Merge lp:~linaro-graphics-wg/glmark2/image-readers into lp:glmark2/2011.11

[Alexandros Frantzis]

Add support for loading texture data from JPEG files, plus some refactoring of the image reading infrastructure.

[Jesse Barker (jesse-barker)]

In the function that fills in the "struct jpeg_source_mgr" function pointer for skip_input_data, while I normally object to C casting in C++, it seems especially odd in this case as there's a mix of C and C++ casts in this function. Aside from that, the code looks great.

review: Approve

[Alexandros Frantzis (afrantzis)]

Thanks, I copied that from the jpeg source code and didn't pay attention to the casts. Fixed in 232.

[lp:~linaro-graphics-wg/glmark2/image-readers]

232. By Alexandros Frantzis on 2012-06-27

ImageReader: Remove C casts.

=== modified file 'android/jni/Android.mk'
--- android/jni/Android.mk	2012-06-21 12:57:43 +0000
+++ android/jni/Android.mk	2012-06-27 16:20:24 +0000
@@ -24,6 +24,18 @@
 
 include $(CLEAR_VARS)
 
+LOCAL_MODULE := libglmark2-jpeg
+LOCAL_CFLAGS := -Werror -Wall -Wextra -Wno-error=attributes \
+                -Wno-error=unused-parameter -Wno-error=unused-function -Wno-error=unused-variable
+LOCAL_C_INCLUDES := $(LOCAL_PATH)/src/libjpeg-turbo/
+LOCAL_SRC_FILES := $(subst $(LOCAL_PATH)/,,$(wildcard $(LOCAL_PATH)/src/libjpeg-turbo/simd/*.c)) \
+                   $(subst $(LOCAL_PATH)/,,$(wildcard $(LOCAL_PATH)/src/libjpeg-turbo/simd/*.S)) \
+                   $(subst $(LOCAL_PATH)/,,$(wildcard $(LOCAL_PATH)/src/libjpeg-turbo/*.c))
+
+include $(BUILD_STATIC_LIBRARY)
+
+include $(CLEAR_VARS)
+
 LOCAL_CPP_EXTENSION := .cc
 LOCAL_MODULE := libglmark2-ideas
 LOCAL_CFLAGS := -DGLMARK_DATA_PATH="" -DUSE_GLESv2 -Werror -Wall -Wextra\
@@ -41,7 +53,7 @@
 
 LOCAL_MODULE_TAGS := optional
 LOCAL_MODULE := libglmark2-android
-LOCAL_STATIC_LIBRARIES := libglmark2-matrix libglmark2-png libglmark2-ideas
+LOCAL_STATIC_LIBRARIES := libglmark2-matrix libglmark2-png libglmark2-ideas libglmark2-jpeg
 LOCAL_CFLAGS := -DGLMARK_DATA_PATH="" -DGLMARK_VERSION="\"2012.06\"" \
                 -DUSE_GLESv2 -Werror -Wall -Wextra -Wnon-virtual-dtor \
                 -Wno-error=unused-parameter
@@ -49,6 +61,7 @@
 LOCAL_C_INCLUDES := $(LOCAL_PATH)/src \
                     $(LOCAL_PATH)/src/libmatrix \
                     $(LOCAL_PATH)/src/scene-ideas \
+                    $(LOCAL_PATH)/src/libjpeg-turbo \
                     $(LOCAL_PATH)/src/libpng \
                     external/zlib
 LOCAL_SRC_FILES := $(filter-out src/canvas% src/main.cpp, \

=== modified file 'android/jni/Android.ndk.mk'
--- android/jni/Android.ndk.mk	2012-06-21 12:57:43 +0000
+++ android/jni/Android.ndk.mk	2012-06-27 16:20:24 +0000
@@ -20,6 +20,18 @@
 
 include $(CLEAR_VARS)
 
+LOCAL_MODULE := libglmark2-jpeg
+LOCAL_CFLAGS := -Werror -Wall -Wextra -Wno-error=attributes \
+                -Wno-error=unused-parameter -Wno-error=unused-function -Wno-error=unused-variable
+LOCAL_C_INCLUDES := $(LOCAL_PATH)/src/libjpeg-turbo/
+LOCAL_SRC_FILES := $(subst $(LOCAL_PATH)/,,$(wildcard $(LOCAL_PATH)/src/libjpeg-turbo/simd/*.c)) \
+                   $(subst $(LOCAL_PATH)/,,$(wildcard $(LOCAL_PATH)/src/libjpeg-turbo/simd/*.S)) \
+                   $(subst $(LOCAL_PATH)/,,$(wildcard $(LOCAL_PATH)/src/libjpeg-turbo/*.c))
+
+include $(BUILD_STATIC_LIBRARY)
+
+include $(CLEAR_VARS)
+
 LOCAL_CPP_EXTENSION := .cc
 LOCAL_MODULE := libglmark2-ideas
 LOCAL_CFLAGS := -DGLMARK_DATA_PATH="" -DUSE_GLESv2 -Werror -Wall -Wextra\
@@ -34,7 +46,7 @@
 
 LOCAL_MODULE_TAGS := optional
 LOCAL_MODULE := libglmark2-android
-LOCAL_STATIC_LIBRARIES := libglmark2-matrix libglmark2-png libglmark2-ideas
+LOCAL_STATIC_LIBRARIES := libglmark2-matrix libglmark2-png libglmark2-ideas libglmark2-jpeg
 LOCAL_CFLAGS := -DGLMARK_DATA_PATH="" -DGLMARK_VERSION="\"2012.06\"" \
                 -DUSE_GLESv2 -Werror -Wall -Wextra -Wnon-virtual-dtor \
                 -Wno-error=unused-parameter
@@ -42,6 +54,7 @@
 LOCAL_C_INCLUDES := $(LOCAL_PATH)/src \
                     $(LOCAL_PATH)/src/libmatrix \
                     $(LOCAL_PATH)/src/scene-ideas \
+                    $(LOCAL_PATH)/src/libjpeg-turbo \
                     $(LOCAL_PATH)/src/libpng
 LOCAL_SRC_FILES := $(filter-out src/canvas% src/main.cpp, \
                      $(subst $(LOCAL_PATH)/,,$(wildcard $(LOCAL_PATH)/src/*.cpp))) \

=== added file 'src/image-reader.cpp'
--- src/image-reader.cpp	1970-01-01 00:00:00 +0000
+++ src/image-reader.cpp	2012-06-27 16:20:24 +0000
@@ -0,0 +1,386 @@
+/*
+ * Copyright © 2012 Linaro Limited
+ *
+ * This file is part of the glmark2 OpenGL (ES) 2.0 benchmark.
+ *
+ * glmark2 is free software: you can redistribute it and/or modify it under the
+ * terms of the GNU General Public License as published by the Free Software
+ * Foundation, either version 3 of the License, or (at your option) any later
+ * version.
+ *
+ * glmark2 is distributed in the hope that it will be useful, but WITHOUT ANY
+ * WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+ * FOR A PARTICULAR PURPOSE.  See the GNU General Public License for more
+ * details.
+ *
+ * You should have received a copy of the GNU General Public License along with
+ * glmark2.  If not, see <http://www.gnu.org/licenses/>.
+ *
+ * Authors:
+ *  Alexandros Frantzis
+ */
+#include <cstdio>
+#include <png.h>
+#include <jpeglib.h>
+#include <memory>
+
+#include "image-reader.h"
+#include "log.h"
+#include "util.h"
+
+/*******
+ * PNG *
+ *******/
+
+struct PNGReaderPrivate
+{
+    PNGReaderPrivate() :
+        png(0), info(0), rows(0), png_error(0),
+        current_row(0), row_stride(0) {}
+
+    static void png_read_fn(png_structp png_ptr, png_bytep data, png_size_t length)
+    {
+        std::istream *is = reinterpret_cast<std::istream*>(png_get_io_ptr(png_ptr));
+        is->read(reinterpret_cast<char *>(data), length);
+    }
+
+    png_structp png;
+    png_infop info;
+    png_bytepp rows;
+    bool png_error;
+    unsigned int current_row;
+    unsigned int row_stride;
+};
+
+PNGReader::PNGReader(const std::string& filename):
+    priv_(new PNGReaderPrivate())
+{
+    priv_->png_error = !init(filename);
+}
+
+PNGReader::~PNGReader()
+{
+    finish();
+    delete priv_;
+}
+
+bool
+PNGReader::error()
+{
+    return priv_->png_error;
+}
+
+bool
+PNGReader::nextRow(unsigned char *dst)
+{
+    bool ret;
+
+    if (priv_->current_row < height()) {
+        memcpy(dst, priv_->rows[priv_->current_row], priv_->row_stride);
+        priv_->current_row++;
+        ret = true;
+    }
+    else {
+        ret = false;
+    }
+
+    return ret;
+}
+
+unsigned int
+PNGReader::width() const
+{ 
+    return png_get_image_width(priv_->png, priv_->info);
+}
+
+unsigned int
+PNGReader::height() const
+{ 
+    return png_get_image_height(priv_->png, priv_->info);
+}
+
+unsigned int
+PNGReader::pixelBytes() const
+{
+    if (png_get_color_type(priv_->png, priv_->info) == PNG_COLOR_TYPE_RGB)
+    {
+        return 3;
+    }
+    return 4;
+}
+
+
+bool
+PNGReader::init(const std::string& filename)
+{
+    static const int png_transforms = PNG_TRANSFORM_STRIP_16 |
+                                      PNG_TRANSFORM_GRAY_TO_RGB |
+                                      PNG_TRANSFORM_PACKING |
+                                      PNG_TRANSFORM_EXPAND;
+
+    Log::debug("Reading PNG file %s\n", filename.c_str());
+
+    const std::auto_ptr<std::istream> is_ptr(Util::get_resource(filename));
+    if (!(*is_ptr)) {
+        Log::error("Cannot open file %s!\n", filename.c_str());
+        return false;
+    }
+
+    /* Set up all the libpng structs we need */
+    priv_->png = png_create_read_struct(PNG_LIBPNG_VER_STRING, 0, 0, 0);
+    if (!priv_->png) {
+        Log::error("Couldn't create libpng read struct\n");
+        return false;
+    }
+
+    priv_->info = png_create_info_struct(priv_->png);
+    if (!priv_->info) {
+        Log::error("Couldn't create libpng info struct\n");
+        return false;
+    }
+
+    /* Set up libpng error handling */
+    if (setjmp(png_jmpbuf(priv_->png))) {
+        Log::error("libpng error while reading file %s\n", filename.c_str());
+        return false;
+    }
+
+    /* Read the image information and data */
+    png_set_read_fn(priv_->png, reinterpret_cast<voidp>(is_ptr.get()),
+                    PNGReaderPrivate::png_read_fn);
+
+    png_read_png(priv_->png, priv_->info, png_transforms, 0);
+
+    priv_->rows = png_get_rows(priv_->png, priv_->info);
+
+    priv_->current_row = 0;
+    priv_->row_stride = width() * pixelBytes();
+
+    return true;
+}
+
+void
+PNGReader::finish()
+{
+    if (priv_->png)
+    {
+        png_destroy_read_struct(&priv_->png, &priv_->info, 0);
+    }
+}
+
+
+/********
+ * JPEG *
+ ********/
+
+struct JPEGErrorMgr
+{
+    struct jpeg_error_mgr pub;
+    jmp_buf jmp_buffer;
+
+    JPEGErrorMgr()
+    {
+        jpeg_std_error(&pub);
+        pub.error_exit = error_exit;
+    }
+
+    static void error_exit(j_common_ptr cinfo)
+    {
+        JPEGErrorMgr *err =
+            reinterpret_cast<JPEGErrorMgr *>(cinfo->err);
+
+        char buffer[JMSG_LENGTH_MAX];
+
+        /* Create the message */
+        (*cinfo->err->format_message)(cinfo, buffer);
+        std::string msg(std::string(buffer) + "\n");
+        Log::error(msg.c_str());
+
+        longjmp(err->jmp_buffer, 1);
+    }
+};
+
+struct JPEGIStreamSourceMgr
+{
+    static const int BUFFER_SIZE = 4096;
+    struct jpeg_source_mgr pub;
+    std::istream *is;
+    JOCTET buffer[BUFFER_SIZE];
+
+    JPEGIStreamSourceMgr(const std::string& filename) : is(0)
+    {
+        is = Util::get_resource(filename);
+
+        /* Fill in jpeg_source_mgr pub struct */
+        pub.init_source = init_source;
+        pub.fill_input_buffer = fill_input_buffer;
+        pub.skip_input_data = skip_input_data;
+        pub.resync_to_restart = jpeg_resync_to_restart; /* use default method */
+        pub.term_source = term_source;
+        pub.bytes_in_buffer = 0; /* forces fill_input_buffer on first read */
+        pub.next_input_byte = NULL; /* until buffer loaded */
+    }
+
+    ~JPEGIStreamSourceMgr()
+    {
+        delete is;
+    }
+
+    bool error()
+    {
+        return !is || (is->fail() && !is->eof());
+    }
+
+    static void init_source(j_decompress_ptr cinfo)
+    {
+        static_cast<void>(cinfo);
+    }
+
+    static boolean fill_input_buffer(j_decompress_ptr cinfo)
+    {
+        JPEGIStreamSourceMgr *src =
+            reinterpret_cast<JPEGIStreamSourceMgr *>(cinfo->src);
+
+        src->is->read(reinterpret_cast<char *>(src->buffer), BUFFER_SIZE);
+
+        src->pub.next_input_byte = src->buffer;
+        src->pub.bytes_in_buffer = src->is->gcount();
+
+        /* 
+         * If the decoder needs more data, but we have no more bytes left to
+         * read mark the end of input.
+         */
+        if (src->pub.bytes_in_buffer == 0) {
+            src->pub.bytes_in_buffer = 2;
+            src->buffer[0] = 0xFF;
+            src->buffer[0] = JPEG_EOI;
+        }
+
+        return TRUE;
+    }
+
+    static void skip_input_data(j_decompress_ptr cinfo, long num_bytes)
+    {
+        JPEGIStreamSourceMgr *src =
+            reinterpret_cast<JPEGIStreamSourceMgr *>(cinfo->src);
+
+        if (num_bytes > 0) {
+            size_t n = static_cast<size_t>(num_bytes);
+            while (n > src->pub.bytes_in_buffer) {
+                n -= src->pub.bytes_in_buffer;
+                (*src->fill_input_buffer)(cinfo);
+            }
+            src->pub.next_input_byte += n;
+            src->pub.bytes_in_buffer -= n;
+        }
+    }
+
+    static void term_source(j_decompress_ptr cinfo)
+    {
+        static_cast<void>(cinfo);
+    }
+};
+
+struct JPEGReaderPrivate
+{
+    JPEGReaderPrivate(const std::string& filename) :
+        source_mgr(filename), jpeg_error(false) {}
+
+    struct jpeg_decompress_struct cinfo;
+    JPEGErrorMgr error_mgr;
+    JPEGIStreamSourceMgr source_mgr;
+    bool jpeg_error;
+};
+
+
+JPEGReader::JPEGReader(const std::string& filename) :
+    priv_(new JPEGReaderPrivate(filename))
+{
+    priv_->jpeg_error = !init(filename);
+}
+
+JPEGReader::~JPEGReader()
+{
+    finish();
+    delete priv_;
+}
+
+bool
+JPEGReader::error()
+{
+    return priv_->jpeg_error || priv_->source_mgr.error();
+}
+
+bool
+JPEGReader::nextRow(unsigned char *dst)
+{
+    bool ret = true;
+    unsigned char *buffer[1];
+    buffer[0] = dst;
+
+    /* Set up error handling */
+    if (setjmp(priv_->error_mgr.jmp_buffer)) {
+        return false;
+    }
+
+    /* While there are lines left, read next line */
+    if (priv_->cinfo.output_scanline < priv_->cinfo.output_height) {
+        jpeg_read_scanlines(&priv_->cinfo, buffer, 1);
+    }
+    else {
+        jpeg_finish_decompress(&priv_->cinfo);
+        ret = false;
+    }
+
+    return ret;
+}
+
+unsigned int
+JPEGReader::width() const
+{ 
+    return priv_->cinfo.output_width;
+}
+
+unsigned int
+JPEGReader::height() const
+{ 
+    return priv_->cinfo.output_height;
+}
+
+unsigned int
+JPEGReader::pixelBytes() const
+{ 
+    return priv_->cinfo.output_components;
+}
+
+bool
+JPEGReader::init(const std::string& filename)
+{
+    Log::debug("Reading JPEG file %s\n", filename.c_str());
+
+    /* Initialize error manager */
+    priv_->cinfo.err = reinterpret_cast<jpeg_error_mgr*>(&priv_->error_mgr);
+
+    if (setjmp(priv_->error_mgr.jmp_buffer)) {
+        return false;
+    }
+
+    jpeg_create_decompress(&priv_->cinfo);
+    priv_->cinfo.src = reinterpret_cast<jpeg_source_mgr*>(&priv_->source_mgr);
+
+    /* Read header */
+    jpeg_read_header(&priv_->cinfo, TRUE);
+
+    jpeg_start_decompress(&priv_->cinfo);
+
+    return true;
+}
+
+void
+JPEGReader::finish()
+{
+    jpeg_destroy_decompress(&priv_->cinfo);
+}
+
+
+

=== added file 'src/image-reader.h'
--- src/image-reader.h	1970-01-01 00:00:00 +0000
+++ src/image-reader.h	2012-06-27 16:20:24 +0000
@@ -0,0 +1,77 @@
+/*
+ * Copyright © 2012 Linaro Limited
+ *
+ * This file is part of the glmark2 OpenGL (ES) 2.0 benchmark.
+ *
+ * glmark2 is free software: you can redistribute it and/or modify it under the
+ * terms of the GNU General Public License as published by the Free Software
+ * Foundation, either version 3 of the License, or (at your option) any later
+ * version.
+ *
+ * glmark2 is distributed in the hope that it will be useful, but WITHOUT ANY
+ * WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+ * FOR A PARTICULAR PURPOSE.  See the GNU General Public License for more
+ * details.
+ *
+ * You should have received a copy of the GNU General Public License along with
+ * glmark2.  If not, see <http://www.gnu.org/licenses/>.
+ *
+ * Authors:
+ *  Alexandros Frantzis
+ */
+#include <string>
+
+class ImageReader
+{
+public:
+    virtual bool error() = 0;
+    virtual bool nextRow(unsigned char *dst) = 0;
+    virtual unsigned int width() const = 0;
+    virtual unsigned int height() const = 0;
+    virtual unsigned int pixelBytes() const = 0;
+    virtual ~ImageReader() {}
+};
+
+class PNGReaderPrivate;
+
+class PNGReader : public ImageReader
+{
+public:
+    PNGReader(const std::string& filename);
+
+    virtual ~PNGReader();
+    bool error();
+    bool nextRow(unsigned char *dst);
+
+    unsigned int width() const;
+    unsigned int height() const;
+    unsigned int pixelBytes() const;
+
+private:
+    bool init(const std::string& filename);
+    void finish();
+
+    PNGReaderPrivate *priv_;
+};
+
+class JPEGReaderPrivate;
+
+class JPEGReader : public ImageReader
+{
+public:
+    JPEGReader(const std::string& filename);
+
+    virtual ~JPEGReader();
+    bool error();
+    bool nextRow(unsigned char *dst);
+    unsigned int width() const;
+    unsigned int height() const;
+    unsigned int pixelBytes() const;
+
+private:
+    bool init(const std::string& filename);
+    void finish();
+
+    JPEGReaderPrivate *priv_;
+};
+

