripts/libpng.pc.in
+++ b/scripts/libpng.pc.in
@@ -5,6 +5,6 @@
 
 Name: libpng
 Description: Loads and saves PNG files
-Version: 1.6.24
+Version: 1.6.25
 Libs: -L${libdir} -lpng16
 Cflags: -I${includedir}
diff --git a/scripts/makefile.cegcc b/scripts/makefile.cegcc
index fb1361b..d723658 100644
--- a/scripts/makefile.cegcc
+++ b/scripts/makefile.cegcc
@@ -23,7 +23,7 @@
 
 VERMAJ = 1
 VERMIN = 6
-VERMIC = 24
+VERMIC = 25
 VER = $(VERMAJ).$(VERMIN).$(VERMIC)
 NAME = libpng
 PACKAGE = $(NAME)-$(VER)
diff --git a/scripts/makefile.linux b/scripts/makefile.linux
index af0d65e..c2bfac6 100644
--- a/scripts/makefile.linux
+++ b/scripts/makefile.linux
@@ -10,7 +10,7 @@
 # Library name:
 LIBNAME = libpng16
 PNGMAJ = 16
-RELEASE = 24
+RELEASE = 25
 
 # Shared library names:
 LIBSO=$(LIBNAME).so
diff --git a/scripts/makefile.msys b/scripts/makefile.msys
index ddbd88c..b0c0e3c 100644
--- a/scripts/makefile.msys
+++ b/scripts/makefile.msys
@@ -18,7 +18,7 @@
 # Library name:
 LIBNAME = libpng16
 PNGMAJ = 16
-RELEASE = 24
+RELEASE = 25
 
 # Shared library names:
 LIBSO=$(LIBNAME).dll
diff --git a/scripts/makefile.ne12bsd b/scripts/makefile.ne12bsd
index d18bdb1..21ffe56 100644
--- a/scripts/makefile.ne12bsd
+++ b/scripts/makefile.ne12bsd
@@ -17,7 +17,7 @@
 
 LIB=	png16
 SHLIB_MAJOR=	0
-SHLIB_MINOR=	1.6.24
+SHLIB_MINOR=	1.6.25
 SRCS=	png.c pngset.c pngget.c pngrutil.c pngtrans.c pngwutil.c \
 	pngread.c pngrio.c pngwio.c pngwrite.c pngrtran.c \
 	pngwtran.c pngmem.c pngerror.c pngpread.c
diff --git a/scripts/makefile.netbsd b/scripts/makefile.netbsd
index 85c2ea4..0046f69 100644
--- a/scripts/makefile.netbsd
+++ b/scripts/makefile.netbsd
@@ -17,7 +17,7 @@
 
 LIB=	png
 SHLIB_MAJOR=	16
-SHLIB_MINOR=	1.6.24
+SHLIB_MINOR=	1.6.25
 SRCS=	png.c pngset.c pngget.c pngrutil.c pngtrans.c pngwutil.c \
 	pngread.c pngrio.c pngwio.c pngwrite.c pngrtran.c \
 	pngwtran.c pngmem.c pngerror.c pngpread.c
diff --git a/scripts/makefile.openbsd b/scripts/makefile.openbsd
index 07b8183..756759f 100644
--- a/scripts/makefile.openbsd
+++ b/scripts/makefile.openbsd
@@ -11,7 +11,7 @@
 MANDIR= ${PREFIX}/man/cat
 
 SHLIB_MAJOR=	16
-SHLIB_MINOR=	1.6.24
+SHLIB_MINOR=	1.6.25
 
 LIB=	png
 SRCS=	png.c pngerror.c pngget.c pngmem.c pngpread.c \
diff --git a/scripts/pnglibconf.h.prebuilt b/scripts/pnglibconf.h.prebuilt
index d91d9f4..d7d21b2 100644
--- a/scripts/pnglibconf.h.prebuilt
+++ b/scripts/pnglibconf.h.prebuilt
@@ -1,8 +1,8 @@
-/* libpng 1.6.24 STANDARD API DEFINITION */
+/* libpng 1.6.25 STANDARD API DEFINITION */
 
 /* pnglibconf.h - library build configuration */
 
-/* Libpng version 1.6.24 - August 4, 2016 */
+/* Libpng version 1.6.25 - September 1, 2016 */
 
 /* Copyright (c) 1998-2015 Glenn Randers-Pehrson */
 
diff --git a/scripts/symbols.def b/scripts/symbols.def
index d3851f5..cdf7896 100644
--- a/scripts/symbols.def
+++ b/scripts/symbols.def
@@ -1,4 +1,4 @@
-;Version 1.6.24
+;Version 1.6.25
 ;--------------------------------------------------------------
 ; LIBPNG symbol list as a Win32 DEF file
 ; Contains all the symbols that can be exported from libpng