=== added directory 'src/libjpeg-turbo'
=== added file 'src/libjpeg-turbo/README'
--- src/libjpeg-turbo/README	1970-01-01 00:00:00 +0000
+++ src/libjpeg-turbo/README	2012-06-27 16:20:24 +0000
@@ -0,0 +1,290 @@
+libjpeg-turbo note:  This file contains portions of the libjpeg v6b and v8
+README files, with additional wordsmithing by The libjpeg-turbo Project.
+It is included only for reference, as some parts of it may not apply to
+libjpeg-turbo.  Please see README-turbo.txt for information specific to
+libjpeg-turbo.
+
+
+The Independent JPEG Group's JPEG software
+==========================================
+
+This distribution contains a release of the Independent JPEG Group's free JPEG
+software.  You are welcome to redistribute this software and to use it for any
+purpose, subject to the conditions under LEGAL ISSUES, below.
+
+This software is the work of Tom Lane, Guido Vollbeding, Philip Gladstone,
+Bill Allombert, Jim Boucher, Lee Crocker, Bob Friesenhahn, Ben Jackson,
+Julian Minguillon, Luis Ortiz, George Phillips, Davide Rossi, Ge' Weijers,
+and other members of the Independent JPEG Group.
+
+IJG is not affiliated with the official ISO JPEG standards committee.
+
+
+DOCUMENTATION ROADMAP
+=====================
+
+This file contains the following sections:
+
+OVERVIEW            General description of JPEG and the IJG software.
+LEGAL ISSUES        Copyright, lack of warranty, terms of distribution.
+REFERENCES          Where to learn more about JPEG.
+ARCHIVE LOCATIONS   Where to find newer versions of this software.
+FILE FORMAT WARS    Software *not* to get.
+TO DO               Plans for future IJG releases.
+
+Other documentation files in the distribution are:
+
+User documentation:
+  install.txt       How to configure and install the IJG software.
+  usage.txt         Usage instructions for cjpeg, djpeg, jpegtran,
+                    rdjpgcom, and wrjpgcom.
+  *.1               Unix-style man pages for programs (same info as usage.txt).
+  wizard.txt        Advanced usage instructions for JPEG wizards only.
+  change.log        Version-to-version change highlights.
+Programmer and internal documentation:
+  libjpeg.txt       How to use the JPEG library in your own programs.
+  example.c         Sample code for calling the JPEG library.
+  structure.txt     Overview of the JPEG library's internal structure.
+  filelist.txt      Road map of IJG files.
+  coderules.txt     Coding style rules --- please read if you contribute code.
+
+Please read at least the files install.txt and usage.txt.  Some information
+can also be found in the JPEG FAQ (Frequently Asked Questions) article.  See
+ARCHIVE LOCATIONS below to find out where to obtain the FAQ article.
+
+If you want to understand how the JPEG code works, we suggest reading one or
+more of the REFERENCES, then looking at the documentation files (in roughly
+the order listed) before diving into the code.
+
+
+OVERVIEW
+========
+
+This package contains C software to implement JPEG image encoding, decoding,
+and transcoding.  JPEG (pronounced "jay-peg") is a standardized compression
+method for full-color and gray-scale images.  JPEG's strong suit is compressing
+photographic images or other types of images that have smooth color and
+brightness transitions between neighboring pixels.  Images with sharp lines or
+other abrupt features may not compress well with JPEG, and a higher JPEG
+quality may have to be used to avoid visible compression artifacts with such
+images.
+
+JPEG is lossy, meaning that the output pixels are not necessarily identical to
+the input pixels.  However, on photographic content and other "smooth" images,
+very good compression ratios can be obtained with no visible compression
+artifacts, and extremely high compression ratios are possible if you are
+willing to sacrifice image quality (by reducing the "quality" setting in the
+compressor.)
+
+This software implements JPEG baseline, extended-sequential, and progressive
+compression processes.  Provision is made for supporting all variants of these
+processes, although some uncommon parameter settings aren't implemented yet.
+We have made no provision for supporting the hierarchical or lossless
+processes defined in the standard.
+
+We provide a set of library routines for reading and writing JPEG image files,
+plus two sample applications "cjpeg" and "djpeg", which use the library to
+perform conversion between JPEG and some other popular image file formats.
+The library is intended to be reused in other applications.
+
+In order to support file conversion and viewing software, we have included
+considerable functionality beyond the bare JPEG coding/decoding capability;
+for example, the color quantization modules are not strictly part of JPEG
+decoding, but they are essential for output to colormapped file formats or
+colormapped displays.  These extra functions can be compiled out of the
+library if not required for a particular application.
+
+We have also included "jpegtran", a utility for lossless transcoding between
+different JPEG processes, and "rdjpgcom" and "wrjpgcom", two simple
+applications for inserting and extracting textual comments in JFIF files.
+
+The emphasis in designing this software has been on achieving portability and
+flexibility, while also making it fast enough to be useful.  In particular,
+the software is not intended to be read as a tutorial on JPEG.  (See the
+REFERENCES section for introductory material.)  Rather, it is intended to
+be reliable, portable, industrial-strength code.  We do not claim to have
+achieved that goal in every aspect of the software, but we strive for it.
+
+We welcome the use of this software as a component of commercial products.
+No royalty is required, but we do ask for an acknowledgement in product
+documentation, as described under LEGAL ISSUES.
+
+
+LEGAL ISSUES
+============
+
+In plain English:
+
+1. We don't promise that this software works.  (But if you find any bugs,
+   please let us know!)
+2. You can use this software for whatever you want.  You don't have to pay us.
+3. You may not pretend that you wrote this software.  If you use it in a
+   program, you must acknowledge somewhere in your documentation that
+   you've used the IJG code.
+
+In legalese:
+
+The authors make NO WARRANTY or representation, either express or implied,
+with respect to this software, its quality, accuracy, merchantability, or
+fitness for a particular purpose.  This software is provided "AS IS", and you,
+its user, assume the entire risk as to its quality and accuracy.
+
+This software is copyright (C) 1991-2010, Thomas G. Lane, Guido Vollbeding.
+All Rights Reserved except as specified below.
+
+Permission is hereby granted to use, copy, modify, and distribute this
+software (or portions thereof) for any purpose, without fee, subject to these
+conditions:
+(1) If any part of the source code for this software is distributed, then this
+README file must be included, with this copyright and no-warranty notice
+unaltered; and any additions, deletions, or changes to the original files
+must be clearly indicated in accompanying documentation.
+(2) If only executable code is distributed, then the accompanying
+documentation must state that "this software is based in part on the work of
+the Independent JPEG Group".
+(3) Permission for use of this software is granted only if the user accepts
+full responsibility for any undesirable consequences; the authors accept
+NO LIABILITY for damages of any kind.
+
+These conditions apply to any software derived from or based on the IJG code,
+not just to the unmodified library.  If you use our work, you ought to
+acknowledge us.
+
+Permission is NOT granted for the use of any IJG author's name or company name
+in advertising or publicity relating to this software or products derived from
+it.  This software may be referred to only as "the Independent JPEG Group's
+software".
+
+We specifically permit and encourage the use of this software as the basis of
+commercial products, provided that all warranty or liability claims are
+assumed by the product vendor.
+
+
+ansi2knr.c is included in this distribution by permission of L. Peter Deutsch,
+sole proprietor of its copyright holder, Aladdin Enterprises of Menlo Park, CA.
+ansi2knr.c is NOT covered by the above copyright and conditions, but instead
+by the usual distribution terms of the Free Software Foundation; principally,
+that you must include source code if you redistribute it.  (See the file
+ansi2knr.c for full details.)  However, since ansi2knr.c is not needed as part
+of any program generated from the IJG code, this does not limit you more than
+the foregoing paragraphs do.
+
+The Unix configuration script "configure" was produced with GNU Autoconf.
+It is copyright by the Free Software Foundation but is freely distributable.
+The same holds for its supporting scripts (config.guess, config.sub,
+ltmain.sh).  Another support script, install-sh, is copyright by X Consortium
+but is also freely distributable.
+
+The IJG distribution formerly included code to read and write GIF files.
+To avoid entanglement with the Unisys LZW patent, GIF reading support has
+been removed altogether, and the GIF writer has been simplified to produce
+"uncompressed GIFs".  This technique does not use the LZW algorithm; the
+resulting GIF files are larger than usual, but are readable by all standard
+GIF decoders.
+
+We are required to state that
+    "The Graphics Interchange Format(c) is the Copyright property of
+    CompuServe Incorporated.  GIF(sm) is a Service Mark property of
+    CompuServe Incorporated."
+
+
+REFERENCES
+==========
+
+We recommend reading one or more of these references before trying to
+understand the innards of the JPEG software.
+
+The best short technical introduction to the JPEG compression algorithm is
+	Wallace, Gregory K.  "The JPEG Still Picture Compression Standard",
+	Communications of the ACM, April 1991 (vol. 34 no. 4), pp. 30-44.
+(Adjacent articles in that issue discuss MPEG motion picture compression,
+applications of JPEG, and related topics.)  If you don't have the CACM issue
+handy, a PostScript file containing a revised version of Wallace's article is
+available at http://www.ijg.org/files/wallace.ps.gz.  The file (actually
+a preprint for an article that appeared in IEEE Trans. Consumer Electronics)
+omits the sample images that appeared in CACM, but it includes corrections
+and some added material.  Note: the Wallace article is copyright ACM and IEEE,
+and it may not be used for commercial purposes.
+
+A somewhat less technical, more leisurely introduction to JPEG can be found in
+"The Data Compression Book" by Mark Nelson and Jean-loup Gailly, published by
+M&T Books (New York), 2nd ed. 1996, ISBN 1-55851-434-1.  This book provides
+good explanations and example C code for a multitude of compression methods
+including JPEG.  It is an excellent source if you are comfortable reading C
+code but don't know much about data compression in general.  The book's JPEG
+sample code is far from industrial-strength, but when you are ready to look
+at a full implementation, you've got one here...
+
+The best currently available description of JPEG is the textbook "JPEG Still
+Image Data Compression Standard" by William B. Pennebaker and Joan L.
+Mitchell, published by Van Nostrand Reinhold, 1993, ISBN 0-442-01272-1.
+Price US$59.95, 638 pp.  The book includes the complete text of the ISO JPEG
+standards (DIS 10918-1 and draft DIS 10918-2).
+
+The original JPEG standard is divided into two parts, Part 1 being the actual
+specification, while Part 2 covers compliance testing methods.  Part 1 is
+titled "Digital Compression and Coding of Continuous-tone Still Images,
+Part 1: Requirements and guidelines" and has document numbers ISO/IEC IS
+10918-1, ITU-T T.81.  Part 2 is titled "Digital Compression and Coding of
+Continuous-tone Still Images, Part 2: Compliance testing" and has document
+numbers ISO/IEC IS 10918-2, ITU-T T.83.
+
+The JPEG standard does not specify all details of an interchangeable file
+format.  For the omitted details we follow the "JFIF" conventions, revision
+1.02.  JFIF 1.02 has been adopted as an Ecma International Technical Report
+and thus received a formal publication status.  It is available as a free
+download in PDF format from
+http://www.ecma-international.org/publications/techreports/E-TR-098.htm.
+A PostScript version of the JFIF document is available at
+http://www.ijg.org/files/jfif.ps.gz.  There is also a plain text version at
+http://www.ijg.org/files/jfif.txt.gz, but it is missing the figures.
+
+The TIFF 6.0 file format specification can be obtained by FTP from
+ftp://ftp.sgi.com/graphics/tiff/TIFF6.ps.gz.  The JPEG incorporation scheme
+found in the TIFF 6.0 spec of 3-June-92 has a number of serious problems.
+IJG does not recommend use of the TIFF 6.0 design (TIFF Compression tag 6).
+Instead, we recommend the JPEG design proposed by TIFF Technical Note #2
+(Compression tag 7).  Copies of this Note can be obtained from
+http://www.ijg.org/files/.  It is expected that the next revision
+of the TIFF spec will replace the 6.0 JPEG design with the Note's design.
+Although IJG's own code does not support TIFF/JPEG, the free libtiff library
+uses our library to implement TIFF/JPEG per the Note.
+
+
+ARCHIVE LOCATIONS
+=================
+
+The "official" archive site for this software is www.ijg.org.
+The most recent released version can always be found there in
+directory "files".  This particular version will be archived as
+http://www.ijg.org/files/jpegsrc.v8d.tar.gz, and in Windows-compatible
+"zip" archive format as http://www.ijg.org/files/jpegsr8d.zip.
+
+The JPEG FAQ (Frequently Asked Questions) article is a source of some
+general information about JPEG.
+It is available on the World Wide Web at http://www.faqs.org/faqs/jpeg-faq/
+and other news.answers archive sites, including the official news.answers
+archive at rtfm.mit.edu: ftp://rtfm.mit.edu/pub/usenet/news.answers/jpeg-faq/.
+If you don't have Web or FTP access, send e-mail to [e-mail]
+with body
+	send usenet/news.answers/jpeg-faq/part1
+	send usenet/news.answers/jpeg-faq/part2
+
+
+FILE FORMAT WARS
+================
+
+The ISO JPEG standards committee actually promotes different formats like
+"JPEG 2000" or "JPEG XR", which are incompatible with original DCT-based
+JPEG.  IJG therefore does not support these formats (see REFERENCES).  Indeed,
+one of the original reasons for developing this free software was to help
+force convergence on common, interoperable format standards for JPEG files.
+Don't use an incompatible file format!
+(In any case, our decoder will remain capable of reading existing JPEG
+image files indefinitely.)
+
+
+TO DO
+=====
+
+Please send bug reports, offers of help, etc. to [e-mail].

=== added file 'src/libjpeg-turbo/README-turbo.txt'
--- src/libjpeg-turbo/README-turbo.txt	1970-01-01 00:00:00 +0000
+++ src/libjpeg-turbo/README-turbo.txt	2012-06-27 16:20:24 +0000
@@ -0,0 +1,361 @@
+*******************************************************************************
+**     Background
+*******************************************************************************
+
+libjpeg-turbo is a derivative of libjpeg that uses SIMD instructions (MMX,
+SSE2, NEON) to accelerate baseline JPEG compression and decompression on x86,
+x86-64, and ARM systems.  On such systems, libjpeg-turbo is generally 2-4x as
+fast as the unmodified version of libjpeg, all else being equal.
+
+libjpeg-turbo was originally based on libjpeg/SIMD by Miyasaka Masaru, but
+the TigerVNC and VirtualGL projects made numerous enhancements to the codec in
+2009, including improved support for Mac OS X, 64-bit support, support for
+32-bit and big-endian pixel formats (RGBX, XBGR, etc.), accelerated Huffman
+encoding/decoding, and various bug fixes.  The goal was to produce a fully
+open-source codec that could replace the partially closed-source TurboJPEG/IPP
+codec used by VirtualGL and TurboVNC.  libjpeg-turbo generally achieves 80-120%
+of the performance of TurboJPEG/IPP.  It is faster in some areas but slower in
+others.
+
+In early 2010, libjpeg-turbo spun off into its own independent project, with
+the goal of making high-speed JPEG compression/decompression technology
+available to a broader range of users and developers.
+
+
+*******************************************************************************
+**     License
+*******************************************************************************
+
+Most of libjpeg-turbo inherits the non-restrictive, BSD-style license used by
+libjpeg (see README.)  The TurboJPEG/OSS wrapper (both C and Java versions) and
+associated test programs bear a similar license, which is reproduced below:
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+- Redistributions of source code must retain the above copyright notice,
+  this list of conditions and the following disclaimer.
+- Redistributions in binary form must reproduce the above copyright notice,
+  this list of conditions and the following disclaimer in the documentation
+  and/or other materials provided with the distribution.
+- Neither the name of the libjpeg-turbo Project nor the names of its
+  contributors may be used to endorse or promote products derived from this
+  software without specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS",
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED.  IN NO EVENT SHALL THE COPYRIGHT HOLDERS OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+POSSIBILITY OF SUCH DAMAGE.
+
+
+*******************************************************************************
+**     Using libjpeg-turbo
+*******************************************************************************
+
+libjpeg-turbo includes two APIs that can be used to compress and decompress
+JPEG images:
+
+  TurboJPEG API:  This API provides an easy-to-use interface for compressing
+  and decompressing JPEG images in memory.  It also provides some functionality
+  that would not be straightforward to achieve using the underlying libjpeg
+  API, such as generating planar YUV images and performing multiple
+  simultaneous lossless transforms on an image.  The Java interface for
+  libjpeg-turbo is written on top of the TurboJPEG API.
+
+  libjpeg API:  This is the de facto industry-standard API for compressing and
+  decompressing JPEG images.  It is more difficult to use than the TurboJPEG
+  API but also more powerful.  libjpeg-turbo is both API/ABI-compatible and
+  mathematically compatible with libjpeg v6b.  It can also optionally be
+  configured to be API/ABI-compatible with libjpeg v7 and v8 (see below.)
+
+
+=============================
+Replacing libjpeg at Run Time
+=============================
+
+If a Unix application is dynamically linked with libjpeg, then you can replace
+libjpeg with libjpeg-turbo at run time by manipulating LD_LIBRARY_PATH.
+For instance:
+
+  [Using libjpeg]
+  > time cjpeg <vgl_5674_0098.ppm >vgl_5674_0098.jpg
+  real  0m0.392s
+  user  0m0.074s
+  sys   0m0.020s
+
+  [Using libjpeg-turbo]
+  > export LD_LIBRARY_PATH=/opt/libjpeg-turbo/{lib}:$LD_LIBRARY_PATH
+  > time cjpeg <vgl_5674_0098.ppm >vgl_5674_0098.jpg
+  real  0m0.109s
+  user  0m0.029s
+  sys   0m0.010s
+
+NOTE: {lib} can be lib, lib32, lib64, or lib/64, depending on the O/S and
+architecture.
+
+System administrators can also replace the libjpeg sym links in /usr/{lib} with
+links to the libjpeg-turbo dynamic library located in /opt/libjpeg-turbo/{lib}.
+This will effectively accelerate every application that uses the libjpeg
+dynamic library on the system.
+
+The libjpeg-turbo SDK for Visual C++ installs the libjpeg-turbo DLL
+(jpeg62.dll, jpeg7.dll, or jpeg8.dll, depending on whether it was built with
+libjpeg v6b, v7, or v8 emulation) into c:\libjpeg-turbo[64]\bin, and the PATH
+environment variable can be modified such that this directory is searched
+before any others that might contain a libjpeg DLL.  However, if a libjpeg
+DLL exists in an application's install directory, then Windows will load this
+DLL first whenever the application is launched.  Thus, if an application ships
+with jpeg62.dll, jpeg7.dll, or jpeg8.dll, then back up the application's
+version of this DLL and copy c:\libjpeg-turbo[64]\bin\jpeg*.dll into the
+application's install directory to accelerate it.
+
+The version of the libjpeg-turbo DLL distributed in the libjpeg-turbo SDK for
+Visual C++ requires the Visual C++ 2008 C run-time DLL (msvcr90.dll).
+msvcr90.dll ships with more recent versions of Windows, but users of older
+Windows releases can obtain it from the Visual C++ 2008 Redistributable
+Package, which is available as a free download from Microsoft's web site.
+
+NOTE:  Features of libjpeg that require passing a C run-time structure, such
+as a file handle, from an application to libjpeg will probably not work with
+the version of the libjpeg-turbo DLL distributed in the libjpeg-turbo SDK for
+Visual C++, unless the application is also built to use the Visual C++ 2008 C
+run-time DLL.  In particular, this affects jpeg_stdio_dest() and
+jpeg_stdio_src().
+
+Mac applications typically embed their own copies of the libjpeg dylib inside
+the (hidden) application bundle, so it is not possible to globally replace
+libjpeg on OS X systems.  If an application uses a shared library version of
+libjpeg, then it may be possible to replace the application's version of it.
+This would generally involve copying libjpeg.*.dylib from libjpeg-turbo into
+the appropriate place in the application bundle and using install_name_tool to
+repoint the dylib to the new directory.  This requires an advanced knowledge of
+OS X and would not survive an upgrade or a re-install of the application.
+Thus, it is not recommended for most users.
+
+=======================
+Replacing TurboJPEG/IPP
+=======================
+
+libjpeg-turbo is a drop-in replacement for the TurboJPEG/IPP SDK used by
+VirtualGL 2.1.x and TurboVNC 0.6 (and prior.)  libjpeg-turbo contains a wrapper
+library (TurboJPEG/OSS) that emulates the TurboJPEG API using libjpeg-turbo
+instead of the closed-source Intel Performance Primitives.  You can replace the
+TurboJPEG/IPP package on Linux systems with the libjpeg-turbo package in order
+to make existing releases of VirtualGL 2.1.x and TurboVNC 0.x use the new codec
+at run time.  Note that the 64-bit libjpeg-turbo packages contain only 64-bit
+binaries, whereas the TurboJPEG/IPP 64-bit packages contained both 64-bit and
+32-bit binaries.  Thus, to replace a TurboJPEG/IPP 64-bit package, install
+both the 64-bit and 32-bit versions of libjpeg-turbo.
+
+You can also build the VirtualGL 2.1.x and TurboVNC 0.6 source code with
+the libjpeg-turbo SDK instead of TurboJPEG/IPP.  It should work identically.
+libjpeg-turbo also includes static library versions of TurboJPEG/OSS, which
+are used to build VirtualGL 2.2 and TurboVNC 1.0 and later.
+
+========================================
+Using libjpeg-turbo in Your Own Programs
+========================================
+
+For the most part, libjpeg-turbo should work identically to libjpeg, so in
+most cases, an application can be built against libjpeg and then run against
+libjpeg-turbo.  On Unix systems (including Cygwin), you can build against
+libjpeg-turbo instead of libjpeg by setting
+
+  CPATH=/opt/libjpeg-turbo/include
+  and
+  LIBRARY_PATH=/opt/libjpeg-turbo/{lib}
+
+({lib} = lib32 or lib64, depending on whether you are building a 32-bit or a
+64-bit application.)
+
+If using MinGW, then set
+
+  CPATH=/c/libjpeg-turbo-gcc[64]/include
+  and
+  LIBRARY_PATH=/c/libjpeg-turbo-gcc[64]/lib
+
+Building against libjpeg-turbo is useful, for instance, if you want to build an
+application that leverages the libjpeg-turbo colorspace extensions (see below.)
+On Linux and Solaris systems, you would still need to manipulate
+LD_LIBRARY_PATH or create appropriate sym links to use libjpeg-turbo at run
+time.  On such systems, you can pass -R /opt/libjpeg-turbo/{lib} to the linker
+to force the use of libjpeg-turbo at run time rather than libjpeg (also useful
+if you want to leverage the colorspace extensions), or you can link against the
+libjpeg-turbo static library.
+
+To force a Linux, Solaris, or MinGW application to link against the static
+version of libjpeg-turbo, you can use the following linker options:
+
+  -Wl,-Bstatic -ljpeg -Wl,-Bdynamic
+
+On OS X, simply add /opt/libjpeg-turbo/lib/libjpeg.a to the linker command
+line (this also works on Linux and Solaris.)
+
+To build Visual C++ applications using libjpeg-turbo, add
+c:\libjpeg-turbo[64]\include to the system or user INCLUDE environment
+variable and c:\libjpeg-turbo[64]\lib to the system or user LIB environment
+variable, and then link against either jpeg.lib (to use the DLL version of
+libjpeg-turbo) or jpeg-static.lib (to use the static version of libjpeg-turbo.)
+
+=====================
+Colorspace Extensions
+=====================
+
+libjpeg-turbo includes extensions that allow JPEG images to be compressed
+directly from (and decompressed directly to) buffers that use BGR, BGRX,
+RGBX, XBGR, and XRGB pixel ordering.  This is implemented with ten new
+colorspace constants:
+
+  JCS_EXT_RGB   /* red/green/blue */
+  JCS_EXT_RGBX  /* red/green/blue/x */
+  JCS_EXT_BGR   /* blue/green/red */
+  JCS_EXT_BGRX  /* blue/green/red/x */
+  JCS_EXT_XBGR  /* x/blue/green/red */
+  JCS_EXT_XRGB  /* x/red/green/blue */
+  JCS_EXT_RGBA  /* red/green/blue/alpha */
+  JCS_EXT_BGRA  /* blue/green/red/alpha */
+  JCS_EXT_ABGR  /* alpha/blue/green/red */
+  JCS_EXT_ARGB  /* alpha/red/green/blue */
+
+Setting cinfo.in_color_space (compression) or cinfo.out_color_space
+(decompression) to one of these values will cause libjpeg-turbo to read the
+red, green, and blue values from (or write them to) the appropriate position in
+the pixel when compressing from/decompressing to an RGB buffer.
+
+Your application can check for the existence of these extensions at compile
+time with:
+
+  #ifdef JCS_EXTENSIONS
+
+At run time, attempting to use these extensions with a version of libjpeg
+that doesn't support them will result in a "Bogus input colorspace" error.
+
+When using the RGBX, BGRX, XBGR, and XRGB colorspaces during decompression, the
+X byte is undefined, and in order to ensure the best performance, libjpeg-turbo
+can set that byte to whatever value it wishes.  If an application expects the X
+byte to be used as an alpha channel, then it should specify JCS_EXT_RGBA,
+JCS_EXT_BGRA, JCS_EXT_ABGR, or JCS_EXT_ARGB.  When these colorspace constants
+are used, the X byte is guaranteed to be 0xFF, which is interpreted as opaque.
+
+Your application can check for the existence of the alpha channel colorspace
+extensions at compile time with:
+
+  #ifdef JCS_ALPHA_EXTENSIONS
+
+jcstest.c, located in the libjpeg-turbo source tree, demonstrates how to check
+for the existence of the colorspace extensions at compile time and run time.
+
+=================================
+libjpeg v7 and v8 API/ABI support
+=================================
+
+With libjpeg v7 and v8, new features were added that necessitated extending the
+compression and decompression structures.  Unfortunately, due to the exposed
+nature of those structures, extending them also necessitated breaking backward
+ABI compatibility with previous libjpeg releases.  Thus, programs that are
+built to use libjpeg v7 or v8 did not work with libjpeg-turbo, since it is
+based on the libjpeg v6b code base.  Although libjpeg v7 and v8 are still not
+as widely used as v6b, enough programs (including a few Linux distros) have
+made the switch that it was desirable to provide support for the libjpeg v7/v8
+API/ABI in libjpeg-turbo.  Although libjpeg-turbo can now be configured as a
+drop-in replacement for libjpeg v7 or v8, it should be noted that not all of
+the features in libjpeg v7 and v8 are supported (see below.)
+
+By passing an argument of --with-jpeg7 or --with-jpeg8 to configure, or an
+argument of -DWITH_JPEG7=1 or -DWITH_JPEG8=1 to cmake, you can build a version
+of libjpeg-turbo that emulates the libjpeg v7 or v8 API/ABI, so that programs
+that are built against libjpeg v7 or v8 can be run with libjpeg-turbo.  The
+following section describes which libjpeg v7+ features are supported and which
+aren't.
+
+libjpeg v7 and v8 Features:
+---------------------------
+
+Fully supported:
+
+-- cjpeg: Separate quality settings for luminance and chrominance
+   Note that the libpjeg v7+ API was extended to accommodate this feature only
+   for convenience purposes.  It has always been possible to implement this
+   feature with libjpeg v6b (see rdswitch.c for an example.)
+
+-- cjpeg: 32-bit BMP support
+
+-- jpegtran: lossless cropping
+
+-- jpegtran: -perfect option
+
+-- rdjpgcom: -raw option
+
+-- rdjpgcom: locale awareness
+
+
+Fully supported when using libjpeg v7/v8 emulation:
+
+-- libjpeg: In-memory source and destination managers
+
+
+Not supported:
+
+-- libjpeg: DCT scaling in compressor
+   cinfo.scale_num and cinfo.scale_denom are silently ignored.
+   There is no technical reason why DCT scaling cannot be supported, but
+   without the SmartScale extension (see below), it would only be able to
+   down-scale using ratios of 1/2, 8/15, 4/7, 8/13, 2/3, 8/11, 4/5, and 8/9,
+   which is of limited usefulness.
+
+-- libjpeg: SmartScale
+   cinfo.block_size is silently ignored.
+   SmartScale is an extension to the JPEG format that allows for DCT block
+   sizes other than 8x8.  It would be difficult to support this feature while
+   retaining backward compatibility with libjpeg v6b.
+
+-- libjpeg: IDCT scaling extensions in decompressor
+   libjpeg-turbo still supports IDCT scaling with scaling factors of 1/2, 1/4,
+   and 1/8 (same as libjpeg v6b.)
+
+-- libjpeg: Fancy downsampling in compressor
+   cinfo.do_fancy_downsampling is silently ignored.
+   This requires the DCT scaling feature, which is not supported.
+
+-- jpegtran: Scaling
+   This requires both the DCT scaling and SmartScale features, which are not
+   supported.
+
+-- Lossless RGB JPEG files
+   This requires the SmartScale feature, which is not supported.
+
+
+*******************************************************************************
+**     Performance pitfalls
+*******************************************************************************
+
+===============
+Restart Markers
+===============
+
+The optimized Huffman decoder in libjpeg-turbo does not handle restart markers
+in a way that makes the rest of the libjpeg infrastructure happy, so it is
+necessary to use the slow Huffman decoder when decompressing a JPEG image that
+has restart markers.  This can cause the decompression performance to drop by
+as much as 20%, but the performance will still be much greater than that of
+libjpeg.  Many consumer packages, such as PhotoShop, use restart markers when
+generating JPEG images, so images generated by those programs will experience
+this issue.
+
+===============================================
+Fast Integer Forward DCT at High Quality Levels
+===============================================
+
+The algorithm used by the SIMD-accelerated quantization function cannot produce
+correct results whenever the fast integer forward DCT is used along with a JPEG
+quality of 98-100.  Thus, libjpeg-turbo must use the non-SIMD quantization
+function in those cases.  This causes performance to drop by as much as 40%.
+It is therefore strongly advised that you use the slow integer forward DCT
+whenever encoding images with a JPEG quality of 98 or higher.

=== added file 'src/libjpeg-turbo/config.h'
--- src/libjpeg-turbo/config.h	1970-01-01 00:00:00 +0000
+++ src/libjpeg-turbo/config.h	2012-06-27 16:20:24 +0000
@@ -0,0 +1,137 @@
+/* config.h.  Generated from config.h.in by configure.  */
+/* config.h.in.  Generated from configure.ac by autoheader.  */
+
+/* Build number */
+#define BUILD "20120626"
+
+/* Support arithmetic encoding */
+#define C_ARITH_CODING_SUPPORTED 1
+
+/* Support arithmetic decoding */
+#define D_ARITH_CODING_SUPPORTED 1
+
+/* Define to 1 if you have the <dlfcn.h> header file. */
+#define HAVE_DLFCN_H 1
+
+/* Define to 1 if you have the <inttypes.h> header file. */
+#define HAVE_INTTYPES_H 1
+
+/* Define to 1 if you have the <jni.h> header file. */
+/* #undef HAVE_JNI_H */
+
+/* Define to 1 if you have the `memcpy' function. */
+#define HAVE_MEMCPY 1
+
+/* Define to 1 if you have the <memory.h> header file. */
+#define HAVE_MEMORY_H 1
+
+/* Define to 1 if you have the `memset' function. */
+#define HAVE_MEMSET 1
+
+/* Define if your compiler supports prototypes */
+#define HAVE_PROTOTYPES 1
+
+/* Define to 1 if you have the <stddef.h> header file. */
+#define HAVE_STDDEF_H 1
+
+/* Define to 1 if you have the <stdint.h> header file. */
+#define HAVE_STDINT_H 1
+
+/* Define to 1 if you have the <stdlib.h> header file. */
+#define HAVE_STDLIB_H 1
+
+/* Define to 1 if you have the <strings.h> header file. */
+#define HAVE_STRINGS_H 1
+
+/* Define to 1 if you have the <string.h> header file. */
+#define HAVE_STRING_H 1
+
+/* Define to 1 if you have the <sys/stat.h> header file. */
+#define HAVE_SYS_STAT_H 1
+
+/* Define to 1 if you have the <sys/types.h> header file. */
+#define HAVE_SYS_TYPES_H 1
+
+/* Define to 1 if you have the <unistd.h> header file. */
+#define HAVE_UNISTD_H 1
+
+/* Define to 1 if the system has the type `unsigned char'. */
+#define HAVE_UNSIGNED_CHAR 1
+
+/* Define to 1 if the system has the type `unsigned short'. */
+#define HAVE_UNSIGNED_SHORT 1
+
+/* Compiler does not support pointers to undefined structures. */
+/* #undef INCOMPLETE_TYPES_BROKEN */
+
+/* How to obtain function inlining. */
+#define INLINE __attribute__((always_inline))
+
+/* libjpeg API version */
+#define JPEG_LIB_VERSION 62
+
+/* libjpeg-turbo version */
+#define LIBJPEG_TURBO_VERSION 1.2.0
+
+/* Define to the sub-directory in which libtool stores uninstalled libraries.
+   */
+#define LT_OBJDIR ".libs/"
+
+/* Define if you have BSD-like bzero and bcopy */
+/* #undef NEED_BSD_STRINGS */
+
+/* Define if you need short function names */
+/* #undef NEED_SHORT_EXTERNAL_NAMES */
+
+/* Define if you have sys/types.h */
+#define NEED_SYS_TYPES_H 1
+
+/* Name of package */
+#define PACKAGE "libjpeg-turbo"
+
+/* Define to the address where bug reports for this package should be sent. */
+#define PACKAGE_BUGREPORT ""
+
+/* Define to the full name of this package. */
+#define PACKAGE_NAME "libjpeg-turbo"
+
+/* Define to the full name and version of this package. */
+#define PACKAGE_STRING "libjpeg-turbo 1.2.0"
+
+/* Define to the one symbol short name of this package. */
+#define PACKAGE_TARNAME "libjpeg-turbo"
+
+/* Define to the home page for this package. */
+#define PACKAGE_URL ""
+
+/* Define to the version of this package. */
+#define PACKAGE_VERSION "1.2.0"
+
+/* Define if shift is unsigned */
+/* #undef RIGHT_SHIFT_IS_UNSIGNED */
+
+/* Define to 1 if you have the ANSI C header files. */
+#define STDC_HEADERS 1
+
+/* Version number of package */
+#define VERSION "1.2.0"
+
+/* Use accelerated SIMD routines. */
+#define WITH_SIMD 1
+
+/* Define to 1 if type `char' is unsigned and you are not using gcc.  */
+#ifndef __CHAR_UNSIGNED__
+/* # undef __CHAR_UNSIGNED__ */
+#endif
+
+/* Define to empty if `const' does not conform to ANSI C. */
+/* #undef const */
+
+/* Define to `__inline__' or `__inline' if that's what the C compiler
+   calls it, or to nothing if 'inline' is not supported under any name.  */
+#ifndef __cplusplus
+/* #undef inline */
+#endif
+
+/* Define to `unsigned int' if <sys/types.h> does not define. */
+/* #undef size_t */

=== added file 'src/libjpeg-turbo/jaricom.c'
--- src/libjpeg-turbo/jaricom.c	1970-01-01 00:00:00 +0000
+++ src/libjpeg-turbo/jaricom.c	2012-06-27 16:20:24 +0000
@@ -0,0 +1,153 @@
+/*
+ * jaricom.c
+ *
+ * Developed 1997-2009 by Guido Vollbeding.
+ * This file is part of the Independent JPEG Group's software.
+ * For conditions of distribution and use, see the accompanying README file.
+ *
+ * This file contains probability estimation tables for common use in
+ * arithmetic entropy encoding and decoding routines.
+ *
+ * This data represents Table D.2 in the JPEG spec (ISO/IEC IS 10918-1
+ * and CCITT Recommendation ITU-T T.81) and Table 24 in the JBIG spec
+ * (ISO/IEC IS 11544 and CCITT Recommendation ITU-T T.82).
+ */
+
+#define JPEG_INTERNALS
+#include "jinclude.h"
+#include "jpeglib.h"
+
+/* The following #define specifies the packing of the four components
+ * into the compact INT32 representation.
+ * Note that this formula must match the actual arithmetic encoder
+ * and decoder implementation.  The implementation has to be changed
+ * if this formula is changed.
+ * The current organization is leaned on Markus Kuhn's JBIG
+ * implementation (jbig_tab.c).
+ */
+
+#define V(i,a,b,c,d) (((INT32)a << 16) | ((INT32)c << 8) | ((INT32)d << 7) | b)
+
+const INT32 jpeg_aritab[113+1] = {
+/*
+ * Index, Qe_Value, Next_Index_LPS, Next_Index_MPS, Switch_MPS
+ */
+  V(   0, 0x5a1d,   1,   1, 1 ),
+  V(   1, 0x2586,  14,   2, 0 ),
+  V(   2, 0x1114,  16,   3, 0 ),
+  V(   3, 0x080b,  18,   4, 0 ),
+  V(   4, 0x03d8,  20,   5, 0 ),
+  V(   5, 0x01da,  23,   6, 0 ),
+  V(   6, 0x00e5,  25,   7, 0 ),
+  V(   7, 0x006f,  28,   8, 0 ),
+  V(   8, 0x0036,  30,   9, 0 ),
+  V(   9, 0x001a,  33,  10, 0 ),
+  V(  10, 0x000d,  35,  11, 0 ),
+  V(  11, 0x0006,   9,  12, 0 ),
+  V(  12, 0x0003,  10,  13, 0 ),
+  V(  13, 0x0001,  12,  13, 0 ),
+  V(  14, 0x5a7f,  15,  15, 1 ),
+  V(  15, 0x3f25,  36,  16, 0 ),
+  V(  16, 0x2cf2,  38,  17, 0 ),
+  V(  17, 0x207c,  39,  18, 0 ),
+  V(  18, 0x17b9,  40,  19, 0 ),
+  V(  19, 0x1182,  42,  20, 0 ),
+  V(  20, 0x0cef,  43,  21, 0 ),
+  V(  21, 0x09a1,  45,  22, 0 ),
+  V(  22, 0x072f,  46,  23, 0 ),
+  V(  23, 0x055c,  48,  24, 0 ),
+  V(  24, 0x0406,  49,  25, 0 ),
+  V(  25, 0x0303,  51,  26, 0 ),
+  V(  26, 0x0240,  52,  27, 0 ),
+  V(  27, 0x01b1,  54,  28, 0 ),
+  V(  28, 0x0144,  56,  29, 0 ),
+  V(  29, 0x00f5,  57,  30, 0 ),
+  V(  30, 0x00b7,  59,  31, 0 ),
+  V(  31, 0x008a,  60,  32, 0 ),
+  V(  32, 0x0068,  62,  33, 0 ),
+  V(  33, 0x004e,  63,  34, 0 ),
+  V(  34, 0x003b,  32,  35, 0 ),
+  V(  35, 0x002c,  33,   9, 0 ),
+  V(  36, 0x5ae1,  37,  37, 1 ),
+  V(  37, 0x484c,  64,  38, 0 ),
+  V(  38, 0x3a0d,  65,  39, 0 ),
+  V(  39, 0x2ef1,  67,  40, 0 ),
+  V(  40, 0x261f,  68,  41, 0 ),
+  V(  41, 0x1f33,  69,  42, 0 ),
+  V(  42, 0x19a8,  70,  43, 0 ),
+  V(  43, 0x1518,  72,  44, 0 ),
+  V(  44, 0x1177,  73,  45, 0 ),
+  V(  45, 0x0e74,  74,  46, 0 ),
+  V(  46, 0x0bfb,  75,  47, 0 ),
+  V(  47, 0x09f8,  77,  48, 0 ),
+  V(  48, 0x0861,  78,  49, 0 ),
+  V(  49, 0x0706,  79,  50, 0 ),
+  V(  50, 0x05cd,  48,  51, 0 ),
+  V(  51, 0x04de,  50,  52, 0 ),
+  V(  52, 0x040f,  50,  53, 0 ),
+  V(  53, 0x0363,  51,  54, 0 ),
+  V(  54, 0x02d4,  52,  55, 0 ),
+  V(  55, 0x025c,  53,  56, 0 ),
+  V(  56, 0x01f8,  54,  57, 0 ),
+  V(  57, 0x01a4,  55,  58, 0 ),
+  V(  58, 0x0160,  56,  59, 0 ),
+  V(  59, 0x0125,  57,  60, 0 ),
+  V(  60, 0x00f6,  58,  61, 0 ),
+  V(  61, 0x00cb,  59,  62, 0 ),
+  V(  62, 0x00ab,  61,  63, 0 ),
+  V(  63, 0x008f,  61,  32, 0 ),
+  V(  64, 0x5b12,  65,  65, 1 ),
+  V(  65, 0x4d04,  80,  66, 0 ),
+  V(  66, 0x412c,  81,  67, 0 ),
+  V(  67, 0x37d8,  82,  68, 0 ),
+  V(  68, 0x2fe8,  83,  69, 0 ),
+  V(  69, 0x293c,  84,  70, 0 ),
+  V(  70, 0x2379,  86,  71, 0 ),
+  V(  71, 0x1edf,  87,  72, 0 ),
+  V(  72, 0x1aa9,  87,  73, 0 ),
+  V(  73, 0x174e,  72,  74, 0 ),
+  V(  74, 0x1424,  72,  75, 0 ),
+  V(  75, 0x119c,  74,  76, 0 ),
+  V(  76, 0x0f6b,  74,  77, 0 ),
+  V(  77, 0x0d51,  75,  78, 0 ),
+  V(  78, 0x0bb6,  77,  79, 0 ),
+  V(  79, 0x0a40,  77,  48, 0 ),
+  V(  80, 0x5832,  80,  81, 1 ),
+  V(  81, 0x4d1c,  88,  82, 0 ),
+  V(  82, 0x438e,  89,  83, 0 ),
+  V(  83, 0x3bdd,  90,  84, 0 ),
+  V(  84, 0x34ee,  91,  85, 0 ),
+  V(  85, 0x2eae,  92,  86, 0 ),
+  V(  86, 0x299a,  93,  87, 0 ),
+  V(  87, 0x2516,  86,  71, 0 ),
+  V(  88, 0x5570,  88,  89, 1 ),
+  V(  89, 0x4ca9,  95,  90, 0 ),
+  V(  90, 0x44d9,  96,  91, 0 ),
+  V(  91, 0x3e22,  97,  92, 0 ),
+  V(  92, 0x3824,  99,  93, 0 ),
+  V(  93, 0x32b4,  99,  94, 0 ),
+  V(  94, 0x2e17,  93,  86, 0 ),
+  V(  95, 0x56a8,  95,  96, 1 ),
+  V(  96, 0x4f46, 101,  97, 0 ),
+  V(  97, 0x47e5, 102,  98, 0 ),
+  V(  98, 0x41cf, 103,  99, 0 ),
+  V(  99, 0x3c3d, 104, 100, 0 ),
+  V( 100, 0x375e,  99,  93, 0 ),
+  V( 101, 0x5231, 105, 102, 0 ),
+  V( 102, 0x4c0f, 106, 103, 0 ),
+  V( 103, 0x4639, 107, 104, 0 ),
+  V( 104, 0x415e, 103,  99, 0 ),
+  V( 105, 0x5627, 105, 106, 1 ),
+  V( 106, 0x50e7, 108, 107, 0 ),
+  V( 107, 0x4b85, 109, 103, 0 ),
+  V( 108, 0x5597, 110, 109, 0 ),
+  V( 109, 0x504f, 111, 107, 0 ),
+  V( 110, 0x5a10, 110, 111, 1 ),
+  V( 111, 0x5522, 112, 109, 0 ),
+  V( 112, 0x59eb, 112, 111, 1 ),
+/*
+ * This last entry is used for fixed probability estimate of 0.5
+ * as recommended in Section 10.3 Table 5 of ITU-T Rec. T.851.
+ */
+  V( 113, 0x5a1d, 113, 113, 0 )
+};

=== added file 'src/libjpeg-turbo/jcapimin.c'
--- src/libjpeg-turbo/jcapimin.c	1970-01-01 00:00:00 +0000
+++ src/libjpeg-turbo/jcapimin.c	2012-06-27 16:20:24 +0000
@@ -0,0 +1,292 @@
+/*
+ * jcapimin.c
+ *
+ * Copyright (C) 1994-1998, Thomas G. Lane.
+ * Modified 2003-2010 by Guido Vollbeding.
+ * This file is part of the Independent JPEG Group's software.
+ * For conditions of distribution and use, see the accompanying README file.
+ *
+ * This file contains application interface code for the compression half
+ * of the JPEG library.  These are the "minimum" API routines that may be
+ * needed in either the normal full-compression case or the transcoding-only
+ * case.
+ *
+ * Most of the routines intended to be called directly by an application
+ * are in this file or in jcapistd.c.  But also see jcparam.c for
+ * parameter-setup helper routines, jcomapi.c for routines shared by
+ * compression and decompression, and jctrans.c for the transcoding case.
+ */
+
+#define JPEG_INTERNALS
+#include "jinclude.h"
+#include "jpeglib.h"
+
+
+/*
+ * Initialization of a JPEG compression object.
+ * The error manager must already be set up (in case memory manager fails).
+ */
+
+GLOBAL(void)
+jpeg_CreateCompress (j_compress_ptr cinfo, int version, size_t structsize)
+{
+  int i;
+
+  /* Guard against version mismatches between library and caller. */
+  cinfo->mem = NULL;		/* so jpeg_destroy knows mem mgr not called */
+  if (version != JPEG_LIB_VERSION)
+    ERREXIT2(cinfo, JERR_BAD_LIB_VERSION, JPEG_LIB_VERSION, version);
+  if (structsize != SIZEOF(struct jpeg_compress_struct))
+    ERREXIT2(cinfo, JERR_BAD_STRUCT_SIZE, 
+	     (int) SIZEOF(struct jpeg_compress_struct), (int) structsize);
+
+  /* For debugging purposes, we zero the whole master structure.
+   * But the application has already set the err pointer, and may have set
+   * client_data, so we have to save and restore those fields.
+   * Note: if application hasn't set client_data, tools like Purify may
+   * complain here.
+   */
+  {
+    struct jpeg_error_mgr * err = cinfo->err;
+    void * client_data = cinfo->client_data; /* ignore Purify complaint here */
+    MEMZERO(cinfo, SIZEOF(struct jpeg_compress_struct));
+    cinfo->err = err;
+    cinfo->client_data = client_data;
+  }
+  cinfo->is_decompressor = FALSE;
+
+  /* Initialize a memory manager instance for this object */
+  jinit_memory_mgr((j_common_ptr) cinfo);
+
+  /* Zero out pointers to permanent structures. */
+  cinfo->progress = NULL;
+  cinfo->dest = NULL;
+
+  cinfo->comp_info = NULL;
+
+  for (i = 0; i < NUM_QUANT_TBLS; i++) {
+    cinfo->quant_tbl_ptrs[i] = NULL;
+#if JPEG_LIB_VERSION >= 70
+    cinfo->q_scale_factor[i] = 100;
+#endif
+  }
+
+  for (i = 0; i < NUM_HUFF_TBLS; i++) {
+    cinfo->dc_huff_tbl_ptrs[i] = NULL;
+    cinfo->ac_huff_tbl_ptrs[i] = NULL;
+  }
+
+#if JPEG_LIB_VERSION >= 80
+  /* Must do it here for emit_dqt in case jpeg_write_tables is used */
+  cinfo->block_size = DCTSIZE;
+  cinfo->natural_order = jpeg_natural_order;
+  cinfo->lim_Se = DCTSIZE2-1;
+#endif
+
+  cinfo->script_space = NULL;
+
+  cinfo->input_gamma = 1.0;	/* in case application forgets */
+
+  /* OK, I'm ready */
+  cinfo->global_state = CSTATE_START;
+}
+
+
+/*
+ * Destruction of a JPEG compression object
+ */
+
+GLOBAL(void)
+jpeg_destroy_compress (j_compress_ptr cinfo)
+{
+  jpeg_destroy((j_common_ptr) cinfo); /* use common routine */
+}
+
+
+/*
+ * Abort processing of a JPEG compression operation,
+ * but don't destroy the object itself.
+ */
+
+GLOBAL(void)
+jpeg_abort_compress (j_compress_ptr cinfo)
+{
+  jpeg_abort((j_common_ptr) cinfo); /* use common routine */
+}
+
+
+/*
+ * Forcibly suppress or un-suppress all quantization and Huffman tables.
+ * Marks all currently defined tables as already written (if suppress)
+ * or not written (if !suppress).  This will control whether they get emitted
+ * by a subsequent jpeg_start_compress call.
+ *
+ * This routine is exported for use by applications that want to produce
+ * abbreviated JPEG datastreams.  It logically belongs in jcparam.c, but
+ * since it is called by jpeg_start_compress, we put it here --- otherwise
+ * jcparam.o would be linked whether the application used it or not.
+ */
+
+GLOBAL(void)
+jpeg_suppress_tables (j_compress_ptr cinfo, boolean suppress)
+{
+  int i;
+  JQUANT_TBL * qtbl;
+  JHUFF_TBL * htbl;
+
+  for (i = 0; i < NUM_QUANT_TBLS; i++) {
+    if ((qtbl = cinfo->quant_tbl_ptrs[i]) != NULL)
+      qtbl->sent_table = suppress;
+  }
+
+  for (i = 0; i < NUM_HUFF_TBLS; i++) {
+    if ((htbl = cinfo->dc_huff_tbl_ptrs[i]) != NULL)
+      htbl->sent_table = suppress;
+    if ((htbl = cinfo->ac_huff_tbl_ptrs[i]) != NULL)
+      htbl->sent_table = suppress;
+  }
+}
+
+
+/*
+ * Finish JPEG compression.
+ *
+ * If a multipass operating mode was selected, this may do a great deal of
+ * work including most of the actual output.
+ */
+
+GLOBAL(void)
+jpeg_finish_compress (j_compress_ptr cinfo)
+{
+  JDIMENSION iMCU_row;
+
+  if (cinfo->global_state == CSTATE_SCANNING ||
+      cinfo->global_state == CSTATE_RAW_OK) {
+    /* Terminate first pass */
+    if (cinfo->next_scanline < cinfo->image_height)
+      ERREXIT(cinfo, JERR_TOO_LITTLE_DATA);
+    (*cinfo->master->finish_pass) (cinfo);
+  } else if (cinfo->global_state != CSTATE_WRCOEFS)
+    ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state);
+  /* Perform any remaining passes */
+  while (! cinfo->master->is_last_pass) {
+    (*cinfo->master->prepare_for_pass) (cinfo);
+    for (iMCU_row = 0; iMCU_row < cinfo->total_iMCU_rows; iMCU_row++) {
+      if (cinfo->progress != NULL) {
+	cinfo->progress->pass_counter = (long) iMCU_row;
+	cinfo->progress->pass_limit = (long) cinfo->total_iMCU_rows;
+	(*cinfo->progress->progress_monitor) ((j_common_ptr) cinfo);
+      }
+      /* We bypass the main controller and invoke coef controller directly;
+       * all work is being done from the coefficient buffer.
+       */
+      if (! (*cinfo->coef->compress_data) (cinfo, (JSAMPIMAGE) NULL))
+	ERREXIT(cinfo, JERR_CANT_SUSPEND);
+    }
+    (*cinfo->master->finish_pass) (cinfo);
+  }
+  /* Write EOI, do final cleanup */
+  (*cinfo->marker->write_file_trailer) (cinfo);
+  (*cinfo->dest->term_destination) (cinfo);
+  /* We can use jpeg_abort to release memory and reset global_state */
+  jpeg_abort((j_common_ptr) cinfo);
+}
+
+
+/*
+ * Write a special marker.
+ * This is only recommended for writing COM or APPn markers.
+ * Must be called after jpeg_start_compress() and before
+ * first call to jpeg_write_scanlines() or jpeg_write_raw_data().
+ */
+
+GLOBAL(void)
+jpeg_write_marker (j_compress_ptr cinfo, int marker,
+		   const JOCTET *dataptr, unsigned int datalen)
+{
+  JMETHOD(void, write_marker_byte, (j_compress_ptr info, int val));
+
+  if (cinfo->next_scanline != 0 ||
+      (cinfo->global_state != CSTATE_SCANNING &&
+       cinfo->global_state != CSTATE_RAW_OK &&
+       cinfo->global_state != CSTATE_WRCOEFS))
+    ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state);
+
+  (*cinfo->marker->write_marker_header) (cinfo, marker, datalen);
+  write_marker_byte = cinfo->marker->write_marker_byte;	/* copy for speed */
+  while (datalen--) {
+    (*write_marker_byte) (cinfo, *dataptr);
+    dataptr++;
+  }
+}
+
+/* Same, but piecemeal. */
+
+GLOBAL(void)
+jpeg_write_m_header (j_compress_ptr cinfo, int marker, unsigned int datalen)
+{
+  if (cinfo->next_scanline != 0 ||
+      (cinfo->global_state != CSTATE_SCANNING &&
+       cinfo->global_state != CSTATE_RAW_OK &&
+       cinfo->global_state != CSTATE_WRCOEFS))
+    ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state);
+
+  (*cinfo->marker->write_marker_header) (cinfo, marker, datalen);
+}
+
+GLOBAL(void)
+jpeg_write_m_byte (j_compress_ptr cinfo, int val)
+{
+  (*cinfo->marker->write_marker_byte) (cinfo, val);
+}
+
+
+/*
+ * Alternate compression function: just write an abbreviated table file.
+ * Before calling this, all parameters and a data destination must be set up.
+ *
+ * To produce a pair of files containing abbreviated tables and abbreviated
+ * image data, one would proceed as follows:
+ *
+ *		initialize JPEG object
+ *		set JPEG parameters
+ *		set destination to table file
+ *		jpeg_write_tables(cinfo);
+ *		set destination to image file
+ *		jpeg_start_compress(cinfo, FALSE);
+ *		write data...
+ *		jpeg_finish_compress(cinfo);
+ *
+ * jpeg_write_tables has the side effect of marking all tables written
+ * (same as jpeg_suppress_tables(..., TRUE)).  Thus a subsequent start_compress
+ * will not re-emit the tables unless it is passed write_all_tables=TRUE.
+ */
+
+GLOBAL(void)
+jpeg_write_tables (j_compress_ptr cinfo)
+{
+  if (cinfo->global_state != CSTATE_START)
+    ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state);
+
+  /* (Re)initialize error mgr and destination modules */
+  (*cinfo->err->reset_error_mgr) ((j_common_ptr) cinfo);
+  (*cinfo->dest->init_destination) (cinfo);
+  /* Initialize the marker writer ... bit of a crock to do it here. */
+  jinit_marker_writer(cinfo);
+  /* Write them tables! */
+  (*cinfo->marker->write_tables_only) (cinfo);
+  /* And clean up. */
+  (*cinfo->dest->term_destination) (cinfo);
+  /*
+   * In library releases up through v6a, we called jpeg_abort() here to free
+   * any working memory allocated by the destination manager and marker
+   * writer.  Some applications had a problem with that: they allocated space
+   * of their own from the library memory manager, and didn't want it to go
+   * away during write_tables.  So now we do nothing.  This will cause a
+   * memory leak if an app calls write_tables repeatedly without doing a full
+   * compression cycle or otherwise resetting the JPEG object.  However, that
+   * seems less bad than unexpectedly freeing memory in the normal case.
+   * An app that prefers the old behavior can call jpeg_abort for itself after
+   * each call to jpeg_write_tables().
+   */
+}

=== added file 'src/libjpeg-turbo/jcapistd.c'
--- src/libjpeg-turbo/jcapistd.c	1970-01-01 00:00:00 +0000
+++ src/libjpeg-turbo/jcapistd.c	2012-06-27 16:20:24 +0000
@@ -0,0 +1,161 @@
+/*
+ * jcapistd.c
+ *
+ * Copyright (C) 1994-1996, Thomas G. Lane.
+ * This file is part of the Independent JPEG Group's software.
+ * For conditions of distribution and use, see the accompanying README file.
+ *
+ * This file contains application interface code for the compression half
+ * of the JPEG library.  These are the "standard" API routines that are
+ * used in the normal full-compression case.  They are not used by a
+ * transcoding-only application.  Note that if an application links in
+ * jpeg_start_compress, it will end up linking in the entire compressor.
+ * We thus must separate this file from jcapimin.c to avoid linking the
+ * whole compression library into a transcoder.
+ */
+
+#define JPEG_INTERNALS
+#include "jinclude.h"
+#include "jpeglib.h"
+
+
+/*
+ * Compression initialization.
+ * Before calling this, all parameters and a data destination must be set up.
+ *
+ * We require a write_all_tables parameter as a failsafe check when writing
+ * multiple datastreams from the same compression object.  Since prior runs
+ * will have left all the tables marked sent_table=TRUE, a subsequent run
+ * would emit an abbreviated stream (no tables) by default.  This may be what
+ * is wanted, but for safety's sake it should not be the default behavior:
+ * programmers should have to make a deliberate choice to emit abbreviated
+ * images.  Therefore the documentation and examples should encourage people
+ * to pass write_all_tables=TRUE; then it will take active thought to do the
+ * wrong thing.
+ */
+
+GLOBAL(void)
+jpeg_start_compress (j_compress_ptr cinfo, boolean write_all_tables)
+{
+  if (cinfo->global_state != CSTATE_START)
+    ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state);
+
+  if (write_all_tables)
+    jpeg_suppress_tables(cinfo, FALSE);	/* mark all tables to be written */
+
+  /* (Re)initialize error mgr and destination modules */
+  (*cinfo->err->reset_error_mgr) ((j_common_ptr) cinfo);
+  (*cinfo->dest->init_destination) (cinfo);
+  /* Perform master selection of active modules */
+  jinit_compress_master(cinfo);
+  /* Set up for the first pass */
+  (*cinfo->master->prepare_for_pass) (cinfo);
+  /* Ready for application to drive first pass through jpeg_write_scanlines
+   * or jpeg_write_raw_data.
+   */
+  cinfo->next_scanline = 0;
+  cinfo->global_state = (cinfo->raw_data_in ? CSTATE_RAW_OK : CSTATE_SCANNING);
+}
+
+
+/*
+ * Write some scanlines of data to the JPEG compressor.
+ *
+ * The return value will be the number of lines actually written.
+ * This should be less than the supplied num_lines only in case that
+ * the data destination module has requested suspension of the compressor,
+ * or if more than image_height scanlines are passed in.
+ *
+ * Note: we warn about excess calls to jpeg_write_scanlines() since
+ * this likely signals an application programmer error.  However,
+ * excess scanlines passed in the last valid call are *silently* ignored,
+ * so that the application need not adjust num_lines for end-of-image
+ * when using a multiple-scanline buffer.
+ */
+
+GLOBAL(JDIMENSION)
+jpeg_write_scanlines (j_compress_ptr cinfo, JSAMPARRAY scanlines,
+		      JDIMENSION num_lines)
+{
+  JDIMENSION row_ctr, rows_left;
+
+  if (cinfo->global_state != CSTATE_SCANNING)
+    ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state);
+  if (cinfo->next_scanline >= cinfo->image_height)
+    WARNMS(cinfo, JWRN_TOO_MUCH_DATA);
+
+  /* Call progress monitor hook if present */
+  if (cinfo->progress != NULL) {
+    cinfo->progress->pass_counter = (long) cinfo->next_scanline;
+    cinfo->progress->pass_limit = (long) cinfo->image_height;
+    (*cinfo->progress->progress_monitor) ((j_common_ptr) cinfo);
+  }
+
+  /* Give master control module another chance if this is first call to
+   * jpeg_write_scanlines.  This lets output of the frame/scan headers be
+   * delayed so that application can write COM, etc, markers between
+   * jpeg_start_compress and jpeg_write_scanlines.
+   */
+  if (cinfo->master->call_pass_startup)
+    (*cinfo->master->pass_startup) (cinfo);
+
+  /* Ignore any extra scanlines at bottom of image. */
+  rows_left = cinfo->image_height - cinfo->next_scanline;
+  if (num_lines > rows_left)
+    num_lines = rows_left;
+
+  row_ctr = 0;
+  (*cinfo->main->process_data) (cinfo, scanlines, &row_ctr, num_lines);
+  cinfo->next_scanline += row_ctr;
+  return row_ctr;
+}
+
+
+/*
+ * Alternate entry point to write raw data.
+ * Processes exactly one iMCU row per call, unless suspended.
+ */
+
+GLOBAL(JDIMENSION)
+jpeg_write_raw_data (j_compress_ptr cinfo, JSAMPIMAGE data,
+		     JDIMENSION num_lines)
+{
+  JDIMENSION lines_per_iMCU_row;
+
+  if (cinfo->global_state != CSTATE_RAW_OK)
+    ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state);
+  if (cinfo->next_scanline >= cinfo->image_height) {
+    WARNMS(cinfo, JWRN_TOO_MUCH_DATA);
+    return 0;
+  }
+
+  /* Call progress monitor hook if present */
+  if (cinfo->progress != NULL) {
+    cinfo->progress->pass_counter = (long) cinfo->next_scanline;
+    cinfo->progress->pass_limit = (long) cinfo->image_height;
+    (*cinfo->progress->progress_monitor) ((j_common_ptr) cinfo);
+  }
+
+  /* Give master control module another chance if this is first call to
+   * jpeg_write_raw_data.  This lets output of the frame/scan headers be
+   * delayed so that application can write COM, etc, markers between
+   * jpeg_start_compress and jpeg_write_raw_data.
+   */
+  if (cinfo->master->call_pass_startup)
+    (*cinfo->master->pass_startup) (cinfo);
+
+  /* Verify that at least one iMCU row has been passed. */
+  lines_per_iMCU_row = cinfo->max_v_samp_factor * DCTSIZE;
+  if (num_lines < lines_per_iMCU_row)
+    ERREXIT(cinfo, JERR_BUFFER_SIZE);
+
+  /* Directly compress the row. */
+  if (! (*cinfo->coef->compress_data) (cinfo, data)) {
+    /* If compressor did not consume the whole row, suspend processing. */
+    return 0;
+  }
+
+  /* OK, we processed one iMCU row. */
+  cinfo->next_scanline += lines_per_iMCU_row;
+  return lines_per_iMCU_row;
+}

=== added file 'src/libjpeg-turbo/jcarith.c'
--- src/libjpeg-turbo/jcarith.c	1970-01-01 00:00:00 +0000
+++ src/libjpeg-turbo/jcarith.c	2012-06-27 16:20:24 +0000
@@ -0,0 +1,925 @@
+/*
+ * jcarith.c
+ *
+ * Developed 1997-2009 by Guido Vollbeding.
+ * This file is part of the Independent JPEG Group's software.
+ * For conditions of distribution and use, see the accompanying README file.
+ *
+ * This file contains portable arithmetic entropy encoding routines for JPEG
+ * (implementing the ISO/IEC IS 10918-1 and CCITT Recommendation ITU-T T.81).
+ *
+ * Both sequential and progressive modes are supported in this single module.
+ *
+ * Suspension is not currently supported in this module.
+ */
+
+#define JPEG_INTERNALS
+#include "jinclude.h"
+#include "jpeglib.h"
+
+
+/* Expanded entropy encoder object for arithmetic encoding. */
+
+typedef struct {
+  struct jpeg_entropy_encoder pub; /* public fields */
+
+  INT32 c; /* C register, base of coding interval, layout as in sec. D.1.3 */
+  INT32 a;               /* A register, normalized size of coding interval */
+  INT32 sc;        /* counter for stacked 0xFF values which might overflow */
+  INT32 zc;          /* counter for pending 0x00 output values which might *
+                          * be discarded at the end ("Pacman" termination) */
+  int ct;  /* bit shift counter, determines when next byte will be written */
+  int buffer;                /* buffer for most recent output byte != 0xFF */
+
+  int last_dc_val[MAX_COMPS_IN_SCAN]; /* last DC coef for each component */
+  int dc_context[MAX_COMPS_IN_SCAN]; /* context index for DC conditioning */
+
+  unsigned int restarts_to_go;	/* MCUs left in this restart interval */
+  int next_restart_num;		/* next restart number to write (0-7) */
+
+  /* Pointers to statistics areas (these workspaces have image lifespan) */
+  unsigned char * dc_stats[NUM_ARITH_TBLS];
+  unsigned char * ac_stats[NUM_ARITH_TBLS];
+
+  /* Statistics bin for coding with fixed probability 0.5 */
+  unsigned char fixed_bin[4];
+} arith_entropy_encoder;
+
+typedef arith_entropy_encoder * arith_entropy_ptr;
+
+/* The following two definitions specify the allocation chunk size
+ * for the statistics area.
+ * According to sections F.1.4.4.1.3 and F.1.4.4.2, we need at least
+ * 49 statistics bins for DC, and 245 statistics bins for AC coding.
+ *
+ * We use a compact representation with 1 byte per statistics bin,
+ * thus the numbers directly represent byte sizes.
+ * This 1 byte per statistics bin contains the meaning of the MPS
+ * (more probable symbol) in the highest bit (mask 0x80), and the
+ * index into the probability estimation state machine table
+ * in the lower bits (mask 0x7F).
+ */
+
+#define DC_STAT_BINS 64
+#define AC_STAT_BINS 256
+
+/* NOTE: Uncomment the following #define if you want to use the
+ * given formula for calculating the AC conditioning parameter Kx
+ * for spectral selection progressive coding in section G.1.3.2
+ * of the spec (Kx = Kmin + SRL (8 + Se - Kmin) 4).
+ * Although the spec and P&M authors claim that this "has proven
+ * to give good results for 8 bit precision samples", I'm not
+ * convinced yet that this is really beneficial.
+ * Early tests gave only very marginal compression enhancements
+ * (a few - around 5 or so - bytes even for very large files),
+ * which would turn out rather negative if we'd suppress the
+ * DAC (Define Arithmetic Conditioning) marker segments for
+ * the default parameters in the future.
+ * Note that currently the marker writing module emits 12-byte
+ * DAC segments for a full-component scan in a color image.
+ * This is not worth worrying about IMHO. However, since the
+ * spec defines the default values to be used if the tables
+ * are omitted (unlike Huffman tables, which are required
+ * anyway), one might optimize this behaviour in the future,
+ * and then it would be disadvantageous to use custom tables if
+ * they don't provide sufficient gain to exceed the DAC size.
+ *
+ * On the other hand, I'd consider it as a reasonable result
+ * that the conditioning has no significant influence on the
+ * compression performance. This means that the basic
+ * statistical model is already rather stable.
+ *
+ * Thus, at the moment, we use the default conditioning values
+ * anyway, and do not use the custom formula.
+ *
+#define CALCULATE_SPECTRAL_CONDITIONING
+ */
+
+/* IRIGHT_SHIFT is like RIGHT_SHIFT, but works on int rather than INT32.
+ * We assume that int right shift is unsigned if INT32 right shift is,
+ * which should be safe.
+ */
+
+#ifdef RIGHT_SHIFT_IS_UNSIGNED
+#define ISHIFT_TEMPS	int ishift_temp;
+#define IRIGHT_SHIFT(x,shft)  \
+	((ishift_temp = (x)) < 0 ? \
+	 (ishift_temp >> (shft)) | ((~0) << (16-(shft))) : \
+	 (ishift_temp >> (shft)))
+#else
+#define ISHIFT_TEMPS
+#define IRIGHT_SHIFT(x,shft)	((x) >> (shft))
+#endif
+
+
+LOCAL(void)
+emit_byte (int val, j_compress_ptr cinfo)
+/* Write next output byte; we do not support suspension in this module. */
+{
+  struct jpeg_destination_mgr * dest = cinfo->dest;
+
+  *dest->next_output_byte++ = (JOCTET) val;
+  if (--dest->free_in_buffer == 0)
+    if (! (*dest->empty_output_buffer) (cinfo))
+      ERREXIT(cinfo, JERR_CANT_SUSPEND);
+}
+
+
+/*
+ * Finish up at the end of an arithmetic-compressed scan.
+ */
+
+METHODDEF(void)
+finish_pass (j_compress_ptr cinfo)
+{
+  arith_entropy_ptr e = (arith_entropy_ptr) cinfo->entropy;
+  INT32 temp;
+
+  /* Section D.1.8: Termination of encoding */
+
+  /* Find the e->c in the coding interval with the largest
+   * number of trailing zero bits */
+  if ((temp = (e->a - 1 + e->c) & 0xFFFF0000L) < e->c)
+    e->c = temp + 0x8000L;
+  else
+    e->c = temp;
+  /* Send remaining bytes to output */
+  e->c <<= e->ct;
+  if (e->c & 0xF8000000L) {
+    /* One final overflow has to be handled */
+    if (e->buffer >= 0) {
+      if (e->zc)
+	do emit_byte(0x00, cinfo);
+	while (--e->zc);
+      emit_byte(e->buffer + 1, cinfo);
+      if (e->buffer + 1 == 0xFF)
+	emit_byte(0x00, cinfo);
+    }
+    e->zc += e->sc;  /* carry-over converts stacked 0xFF bytes to 0x00 */
+    e->sc = 0;
+  } else {
+    if (e->buffer == 0)
+      ++e->zc;
+    else if (e->buffer >= 0) {
+      if (e->zc)
+	do emit_byte(0x00, cinfo);
+	while (--e->zc);
+      emit_byte(e->buffer, cinfo);
+    }
+    if (e->sc) {
+      if (e->zc)
+	do emit_byte(0x00, cinfo);
+	while (--e->zc);
+      do {
+	emit_byte(0xFF, cinfo);
+	emit_byte(0x00, cinfo);
+      } while (--e->sc);
+    }
+  }
+  /* Output final bytes only if they are not 0x00 */
+  if (e->c & 0x7FFF800L) {
+    if (e->zc)  /* output final pending zero bytes */
+      do emit_byte(0x00, cinfo);
+      while (--e->zc);
+    emit_byte((e->c >> 19) & 0xFF, cinfo);
+    if (((e->c >> 19) & 0xFF) == 0xFF)
+      emit_byte(0x00, cinfo);
+    if (e->c & 0x7F800L) {
+      emit_byte((e->c >> 11) & 0xFF, cinfo);
+      if (((e->c >> 11) & 0xFF) == 0xFF)
+	emit_byte(0x00, cinfo);
+    }
+  }
+}
+
+
+/*
+ * The core arithmetic encoding routine (common in JPEG and JBIG).
+ * This needs to go as fast as possible.
+ * Machine-dependent optimization facilities
+ * are not utilized in this portable implementation.
+ * However, this code should be fairly efficient and
+ * may be a good base for further optimizations anyway.
+ *
+ * Parameter 'val' to be encoded may be 0 or 1 (binary decision).
+ *
+ * Note: I've added full "Pacman" termination support to the
+ * byte output routines, which is equivalent to the optional
+ * Discard_final_zeros procedure (Figure D.15) in the spec.
+ * Thus, we always produce the shortest possible output
+ * stream compliant to the spec (no trailing zero bytes,
+ * except for FF stuffing).
+ *
+ * I've also introduced a new scheme for accessing
+ * the probability estimation state machine table,
+ * derived from Markus Kuhn's JBIG implementation.
+ */
+
+LOCAL(void)
+arith_encode (j_compress_ptr cinfo, unsigned char *st, int val) 
+{
+  register arith_entropy_ptr e = (arith_entropy_ptr) cinfo->entropy;
+  register unsigned char nl, nm;
+  register INT32 qe, temp;
+  register int sv;
+
+  /* Fetch values from our compact representation of Table D.2:
+   * Qe values and probability estimation state machine
+   */
+  sv = *st;
+  qe = jpeg_aritab[sv & 0x7F];	/* => Qe_Value */
+  nl = qe & 0xFF; qe >>= 8;	/* Next_Index_LPS + Switch_MPS */
+  nm = qe & 0xFF; qe >>= 8;	/* Next_Index_MPS */
+
+  /* Encode & estimation procedures per sections D.1.4 & D.1.5 */
+  e->a -= qe;
+  if (val != (sv >> 7)) {
+    /* Encode the less probable symbol */
+    if (e->a >= qe) {
+      /* If the interval size (qe) for the less probable symbol (LPS)
+       * is larger than the interval size for the MPS, then exchange
+       * the two symbols for coding efficiency, otherwise code the LPS
+       * as usual: */
+      e->c += e->a;
+      e->a = qe;
+    }
+    *st = (sv & 0x80) ^ nl;	/* Estimate_after_LPS */
+  } else {
+    /* Encode the more probable symbol */
+    if (e->a >= 0x8000L)
+      return;  /* A >= 0x8000 -> ready, no renormalization required */
+    if (e->a < qe) {
+      /* If the interval size (qe) for the less probable symbol (LPS)
+       * is larger than the interval size for the MPS, then exchange
+       * the two symbols for coding efficiency: */
+      e->c += e->a;
+      e->a = qe;
+    }
+    *st = (sv & 0x80) ^ nm;	/* Estimate_after_MPS */
+  }
+
+  /* Renormalization & data output per section D.1.6 */
+  do {
+    e->a <<= 1;
+    e->c <<= 1;
+    if (--e->ct == 0) {
+      /* Another byte is ready for output */
+      temp = e->c >> 19;
+      if (temp > 0xFF) {
+	/* Handle overflow over all stacked 0xFF bytes */
+	if (e->buffer >= 0) {
+	  if (e->zc)
+	    do emit_byte(0x00, cinfo);
+	    while (--e->zc);
+	  emit_byte(e->buffer + 1, cinfo);
+	  if (e->buffer + 1 == 0xFF)
+	    emit_byte(0x00, cinfo);
+	}
+	e->zc += e->sc;  /* carry-over converts stacked 0xFF bytes to 0x00 */
+	e->sc = 0;
+	/* Note: The 3 spacer bits in the C register guarantee
+	 * that the new buffer byte can't be 0xFF here
+	 * (see page 160 in the P&M JPEG book). */
+	e->buffer = temp & 0xFF;  /* new output byte, might overflow later */
+      } else if (temp == 0xFF) {
+	++e->sc;  /* stack 0xFF byte (which might overflow later) */
+      } else {
+	/* Output all stacked 0xFF bytes, they will not overflow any more */
+	if (e->buffer == 0)
+	  ++e->zc;
+	else if (e->buffer >= 0) {
+	  if (e->zc)
+	    do emit_byte(0x00, cinfo);
+	    while (--e->zc);
+	  emit_byte(e->buffer, cinfo);
+	}
+	if (e->sc) {
+	  if (e->zc)
+	    do emit_byte(0x00, cinfo);
+	    while (--e->zc);
+	  do {
+	    emit_byte(0xFF, cinfo);
+	    emit_byte(0x00, cinfo);
+	  } while (--e->sc);
+	}
+	e->buffer = temp & 0xFF;  /* new output byte (can still overflow) */
+      }
+      e->c &= 0x7FFFFL;
+      e->ct += 8;
+    }
+  } while (e->a < 0x8000L);
+}
+
+
+/*
+ * Emit a restart marker & resynchronize predictions.
+ */
+
+LOCAL(void)
+emit_restart (j_compress_ptr cinfo, int restart_num)
+{
+  arith_entropy_ptr entropy = (arith_entropy_ptr) cinfo->entropy;
+  int ci;
+  jpeg_component_info * compptr;
+
+  finish_pass(cinfo);
+
+  emit_byte(0xFF, cinfo);
+  emit_byte(JPEG_RST0 + restart_num, cinfo);
+
+  /* Re-initialize statistics areas */
+  for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
+    compptr = cinfo->cur_comp_info[ci];
+    /* DC needs no table for refinement scan */
+    if (cinfo->progressive_mode == 0 || (cinfo->Ss == 0 && cinfo->Ah == 0)) {
+      MEMZERO(entropy->dc_stats[compptr->dc_tbl_no], DC_STAT_BINS);
+      /* Reset DC predictions to 0 */
+      entropy->last_dc_val[ci] = 0;
+      entropy->dc_context[ci] = 0;
+    }
+    /* AC needs no table when not present */
+    if (cinfo->progressive_mode == 0 || cinfo->Se) {
+      MEMZERO(entropy->ac_stats[compptr->ac_tbl_no], AC_STAT_BINS);
+    }
+  }
+
+  /* Reset arithmetic encoding variables */
+  entropy->c = 0;
+  entropy->a = 0x10000L;
+  entropy->sc = 0;
+  entropy->zc = 0;
+  entropy->ct = 11;
+  entropy->buffer = -1;  /* empty */
+}
+
+
+/*
+ * MCU encoding for DC initial scan (either spectral selection,
+ * or first pass of successive approximation).
+ */
+
+METHODDEF(boolean)
+encode_mcu_DC_first (j_compress_ptr cinfo, JBLOCKROW *MCU_data)
+{
+  arith_entropy_ptr entropy = (arith_entropy_ptr) cinfo->entropy;
+  JBLOCKROW block;
+  unsigned char *st;
+  int blkn, ci, tbl;
+  int v, v2, m;
+  ISHIFT_TEMPS
+
+  /* Emit restart marker if needed */
+  if (cinfo->restart_interval) {
+    if (entropy->restarts_to_go == 0) {
+      emit_restart(cinfo, entropy->next_restart_num);
+      entropy->restarts_to_go = cinfo->restart_interval;
+      entropy->next_restart_num++;
+      entropy->next_restart_num &= 7;
+    }
+    entropy->restarts_to_go--;
+  }
+
+  /* Encode the MCU data blocks */
+  for (blkn = 0; blkn < cinfo->blocks_in_MCU; blkn++) {
+    block = MCU_data[blkn];
+    ci = cinfo->MCU_membership[blkn];
+    tbl = cinfo->cur_comp_info[ci]->dc_tbl_no;
+
+    /* Compute the DC value after the required point transform by Al.
+     * This is simply an arithmetic right shift.
+     */
+    m = IRIGHT_SHIFT((int) ((*block)[0]), cinfo->Al);
+
+    /* Sections F.1.4.1 & F.1.4.4.1: Encoding of DC coefficients */
+
+    /* Table F.4: Point to statistics bin S0 for DC coefficient coding */
+    st = entropy->dc_stats[tbl] + entropy->dc_context[ci];
+
+    /* Figure F.4: Encode_DC_DIFF */
+    if ((v = m - entropy->last_dc_val[ci]) == 0) {
+      arith_encode(cinfo, st, 0);
+      entropy->dc_context[ci] = 0;	/* zero diff category */
+    } else {
+      entropy->last_dc_val[ci] = m;
+      arith_encode(cinfo, st, 1);
+      /* Figure F.6: Encoding nonzero value v */
+      /* Figure F.7: Encoding the sign of v */
+      if (v > 0) {
+	arith_encode(cinfo, st + 1, 0);	/* Table F.4: SS = S0 + 1 */
+	st += 2;			/* Table F.4: SP = S0 + 2 */
+	entropy->dc_context[ci] = 4;	/* small positive diff category */
+      } else {
+	v = -v;
+	arith_encode(cinfo, st + 1, 1);	/* Table F.4: SS = S0 + 1 */
+	st += 3;			/* Table F.4: SN = S0 + 3 */
+	entropy->dc_context[ci] = 8;	/* small negative diff category */
+      }
+      /* Figure F.8: Encoding the magnitude category of v */
+      m = 0;
+      if (v -= 1) {
+	arith_encode(cinfo, st, 1);
+	m = 1;
+	v2 = v;
+	st = entropy->dc_stats[tbl] + 20; /* Table F.4: X1 = 20 */
+	while (v2 >>= 1) {
+	  arith_encode(cinfo, st, 1);
+	  m <<= 1;
+	  st += 1;
+	}
+      }
+      arith_encode(cinfo, st, 0);
+      /* Section F.1.4.4.1.2: Establish dc_context conditioning category */
+      if (m < (int) ((1L << cinfo->arith_dc_L[tbl]) >> 1))
+	entropy->dc_context[ci] = 0;	/* zero diff category */
+      else if (m > (int) ((1L << cinfo->arith_dc_U[tbl]) >> 1))
+	entropy->dc_context[ci] += 8;	/* large diff category */
+      /* Figure F.9: Encoding the magnitude bit pattern of v */
+      st += 14;
+      while (m >>= 1)
+	arith_encode(cinfo, st, (m & v) ? 1 : 0);
+    }
+  }
+
+  return TRUE;
+}
+
+
+/*
+ * MCU encoding for AC initial scan (either spectral selection,
+ * or first pass of successive approximation).
+ */
+
+METHODDEF(boolean)
+encode_mcu_AC_first (j_compress_ptr cinfo, JBLOCKROW *MCU_data)
+{
+  arith_entropy_ptr entropy = (arith_entropy_ptr) cinfo->entropy;
+  JBLOCKROW block;
+  unsigned char *st;
+  int tbl, k, ke;
+  int v, v2, m;
+
+  /* Emit restart marker if needed */
+  if (cinfo->restart_interval) {
+    if (entropy->restarts_to_go == 0) {
+      emit_restart(cinfo, entropy->next_restart_num);
+      entropy->restarts_to_go = cinfo->restart_interval;
+      entropy->next_restart_num++;
+      entropy->next_restart_num &= 7;
+    }
+    entropy->restarts_to_go--;
+  }
+
+  /* Encode the MCU data block */
+  block = MCU_data[0];
+  tbl = cinfo->cur_comp_info[0]->ac_tbl_no;
+
+  /* Sections F.1.4.2 & F.1.4.4.2: Encoding of AC coefficients */
+
+  /* Establish EOB (end-of-block) index */
+  for (ke = cinfo->Se; ke > 0; ke--)
+    /* We must apply the point transform by Al.  For AC coefficients this
+     * is an integer division with rounding towards 0.  To do this portably
+     * in C, we shift after obtaining the absolute value.
+     */
+    if ((v = (*block)[jpeg_natural_order[ke]]) >= 0) {
+      if (v >>= cinfo->Al) break;
+    } else {
+      v = -v;
+      if (v >>= cinfo->Al) break;
+    }
+
+  /* Figure F.5: Encode_AC_Coefficients */
+  for (k = cinfo->Ss; k <= ke; k++) {
+    st = entropy->ac_stats[tbl] + 3 * (k - 1);
+    arith_encode(cinfo, st, 0);		/* EOB decision */
+    for (;;) {
+      if ((v = (*block)[jpeg_natural_order[k]]) >= 0) {
+	if (v >>= cinfo->Al) {
+	  arith_encode(cinfo, st + 1, 1);
+	  arith_encode(cinfo, entropy->fixed_bin, 0);
+	  break;
+	}
+      } else {
+	v = -v;
+	if (v >>= cinfo->Al) {
+	  arith_encode(cinfo, st + 1, 1);
+	  arith_encode(cinfo, entropy->fixed_bin, 1);
+	  break;
+	}
+      }
+      arith_encode(cinfo, st + 1, 0); st += 3; k++;
+    }
+    st += 2;
+    /* Figure F.8: Encoding the magnitude category of v */
+    m = 0;
+    if (v -= 1) {
+      arith_encode(cinfo, st, 1);
+      m = 1;
+      v2 = v;
+      if (v2 >>= 1) {
+	arith_encode(cinfo, st, 1);
+	m <<= 1;
+	st = entropy->ac_stats[tbl] +
+	     (k <= cinfo->arith_ac_K[tbl] ? 189 : 217);
+	while (v2 >>= 1) {
+	  arith_encode(cinfo, st, 1);
+	  m <<= 1;
+	  st += 1;
+	}
+      }
+    }
+    arith_encode(cinfo, st, 0);
+    /* Figure F.9: Encoding the magnitude bit pattern of v */
+    st += 14;
+    while (m >>= 1)
+      arith_encode(cinfo, st, (m & v) ? 1 : 0);
+  }
+  /* Encode EOB decision only if k <= cinfo->Se */
+  if (k <= cinfo->Se) {
+    st = entropy->ac_stats[tbl] + 3 * (k - 1);
+    arith_encode(cinfo, st, 1);
+  }
+
+  return TRUE;
+}
+
+
+/*
+ * MCU encoding for DC successive approximation refinement scan.
+ */
+
+METHODDEF(boolean)
+encode_mcu_DC_refine (j_compress_ptr cinfo, JBLOCKROW *MCU_data)
+{
+  arith_entropy_ptr entropy = (arith_entropy_ptr) cinfo->entropy;
+  unsigned char *st;
+  int Al, blkn;
+
+  /* Emit restart marker if needed */
+  if (cinfo->restart_interval) {
+    if (entropy->restarts_to_go == 0) {
+      emit_restart(cinfo, entropy->next_restart_num);
+      entropy->restarts_to_go = cinfo->restart_interval;
+      entropy->next_restart_num++;
+      entropy->next_restart_num &= 7;
+    }
+    entropy->restarts_to_go--;
+  }
+
+  st = entropy->fixed_bin;	/* use fixed probability estimation */
+  Al = cinfo->Al;
+
+  /* Encode the MCU data blocks */
+  for (blkn = 0; blkn < cinfo->blocks_in_MCU; blkn++) {
+    /* We simply emit the Al'th bit of the DC coefficient value. */
+    arith_encode(cinfo, st, (MCU_data[blkn][0][0] >> Al) & 1);
+  }
+
+  return TRUE;
+}
+
+
+/*
+ * MCU encoding for AC successive approximation refinement scan.
+ */
+
+METHODDEF(boolean)
+encode_mcu_AC_refine (j_compress_ptr cinfo, JBLOCKROW *MCU_data)
+{
+  arith_entropy_ptr entropy = (arith_entropy_ptr) cinfo->entropy;
+  JBLOCKROW block;
+  unsigned char *st;
+  int tbl, k, ke, kex;
+  int v;
+
+  /* Emit restart marker if needed */
+  if (cinfo->restart_interval) {
+    if (entropy->restarts_to_go == 0) {
+      emit_restart(cinfo, entropy->next_restart_num);
+      entropy->restarts_to_go = cinfo->restart_interval;
+      entropy->next_restart_num++;
+      entropy->next_restart_num &= 7;
+    }
+    entropy->restarts_to_go--;
+  }
+
+  /* Encode the MCU data block */
+  block = MCU_data[0];
+  tbl = cinfo->cur_comp_info[0]->ac_tbl_no;
+
+  /* Section G.1.3.3: Encoding of AC coefficients */
+
+  /* Establish EOB (end-of-block) index */
+  for (ke = cinfo->Se; ke > 0; ke--)
+    /* We must apply the point transform by Al.  For AC coefficients this
+     * is an integer division with rounding towards 0.  To do this portably
+     * in C, we shift after obtaining the absolute value.
+     */
+    if ((v = (*block)[jpeg_natural_order[ke]]) >= 0) {
+      if (v >>= cinfo->Al) break;
+    } else {
+      v = -v;
+      if (v >>= cinfo->Al) break;
+    }
+
+  /* Establish EOBx (previous stage end-of-block) index */
+  for (kex = ke; kex > 0; kex--)
+    if ((v = (*block)[jpeg_natural_order[kex]]) >= 0) {
+      if (v >>= cinfo->Ah) break;
+    } else {
+      v = -v;
+      if (v >>= cinfo->Ah) break;
+    }
+
+  /* Figure G.10: Encode_AC_Coefficients_SA */
+  for (k = cinfo->Ss; k <= ke; k++) {
+    st = entropy->ac_stats[tbl] + 3 * (k - 1);
+    if (k > kex)
+      arith_encode(cinfo, st, 0);	/* EOB decision */
+    for (;;) {
+      if ((v = (*block)[jpeg_natural_order[k]]) >= 0) {
+	if (v >>= cinfo->Al) {
+	  if (v >> 1)			/* previously nonzero coef */
+	    arith_encode(cinfo, st + 2, (v & 1));
+	  else {			/* newly nonzero coef */
+	    arith_encode(cinfo, st + 1, 1);
+	    arith_encode(cinfo, entropy->fixed_bin, 0);
+	  }
+	  break;
+	}
+      } else {
+	v = -v;
+	if (v >>= cinfo->Al) {
+	  if (v >> 1)			/* previously nonzero coef */
+	    arith_encode(cinfo, st + 2, (v & 1));
+	  else {			/* newly nonzero coef */
+	    arith_encode(cinfo, st + 1, 1);
+	    arith_encode(cinfo, entropy->fixed_bin, 1);
+	  }
+	  break;
+	}
+      }
+      arith_encode(cinfo, st + 1, 0); st += 3; k++;
+    }
+  }
+  /* Encode EOB decision only if k <= cinfo->Se */
+  if (k <= cinfo->Se) {
+    st = entropy->ac_stats[tbl] + 3 * (k - 1);
+    arith_encode(cinfo, st, 1);
+  }
+
+  return TRUE;
+}
+
+
+/*
+ * Encode and output one MCU's worth of arithmetic-compressed coefficients.
+ */
+
+METHODDEF(boolean)
+encode_mcu (j_compress_ptr cinfo, JBLOCKROW *MCU_data)
+{
+  arith_entropy_ptr entropy = (arith_entropy_ptr) cinfo->entropy;
+  jpeg_component_info * compptr;
+  JBLOCKROW block;
+  unsigned char *st;
+  int blkn, ci, tbl, k, ke;
+  int v, v2, m;
+
+  /* Emit restart marker if needed */
+  if (cinfo->restart_interval) {
+    if (entropy->restarts_to_go == 0) {
+      emit_restart(cinfo, entropy->next_restart_num);
+      entropy->restarts_to_go = cinfo->restart_interval;
+      entropy->next_restart_num++;
+      entropy->next_restart_num &= 7;
+    }
+    entropy->restarts_to_go--;
+  }
+
+  /* Encode the MCU data blocks */
+  for (blkn = 0; blkn < cinfo->blocks_in_MCU; blkn++) {
+    block = MCU_data[blkn];
+    ci = cinfo->MCU_membership[blkn];
+    compptr = cinfo->cur_comp_info[ci];
+
+    /* Sections F.1.4.1 & F.1.4.4.1: Encoding of DC coefficients */
+
+    tbl = compptr->dc_tbl_no;
+
+    /* Table F.4: Point to statistics bin S0 for DC coefficient coding */
+    st = entropy->dc_stats[tbl] + entropy->dc_context[ci];
+
+    /* Figure F.4: Encode_DC_DIFF */
+    if ((v = (*block)[0] - entropy->last_dc_val[ci]) == 0) {
+      arith_encode(cinfo, st, 0);
+      entropy->dc_context[ci] = 0;	/* zero diff category */
+    } else {
+      entropy->last_dc_val[ci] = (*block)[0];
+      arith_encode(cinfo, st, 1);
+      /* Figure F.6: Encoding nonzero value v */
+      /* Figure F.7: Encoding the sign of v */
+      if (v > 0) {
+	arith_encode(cinfo, st + 1, 0);	/* Table F.4: SS = S0 + 1 */
+	st += 2;			/* Table F.4: SP = S0 + 2 */
+	entropy->dc_context[ci] = 4;	/* small positive diff category */
+      } else {
+	v = -v;
+	arith_encode(cinfo, st + 1, 1);	/* Table F.4: SS = S0 + 1 */
+	st += 3;			/* Table F.4: SN = S0 + 3 */
+	entropy->dc_context[ci] = 8;	/* small negative diff category */
+      }
+      /* Figure F.8: Encoding the magnitude category of v */
+      m = 0;
+      if (v -= 1) {
+	arith_encode(cinfo, st, 1);
+	m = 1;
+	v2 = v;
+	st = entropy->dc_stats[tbl] + 20; /* Table F.4: X1 = 20 */
+	while (v2 >>= 1) {
+	  arith_encode(cinfo, st, 1);
+	  m <<= 1;
+	  st += 1;
+	}
+      }
+      arith_encode(cinfo, st, 0);
+      /* Section F.1.4.4.1.2: Establish dc_context conditioning category */
+      if (m < (int) ((1L << cinfo->arith_dc_L[tbl]) >> 1))
+	entropy->dc_context[ci] = 0;	/* zero diff category */
+      else if (m > (int) ((1L << cinfo->arith_dc_U[tbl]) >> 1))
+	entropy->dc_context[ci] += 8;	/* large diff category */
+      /* Figure F.9: Encoding the magnitude bit pattern of v */
+      st += 14;
+      while (m >>= 1)
+	arith_encode(cinfo, st, (m & v) ? 1 : 0);
+    }
+
+    /* Sections F.1.4.2 & F.1.4.4.2: Encoding of AC coefficients */
+
+    tbl = compptr->ac_tbl_no;
+
+    /* Establish EOB (end-of-block) index */
+    for (ke = DCTSIZE2 - 1; ke > 0; ke--)
+      if ((*block)[jpeg_natural_order[ke]]) break;
+
+    /* Figure F.5: Encode_AC_Coefficients */
+    for (k = 1; k <= ke; k++) {
+      st = entropy->ac_stats[tbl] + 3 * (k - 1);
+      arith_encode(cinfo, st, 0);	/* EOB decision */
+      while ((v = (*block)[jpeg_natural_order[k]]) == 0) {
+	arith_encode(cinfo, st + 1, 0); st += 3; k++;
+      }
+      arith_encode(cinfo, st + 1, 1);
+      /* Figure F.6: Encoding nonzero value v */
+      /* Figure F.7: Encoding the sign of v */
+      if (v > 0) {
+	arith_encode(cinfo, entropy->fixed_bin, 0);
+      } else {
+	v = -v;
+	arith_encode(cinfo, entropy->fixed_bin, 1);
+      }
+      st += 2;
+      /* Figure F.8: Encoding the magnitude category of v */
+      m = 0;
+      if (v -= 1) {
+	arith_encode(cinfo, st, 1);
+	m = 1;
+	v2 = v;
+	if (v2 >>= 1) {
+	  arith_encode(cinfo, st, 1);
+	  m <<= 1;
+	  st = entropy->ac_stats[tbl] +
+	       (k <= cinfo->arith_ac_K[tbl] ? 189 : 217);
+	  while (v2 >>= 1) {
+	    arith_encode(cinfo, st, 1);
+	    m <<= 1;
+	    st += 1;
+	  }
+	}
+      }
+      arith_encode(cinfo, st, 0);
+      /* Figure F.9: Encoding the magnitude bit pattern of v */
+      st += 14;
+      while (m >>= 1)
+	arith_encode(cinfo, st, (m & v) ? 1 : 0);
+    }
+    /* Encode EOB decision only if k <= DCTSIZE2 - 1 */
+    if (k <= DCTSIZE2 - 1) {
+      st = entropy->ac_stats[tbl] + 3 * (k - 1);
+      arith_encode(cinfo, st, 1);
+    }
+  }
+
+  return TRUE;
+}
+
+
+/*
+ * Initialize for an arithmetic-compressed scan.
+ */
+
+METHODDEF(void)
+start_pass (j_compress_ptr cinfo, boolean gather_statistics)
+{
+  arith_entropy_ptr entropy = (arith_entropy_ptr) cinfo->entropy;
+  int ci, tbl;
+  jpeg_component_info * compptr;
+
+  if (gather_statistics)
+    /* Make sure to avoid that in the master control logic!
+     * We are fully adaptive here and need no extra
+     * statistics gathering pass!
+     */
+    ERREXIT(cinfo, JERR_NOT_COMPILED);
+
+  /* We assume jcmaster.c already validated the progressive scan parameters. */
+
+  /* Select execution routines */
+  if (cinfo->progressive_mode) {
+    if (cinfo->Ah == 0) {
+      if (cinfo->Ss == 0)
+	entropy->pub.encode_mcu = encode_mcu_DC_first;
+      else
+	entropy->pub.encode_mcu = encode_mcu_AC_first;
+    } else {
+      if (cinfo->Ss == 0)
+	entropy->pub.encode_mcu = encode_mcu_DC_refine;
+      else
+	entropy->pub.encode_mcu = encode_mcu_AC_refine;
+    }
+  } else
+    entropy->pub.encode_mcu = encode_mcu;
+
+  /* Allocate & initialize requested statistics areas */
+  for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
+    compptr = cinfo->cur_comp_info[ci];
+    /* DC needs no table for refinement scan */
+    if (cinfo->progressive_mode == 0 || (cinfo->Ss == 0 && cinfo->Ah == 0)) {
+      tbl = compptr->dc_tbl_no;
+      if (tbl < 0 || tbl >= NUM_ARITH_TBLS)
+	ERREXIT1(cinfo, JERR_NO_ARITH_TABLE, tbl);
+      if (entropy->dc_stats[tbl] == NULL)
+	entropy->dc_stats[tbl] = (unsigned char *) (*cinfo->mem->alloc_small)
+	  ((j_common_ptr) cinfo, JPOOL_IMAGE, DC_STAT_BINS);
+      MEMZERO(entropy->dc_stats[tbl], DC_STAT_BINS);
+      /* Initialize DC predictions to 0 */
+      entropy->last_dc_val[ci] = 0;
+      entropy->dc_context[ci] = 0;
+    }
+    /* AC needs no table when not present */
+    if (cinfo->progressive_mode == 0 || cinfo->Se) {
+      tbl = compptr->ac_tbl_no;
+      if (tbl < 0 || tbl >= NUM_ARITH_TBLS)
+	ERREXIT1(cinfo, JERR_NO_ARITH_TABLE, tbl);
+      if (entropy->ac_stats[tbl] == NULL)
+	entropy->ac_stats[tbl] = (unsigned char *) (*cinfo->mem->alloc_small)
+	  ((j_common_ptr) cinfo, JPOOL_IMAGE, AC_STAT_BINS);
+      MEMZERO(entropy->ac_stats[tbl], AC_STAT_BINS);
+#ifdef CALCULATE_SPECTRAL_CONDITIONING
+      if (cinfo->progressive_mode)
+	/* Section G.1.3.2: Set appropriate arithmetic conditioning value Kx */
+	cinfo->arith_ac_K[tbl] = cinfo->Ss + ((8 + cinfo->Se - cinfo->Ss) >> 4);
+#endif
+    }
+  }
+
+  /* Initialize arithmetic encoding variables */
+  entropy->c = 0;
+  entropy->a = 0x10000L;
+  entropy->sc = 0;
+  entropy->zc = 0;
+  entropy->ct = 11;
+  entropy->buffer = -1;  /* empty */
+
+  /* Initialize restart stuff */
+  entropy->restarts_to_go = cinfo->restart_interval;
+  entropy->next_restart_num = 0;
+}
+
+
+/*
+ * Module initialization routine for arithmetic entropy encoding.
+ */
+
+GLOBAL(void)
+jinit_arith_encoder (j_compress_ptr cinfo)
+{
+  arith_entropy_ptr entropy;
+  int i;
+
+  entropy = (arith_entropy_ptr)
+    (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
+				SIZEOF(arith_entropy_encoder));
+  cinfo->entropy = (struct jpeg_entropy_encoder *) entropy;
+  entropy->pub.start_pass = start_pass;
+  entropy->pub.finish_pass = finish_pass;
+
+  /* Mark tables unallocated */
+  for (i = 0; i < NUM_ARITH_TBLS; i++) {
+    entropy->dc_stats[i] = NULL;
+    entropy->ac_stats[i] = NULL;
+  }
+
+  /* Initialize index for fixed probability estimation */
+  entropy->fixed_bin[0] = 113;
+}

=== added file 'src/libjpeg-turbo/jccoefct.c'
--- src/libjpeg-turbo/jccoefct.c	1970-01-01 00:00:00 +0000
+++ src/libjpeg-turbo/jccoefct.c	2012-06-27 16:20:24 +0000
@@ -0,0 +1,449 @@
+/*
+ * jccoefct.c
+ *
+ * Copyright (C) 1994-1997, Thomas G. Lane.
+ * This file is part of the Independent JPEG Group's software.
+ * For conditions of distribution and use, see the accompanying README file.
+ *
+ * This file contains the coefficient buffer controller for compression.
+ * This controller is the top level of the JPEG compressor proper.
+ * The coefficient buffer lies between forward-DCT and entropy encoding steps.
+ */
+
+#define JPEG_INTERNALS
+#include "jinclude.h"
+#include "jpeglib.h"
+
+
+/* We use a full-image coefficient buffer when doing Huffman optimization,
+ * and also for writing multiple-scan JPEG files.  In all cases, the DCT
+ * step is run during the first pass, and subsequent passes need only read
+ * the buffered coefficients.
+ */
+#ifdef ENTROPY_OPT_SUPPORTED
+#define FULL_COEF_BUFFER_SUPPORTED
+#else
+#ifdef C_MULTISCAN_FILES_SUPPORTED
+#define FULL_COEF_BUFFER_SUPPORTED
+#endif
+#endif
+
+
+/* Private buffer controller object */
+
+typedef struct {
+  struct jpeg_c_coef_controller pub; /* public fields */
+
+  JDIMENSION iMCU_row_num;	/* iMCU row # within image */
+  JDIMENSION mcu_ctr;		/* counts MCUs processed in current row */
+  int MCU_vert_offset;		/* counts MCU rows within iMCU row */
+  int MCU_rows_per_iMCU_row;	/* number of such rows needed */
+
+  /* For single-pass compression, it's sufficient to buffer just one MCU
+   * (although this may prove a bit slow in practice).  We allocate a
+   * workspace of C_MAX_BLOCKS_IN_MCU coefficient blocks, and reuse it for each
+   * MCU constructed and sent.  (On 80x86, the workspace is FAR even though
+   * it's not really very big; this is to keep the module interfaces unchanged
+   * when a large coefficient buffer is necessary.)
+   * In multi-pass modes, this array points to the current MCU's blocks
+   * within the virtual arrays.
+   */
+  JBLOCKROW MCU_buffer[C_MAX_BLOCKS_IN_MCU];
+
+  /* In multi-pass modes, we need a virtual block array for each component. */
+  jvirt_barray_ptr whole_image[MAX_COMPONENTS];
+} my_coef_controller;
+
+typedef my_coef_controller * my_coef_ptr;
+
+
+/* Forward declarations */
+METHODDEF(boolean) compress_data
+    JPP((j_compress_ptr cinfo, JSAMPIMAGE input_buf));
+#ifdef FULL_COEF_BUFFER_SUPPORTED
+METHODDEF(boolean) compress_first_pass
+    JPP((j_compress_ptr cinfo, JSAMPIMAGE input_buf));
+METHODDEF(boolean) compress_output
+    JPP((j_compress_ptr cinfo, JSAMPIMAGE input_buf));
+#endif
+
+
+LOCAL(void)
+start_iMCU_row (j_compress_ptr cinfo)
+/* Reset within-iMCU-row counters for a new row */
+{
+  my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
+
+  /* In an interleaved scan, an MCU row is the same as an iMCU row.
+   * In a noninterleaved scan, an iMCU row has v_samp_factor MCU rows.
+   * But at the bottom of the image, process only what's left.
+   */
+  if (cinfo->comps_in_scan > 1) {
+    coef->MCU_rows_per_iMCU_row = 1;
+  } else {
+    if (coef->iMCU_row_num < (cinfo->total_iMCU_rows-1))
+      coef->MCU_rows_per_iMCU_row = cinfo->cur_comp_info[0]->v_samp_factor;
+    else
+      coef->MCU_rows_per_iMCU_row = cinfo->cur_comp_info[0]->last_row_height;
+  }
+
+  coef->mcu_ctr = 0;
+  coef->MCU_vert_offset = 0;
+}
+
+
+/*
+ * Initialize for a processing pass.
+ */
+
+METHODDEF(void)
+start_pass_coef (j_compress_ptr cinfo, J_BUF_MODE pass_mode)
+{
+  my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
+
+  coef->iMCU_row_num = 0;
+  start_iMCU_row(cinfo);
+
+  switch (pass_mode) {
+  case JBUF_PASS_THRU:
+    if (coef->whole_image[0] != NULL)
+      ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
+    coef->pub.compress_data = compress_data;
+    break;
+#ifdef FULL_COEF_BUFFER_SUPPORTED
+  case JBUF_SAVE_AND_PASS:
+    if (coef->whole_image[0] == NULL)
+      ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
+    coef->pub.compress_data = compress_first_pass;
+    break;
+  case JBUF_CRANK_DEST:
+    if (coef->whole_image[0] == NULL)
+      ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
+    coef->pub.compress_data = compress_output;
+    break;
+#endif
+  default:
+    ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
+    break;
+  }
+}
+
+
+/*
+ * Process some data in the single-pass case.
+ * We process the equivalent of one fully interleaved MCU row ("iMCU" row)
+ * per call, ie, v_samp_factor block rows for each component in the image.
+ * Returns TRUE if the iMCU row is completed, FALSE if suspended.
+ *
+ * NB: input_buf contains a plane for each component in image,
+ * which we index according to the component's SOF position.
+ */
+
+METHODDEF(boolean)
+compress_data (j_compress_ptr cinfo, JSAMPIMAGE input_buf)
+{
+  my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
+  JDIMENSION MCU_col_num;	/* index of current MCU within row */
+  JDIMENSION last_MCU_col = cinfo->MCUs_per_row - 1;
+  JDIMENSION last_iMCU_row = cinfo->total_iMCU_rows - 1;
+  int blkn, bi, ci, yindex, yoffset, blockcnt;
+  JDIMENSION ypos, xpos;
+  jpeg_component_info *compptr;
+
+  /* Loop to write as much as one whole iMCU row */
+  for (yoffset = coef->MCU_vert_offset; yoffset < coef->MCU_rows_per_iMCU_row;
+       yoffset++) {
+    for (MCU_col_num = coef->mcu_ctr; MCU_col_num <= last_MCU_col;
+	 MCU_col_num++) {
+      /* Determine where data comes from in input_buf and do the DCT thing.
+       * Each call on forward_DCT processes a horizontal row of DCT blocks
+       * as wide as an MCU; we rely on having allocated the MCU_buffer[] blocks
+       * sequentially.  Dummy blocks at the right or bottom edge are filled in
+       * specially.  The data in them does not matter for image reconstruction,
+       * so we fill them with values that will encode to the smallest amount of
+       * data, viz: all zeroes in the AC entries, DC entries equal to previous
+       * block's DC value.  (Thanks to Thomas Kinsman for this idea.)
+       */
+      blkn = 0;
+      for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
+	compptr = cinfo->cur_comp_info[ci];
+	blockcnt = (MCU_col_num < last_MCU_col) ? compptr->MCU_width
+						: compptr->last_col_width;
+	xpos = MCU_col_num * compptr->MCU_sample_width;
+	ypos = yoffset * DCTSIZE; /* ypos == (yoffset+yindex) * DCTSIZE */
+	for (yindex = 0; yindex < compptr->MCU_height; yindex++) {
+	  if (coef->iMCU_row_num < last_iMCU_row ||
+	      yoffset+yindex < compptr->last_row_height) {
+	    (*cinfo->fdct->forward_DCT) (cinfo, compptr,
+					 input_buf[compptr->component_index],
+					 coef->MCU_buffer[blkn],
+					 ypos, xpos, (JDIMENSION) blockcnt);
+	    if (blockcnt < compptr->MCU_width) {
+	      /* Create some dummy blocks at the right edge of the image. */
+	      jzero_far((void FAR *) coef->MCU_buffer[blkn + blockcnt],
+			(compptr->MCU_width - blockcnt) * SIZEOF(JBLOCK));
+	      for (bi = blockcnt; bi < compptr->MCU_width; bi++) {
+		coef->MCU_buffer[blkn+bi][0][0] = coef->MCU_buffer[blkn+bi-1][0][0];
+	      }
+	    }
+	  } else {
+	    /* Create a row of dummy blocks at the bottom of the image. */
+	    jzero_far((void FAR *) coef->MCU_buffer[blkn],
+		      compptr->MCU_width * SIZEOF(JBLOCK));
+	    for (bi = 0; bi < compptr->MCU_width; bi++) {
+	      coef->MCU_buffer[blkn+bi][0][0] = coef->MCU_buffer[blkn-1][0][0];
+	    }
+	  }
+	  blkn += compptr->MCU_width;
+	  ypos += DCTSIZE;
+	}
+      }
+      /* Try to write the MCU.  In event of a suspension failure, we will
+       * re-DCT the MCU on restart (a bit inefficient, could be fixed...)
+       */
+      if (! (*cinfo->entropy->encode_mcu) (cinfo, coef->MCU_buffer)) {
+	/* Suspension forced; update state counters and exit */
+	coef->MCU_vert_offset = yoffset;
+	coef->mcu_ctr = MCU_col_num;
+	return FALSE;
+      }
+    }
+    /* Completed an MCU row, but perhaps not an iMCU row */
+    coef->mcu_ctr = 0;
+  }
+  /* Completed the iMCU row, advance counters for next one */
+  coef->iMCU_row_num++;
+  start_iMCU_row(cinfo);
+  return TRUE;
+}
+
+
+#ifdef FULL_COEF_BUFFER_SUPPORTED
+
+/*
+ * Process some data in the first pass of a multi-pass case.
+ * We process the equivalent of one fully interleaved MCU row ("iMCU" row)
+ * per call, ie, v_samp_factor block rows for each component in the image.
+ * This amount of data is read from the source buffer, DCT'd and quantized,
+ * and saved into the virtual arrays.  We also generate suitable dummy blocks
+ * as needed at the right and lower edges.  (The dummy blocks are constructed
+ * in the virtual arrays, which have been padded appropriately.)  This makes
+ * it possible for subsequent passes not to worry about real vs. dummy blocks.
+ *
+ * We must also emit the data to the entropy encoder.  This is conveniently
+ * done by calling compress_output() after we've loaded the current strip
+ * of the virtual arrays.
+ *
+ * NB: input_buf contains a plane for each component in image.  All
+ * components are DCT'd and loaded into the virtual arrays in this pass.
+ * However, it may be that only a subset of the components are emitted to
+ * the entropy encoder during this first pass; be careful about looking
+ * at the scan-dependent variables (MCU dimensions, etc).
+ */
+
+METHODDEF(boolean)
+compress_first_pass (j_compress_ptr cinfo, JSAMPIMAGE input_buf)
+{
+  my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
+  JDIMENSION last_iMCU_row = cinfo->total_iMCU_rows - 1;
+  JDIMENSION blocks_across, MCUs_across, MCUindex;
+  int bi, ci, h_samp_factor, block_row, block_rows, ndummy;
+  JCOEF lastDC;
+  jpeg_component_info *compptr;
+  JBLOCKARRAY buffer;
+  JBLOCKROW thisblockrow, lastblockrow;
+
+  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
+       ci++, compptr++) {
+    /* Align the virtual buffer for this component. */
+    buffer = (*cinfo->mem->access_virt_barray)
+      ((j_common_ptr) cinfo, coef->whole_image[ci],
+       coef->iMCU_row_num * compptr->v_samp_factor,
+       (JDIMENSION) compptr->v_samp_factor, TRUE);
+    /* Count non-dummy DCT block rows in this iMCU row. */
+    if (coef->iMCU_row_num < last_iMCU_row)
+      block_rows = compptr->v_samp_factor;
+    else {
+      /* NB: can't use last_row_height here, since may not be set! */
+      block_rows = (int) (compptr->height_in_blocks % compptr->v_samp_factor);
+      if (block_rows == 0) block_rows = compptr->v_samp_factor;
+    }
+    blocks_across = compptr->width_in_blocks;
+    h_samp_factor = compptr->h_samp_factor;
+    /* Count number of dummy blocks to be added at the right margin. */
+    ndummy = (int) (blocks_across % h_samp_factor);
+    if (ndummy > 0)
+      ndummy = h_samp_factor - ndummy;
+    /* Perform DCT for all non-dummy blocks in this iMCU row.  Each call
+     * on forward_DCT processes a complete horizontal row of DCT blocks.
+     */
+    for (block_row = 0; block_row < block_rows; block_row++) {
+      thisblockrow = buffer[block_row];
+      (*cinfo->fdct->forward_DCT) (cinfo, compptr,
+				   input_buf[ci], thisblockrow,
+				   (JDIMENSION) (block_row * DCTSIZE),
+				   (JDIMENSION) 0, blocks_across);
+      if (ndummy > 0) {
+	/* Create dummy blocks at the right edge of the image. */
+	thisblockrow += blocks_across; /* => first dummy block */
+	jzero_far((void FAR *) thisblockrow, ndummy * SIZEOF(JBLOCK));
+	lastDC = thisblockrow[-1][0];
+	for (bi = 0; bi < ndummy; bi++) {
+	  thisblockrow[bi][0] = lastDC;
+	}
+      }
+    }
+    /* If at end of image, create dummy block rows as needed.
+     * The tricky part here is that within each MCU, we want the DC values
+     * of the dummy blocks to match the last real block's DC value.
+     * This squeezes a few more bytes out of the resulting file...
+     */
+    if (coef->iMCU_row_num == last_iMCU_row) {
+      blocks_across += ndummy;	/* include lower right corner */
+      MCUs_across = blocks_across / h_samp_factor;
+      for (block_row = block_rows; block_row < compptr->v_samp_factor;
+	   block_row++) {
+	thisblockrow = buffer[block_row];
+	lastblockrow = buffer[block_row-1];
+	jzero_far((void FAR *) thisblockrow,
+		  (size_t) (blocks_across * SIZEOF(JBLOCK)));
+	for (MCUindex = 0; MCUindex < MCUs_across; MCUindex++) {
+	  lastDC = lastblockrow[h_samp_factor-1][0];
+	  for (bi = 0; bi < h_samp_factor; bi++) {
+	    thisblockrow[bi][0] = lastDC;
+	  }
+	  thisblockrow += h_samp_factor; /* advance to next MCU in row */
+	  lastblockrow += h_samp_factor;
+	}
+      }
+    }
+  }
+  /* NB: compress_output will increment iMCU_row_num if successful.
+   * A suspension return will result in redoing all the work above next time.
+   */
+
+  /* Emit data to the entropy encoder, sharing code with subsequent passes */
+  return compress_output(cinfo, input_buf);
+}
+
+
+/*
+ * Process some data in subsequent passes of a multi-pass case.
+ * We process the equivalent of one fully interleaved MCU row ("iMCU" row)
+ * per call, ie, v_samp_factor block rows for each component in the scan.
+ * The data is obtained from the virtual arrays and fed to the entropy coder.
+ * Returns TRUE if the iMCU row is completed, FALSE if suspended.
+ *
+ * NB: input_buf is ignored; it is likely to be a NULL pointer.
+ */
+
+METHODDEF(boolean)
+compress_output (j_compress_ptr cinfo, JSAMPIMAGE input_buf)
+{
+  my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
+  JDIMENSION MCU_col_num;	/* index of current MCU within row */
+  int blkn, ci, xindex, yindex, yoffset;
+  JDIMENSION start_col;
+  JBLOCKARRAY buffer[MAX_COMPS_IN_SCAN];
+  JBLOCKROW buffer_ptr;
+  jpeg_component_info *compptr;
+
+  /* Align the virtual buffers for the components used in this scan.
+   * NB: during first pass, this is safe only because the buffers will
+   * already be aligned properly, so jmemmgr.c won't need to do any I/O.
+   */
+  for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
+    compptr = cinfo->cur_comp_info[ci];
+    buffer[ci] = (*cinfo->mem->access_virt_barray)
+      ((j_common_ptr) cinfo, coef->whole_image[compptr->component_index],
+       coef->iMCU_row_num * compptr->v_samp_factor,
+       (JDIMENSION) compptr->v_samp_factor, FALSE);
+  }
+
+  /* Loop to process one whole iMCU row */
+  for (yoffset = coef->MCU_vert_offset; yoffset < coef->MCU_rows_per_iMCU_row;
+       yoffset++) {
+    for (MCU_col_num = coef->mcu_ctr; MCU_col_num < cinfo->MCUs_per_row;
+	 MCU_col_num++) {
+      /* Construct list of pointers to DCT blocks belonging to this MCU */
+      blkn = 0;			/* index of current DCT block within MCU */
+      for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
+	compptr = cinfo->cur_comp_info[ci];
+	start_col = MCU_col_num * compptr->MCU_width;
+	for (yindex = 0; yindex < compptr->MCU_height; yindex++) {
+	  buffer_ptr = buffer[ci][yindex+yoffset] + start_col;
+	  for (xindex = 0; xindex < compptr->MCU_width; xindex++) {
+	    coef->MCU_buffer[blkn++] = buffer_ptr++;
+	  }
+	}
+      }
+      /* Try to write the MCU. */
+      if (! (*cinfo->entropy->encode_mcu) (cinfo, coef->MCU_buffer)) {
+	/* Suspension forced; update state counters and exit */
+	coef->MCU_vert_offset = yoffset;
+	coef->mcu_ctr = MCU_col_num;
+	return FALSE;
+      }
+    }
+    /* Completed an MCU row, but perhaps not an iMCU row */
+    coef->mcu_ctr = 0;
+  }
+  /* Completed the iMCU row, advance counters for next one */
+  coef->iMCU_row_num++;
+  start_iMCU_row(cinfo);
+  return TRUE;
+}
+
+#endif /* FULL_COEF_BUFFER_SUPPORTED */
+
+
+/*
+ * Initialize coefficient buffer controller.
+ */
+
+GLOBAL(void)
+jinit_c_coef_controller (j_compress_ptr cinfo, boolean need_full_buffer)
+{
+  my_coef_ptr coef;
+
+  coef = (my_coef_ptr)
+    (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
+				SIZEOF(my_coef_controller));
+  cinfo->coef = (struct jpeg_c_coef_controller *) coef;
+  coef->pub.start_pass = start_pass_coef;
+
+  /* Create the coefficient buffer. */
+  if (need_full_buffer) {
+#ifdef FULL_COEF_BUFFER_SUPPORTED
+    /* Allocate a full-image virtual array for each component, */
+    /* padded to a multiple of samp_factor DCT blocks in each direction. */
+    int ci;
+    jpeg_component_info *compptr;
+
+    for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
+	 ci++, compptr++) {
+      coef->whole_image[ci] = (*cinfo->mem->request_virt_barray)
+	((j_common_ptr) cinfo, JPOOL_IMAGE, FALSE,
+	 (JDIMENSION) jround_up((long) compptr->width_in_blocks,
+				(long) compptr->h_samp_factor),
+	 (JDIMENSION) jround_up((long) compptr->height_in_blocks,
+				(long) compptr->v_samp_factor),
+	 (JDIMENSION) compptr->v_samp_factor);
+    }
+#else
+    ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
+#endif
+  } else {
+    /* We only need a single-MCU buffer. */
+    JBLOCKROW buffer;
+    int i;
+
+    buffer = (JBLOCKROW)
+      (*cinfo->mem->alloc_large) ((j_common_ptr) cinfo, JPOOL_IMAGE,
+				  C_MAX_BLOCKS_IN_MCU * SIZEOF(JBLOCK));
+    for (i = 0; i < C_MAX_BLOCKS_IN_MCU; i++) {
+      coef->MCU_buffer[i] = buffer + i;
+    }
+    coef->whole_image[0] = NULL; /* flag for no virtual arrays */
+  }
+}

=== added file 'src/libjpeg-turbo/jccolext.c.inc'
--- src/libjpeg-turbo/jccolext.c.inc	1970-01-01 00:00:00 +0000
+++ src/libjpeg-turbo/jccolext.c.inc	2012-06-27 16:20:24 +0000
@@ -0,0 +1,114 @@
+/*
+ * jccolext.c
+ *
+ * Copyright (C) 1991-1996, Thomas G. Lane.
+ * Copyright (C) 2009-2011, D. R. Commander.
+ * This file is part of the Independent JPEG Group's software.
+ * For conditions of distribution and use, see the accompanying README file.
+ *
+ * This file contains input colorspace conversion routines.
+ */
+
+
+/* This file is included by jccolor.c */
+
+
+/*
+ * Convert some rows of samples to the JPEG colorspace.
+ *
+ * Note that we change from the application's interleaved-pixel format
+ * to our internal noninterleaved, one-plane-per-component format.
+ * The input buffer is therefore three times as wide as the output buffer.
+ *
+ * A starting row offset is provided only for the output buffer.  The caller
+ * can easily adjust the passed input_buf value to accommodate any row
+ * offset required on that side.
+ */
+
+INLINE
+LOCAL(void)
+rgb_ycc_convert_internal (j_compress_ptr cinfo,
+                          JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+                          JDIMENSION output_row, int num_rows)
+{
+  my_cconvert_ptr cconvert = (my_cconvert_ptr) cinfo->cconvert;
+  register int r, g, b;
+  register INT32 * ctab = cconvert->rgb_ycc_tab;
+  register JSAMPROW inptr;
+  register JSAMPROW outptr0, outptr1, outptr2;
+  register JDIMENSION col;
+  JDIMENSION num_cols = cinfo->image_width;
+
+  while (--num_rows >= 0) {
+    inptr = *input_buf++;
+    outptr0 = output_buf[0][output_row];
+    outptr1 = output_buf[1][output_row];
+    outptr2 = output_buf[2][output_row];
+    output_row++;
+    for (col = 0; col < num_cols; col++) {
+      r = GETJSAMPLE(inptr[RGB_RED]);
+      g = GETJSAMPLE(inptr[RGB_GREEN]);
+      b = GETJSAMPLE(inptr[RGB_BLUE]);
+      inptr += RGB_PIXELSIZE;
+      /* If the inputs are 0..MAXJSAMPLE, the outputs of these equations
+       * must be too; we do not need an explicit range-limiting operation.
+       * Hence the value being shifted is never negative, and we don't
+       * need the general RIGHT_SHIFT macro.
+       */
+      /* Y */
+      outptr0[col] = (JSAMPLE)
+		((ctab[r+R_Y_OFF] + ctab[g+G_Y_OFF] + ctab[b+B_Y_OFF])
+		 >> SCALEBITS);
+      /* Cb */
+      outptr1[col] = (JSAMPLE)
+		((ctab[r+R_CB_OFF] + ctab[g+G_CB_OFF] + ctab[b+B_CB_OFF])
+		 >> SCALEBITS);
+      /* Cr */
+      outptr2[col] = (JSAMPLE)
+		((ctab[r+R_CR_OFF] + ctab[g+G_CR_OFF] + ctab[b+B_CR_OFF])
+		 >> SCALEBITS);
+    }
+  }
+}
+
+
+/**************** Cases other than RGB -> YCbCr **************/
+
+
+/*
+ * Convert some rows of samples to the JPEG colorspace.
+ * This version handles RGB->grayscale conversion, which is the same
+ * as the RGB->Y portion of RGB->YCbCr.
+ * We assume rgb_ycc_start has been called (we only use the Y tables).
+ */
+
+INLINE
+LOCAL(void)
+rgb_gray_convert_internal (j_compress_ptr cinfo,
+                           JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+                           JDIMENSION output_row, int num_rows)
+{
+  my_cconvert_ptr cconvert = (my_cconvert_ptr) cinfo->cconvert;
+  register int r, g, b;
+  register INT32 * ctab = cconvert->rgb_ycc_tab;
+  register JSAMPROW inptr;
+  register JSAMPROW outptr;
+  register JDIMENSION col;
+  JDIMENSION num_cols = cinfo->image_width;
+
+  while (--num_rows >= 0) {
+    inptr = *input_buf++;
+    outptr = output_buf[0][output_row];
+    output_row++;
+    for (col = 0; col < num_cols; col++) {
+      r = GETJSAMPLE(inptr[RGB_RED]);
+      g = GETJSAMPLE(inptr[RGB_GREEN]);
+      b = GETJSAMPLE(inptr[RGB_BLUE]);
+      inptr += RGB_PIXELSIZE;
+      /* Y */
+      outptr[col] = (JSAMPLE)
+		((ctab[r+R_Y_OFF] + ctab[g+G_Y_OFF] + ctab[b+B_Y_OFF])
+		 >> SCALEBITS);
+    }
+  }
+}

=== added file 'src/libjpeg-turbo/jccolor.c'
--- src/libjpeg-turbo/jccolor.c	1970-01-01 00:00:00 +0000
+++ src/libjpeg-turbo/jccolor.c	2012-06-27 16:20:24 +0000
@@ -0,0 +1,599 @@
+/*
+ * jccolor.c
+ *
+ * Copyright (C) 1991-1996, Thomas G. Lane.
+ * Copyright 2009 Pierre Ossman <[e-mail]> for Cendio AB
+ * Copyright (C) 2009-2011, D. R. Commander.
+ * This file is part of the Independent JPEG Group's software.
+ * For conditions of distribution and use, see the accompanying README file.
+ *
+ * This file contains input colorspace conversion routines.
+ */
+
+#define JPEG_INTERNALS
+#include "jinclude.h"
+#include "jpeglib.h"
+#include "jsimd.h"
+#include "config.h"
+
+
+/* Private subobject */
+
+typedef struct {
+  struct jpeg_color_converter pub; /* public fields */
+
+  /* Private state for RGB->YCC conversion */
+  INT32 * rgb_ycc_tab;		/* => table for RGB to YCbCr conversion */
+} my_color_converter;
+
+typedef my_color_converter * my_cconvert_ptr;
+
+
+/**************** RGB -> YCbCr conversion: most common case **************/
+
+/*
+ * YCbCr is defined per CCIR 601-1, except that Cb and Cr are
+ * normalized to the range 0..MAXJSAMPLE rather than -0.5 .. 0.5.
+ * The conversion equations to be implemented are therefore
+ *	Y  =  0.29900 * R + 0.58700 * G + 0.11400 * B
+ *	Cb = -0.16874 * R - 0.33126 * G + 0.50000 * B  + CENTERJSAMPLE
+ *	Cr =  0.50000 * R - 0.41869 * G - 0.08131 * B  + CENTERJSAMPLE
+ * (These numbers are derived from TIFF 6.0 section 21, dated 3-June-92.)
+ * Note: older versions of the IJG code used a zero offset of MAXJSAMPLE/2,
+ * rather than CENTERJSAMPLE, for Cb and Cr.  This gave equal positive and
+ * negative swings for Cb/Cr, but meant that grayscale values (Cb=Cr=0)
+ * were not represented exactly.  Now we sacrifice exact representation of
+ * maximum red and maximum blue in order to get exact grayscales.
+ *
+ * To avoid floating-point arithmetic, we represent the fractional constants
+ * as integers scaled up by 2^16 (about 4 digits precision); we have to divide
+ * the products by 2^16, with appropriate rounding, to get the correct answer.
+ *
+ * For even more speed, we avoid doing any multiplications in the inner loop
+ * by precalculating the constants times R,G,B for all possible values.
+ * For 8-bit JSAMPLEs this is very reasonable (only 256 entries per table);
+ * for 12-bit samples it is still acceptable.  It's not very reasonable for
+ * 16-bit samples, but if you want lossless storage you shouldn't be changing
+ * colorspace anyway.
+ * The CENTERJSAMPLE offsets and the rounding fudge-factor of 0.5 are included
+ * in the tables to save adding them separately in the inner loop.
+ */
+
+#define SCALEBITS	16	/* speediest right-shift on some machines */
+#define CBCR_OFFSET	((INT32) CENTERJSAMPLE << SCALEBITS)
+#define ONE_HALF	((INT32) 1 << (SCALEBITS-1))
+#define FIX(x)		((INT32) ((x) * (1L<<SCALEBITS) + 0.5))
+
+/* We allocate one big table and divide it up into eight parts, instead of
+ * doing eight alloc_small requests.  This lets us use a single table base
+ * address, which can be held in a register in the inner loops on many
+ * machines (more than can hold all eight addresses, anyway).
+ */
+
+#define R_Y_OFF		0			/* offset to R => Y section */
+#define G_Y_OFF		(1*(MAXJSAMPLE+1))	/* offset to G => Y section */
+#define B_Y_OFF		(2*(MAXJSAMPLE+1))	/* etc. */
+#define R_CB_OFF	(3*(MAXJSAMPLE+1))
+#define G_CB_OFF	(4*(MAXJSAMPLE+1))
+#define B_CB_OFF	(5*(MAXJSAMPLE+1))
+#define R_CR_OFF	B_CB_OFF		/* B=>Cb, R=>Cr are the same */
+#define G_CR_OFF	(6*(MAXJSAMPLE+1))
+#define B_CR_OFF	(7*(MAXJSAMPLE+1))
+#define TABLE_SIZE	(8*(MAXJSAMPLE+1))
+
+
+/* Include inline routines for colorspace extensions */
+
+#include "jccolext.c.inc"
+#undef RGB_RED
+#undef RGB_GREEN
+#undef RGB_BLUE
+#undef RGB_PIXELSIZE
+
+#define RGB_RED EXT_RGB_RED
+#define RGB_GREEN EXT_RGB_GREEN
+#define RGB_BLUE EXT_RGB_BLUE
+#define RGB_PIXELSIZE EXT_RGB_PIXELSIZE
+#define rgb_ycc_convert_internal extrgb_ycc_convert_internal
+#define rgb_gray_convert_internal extrgb_gray_convert_internal
+#include "jccolext.c.inc"
+#undef RGB_RED
+#undef RGB_GREEN
+#undef RGB_BLUE
+#undef RGB_PIXELSIZE
+#undef rgb_ycc_convert_internal
+#undef rgb_gray_convert_internal
+
+#define RGB_RED EXT_RGBX_RED
+#define RGB_GREEN EXT_RGBX_GREEN
+#define RGB_BLUE EXT_RGBX_BLUE
+#define RGB_PIXELSIZE EXT_RGBX_PIXELSIZE
+#define rgb_ycc_convert_internal extrgbx_ycc_convert_internal
+#define rgb_gray_convert_internal extrgbx_gray_convert_internal
+#include "jccolext.c.inc"
+#undef RGB_RED
+#undef RGB_GREEN
+#undef RGB_BLUE
+#undef RGB_PIXELSIZE
+#undef rgb_ycc_convert_internal
+#undef rgb_gray_convert_internal
+
+#define RGB_RED EXT_BGR_RED
+#define RGB_GREEN EXT_BGR_GREEN
+#define RGB_BLUE EXT_BGR_BLUE
+#define RGB_PIXELSIZE EXT_BGR_PIXELSIZE
+#define rgb_ycc_convert_internal extbgr_ycc_convert_internal
+#define rgb_gray_convert_internal extbgr_gray_convert_internal
+#include "jccolext.c.inc"
+#undef RGB_RED
+#undef RGB_GREEN
+#undef RGB_BLUE
+#undef RGB_PIXELSIZE
+#undef rgb_ycc_convert_internal
+#undef rgb_gray_convert_internal
+
+#define RGB_RED EXT_BGRX_RED
+#define RGB_GREEN EXT_BGRX_GREEN
+#define RGB_BLUE EXT_BGRX_BLUE
+#define RGB_PIXELSIZE EXT_BGRX_PIXELSIZE
+#define rgb_ycc_convert_internal extbgrx_ycc_convert_internal
+#define rgb_gray_convert_internal extbgrx_gray_convert_internal
+#include "jccolext.c.inc"
+#undef RGB_RED
+#undef RGB_GREEN
+#undef RGB_BLUE
+#undef RGB_PIXELSIZE
+#undef rgb_ycc_convert_internal
+#undef rgb_gray_convert_internal
+
+#define RGB_RED EXT_XBGR_RED
+#define RGB_GREEN EXT_XBGR_GREEN
+#define RGB_BLUE EXT_XBGR_BLUE
+#define RGB_PIXELSIZE EXT_XBGR_PIXELSIZE
+#define rgb_ycc_convert_internal extxbgr_ycc_convert_internal
+#define rgb_gray_convert_internal extxbgr_gray_convert_internal
+#include "jccolext.c.inc"
+#undef RGB_RED
+#undef RGB_GREEN
+#undef RGB_BLUE
+#undef RGB_PIXELSIZE
+#undef rgb_ycc_convert_internal
+#undef rgb_gray_convert_internal
+
+#define RGB_RED EXT_XRGB_RED
+#define RGB_GREEN EXT_XRGB_GREEN
+#define RGB_BLUE EXT_XRGB_BLUE
+#define RGB_PIXELSIZE EXT_XRGB_PIXELSIZE
+#define rgb_ycc_convert_internal extxrgb_ycc_convert_internal
+#define rgb_gray_convert_internal extxrgb_gray_convert_internal
+#include "jccolext.c.inc"
+#undef RGB_RED
+#undef RGB_GREEN
+#undef RGB_BLUE
+#undef RGB_PIXELSIZE
+#undef rgb_ycc_convert_internal
+#undef rgb_gray_convert_internal
+
+
+/*
+ * Initialize for RGB->YCC colorspace conversion.
+ */
+
+METHODDEF(void)
+rgb_ycc_start (j_compress_ptr cinfo)
+{
+  my_cconvert_ptr cconvert = (my_cconvert_ptr) cinfo->cconvert;
+  INT32 * rgb_ycc_tab;
+  INT32 i;
+
+  /* Allocate and fill in the conversion tables. */
+  cconvert->rgb_ycc_tab = rgb_ycc_tab = (INT32 *)
+    (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
+				(TABLE_SIZE * SIZEOF(INT32)));
+
+  for (i = 0; i <= MAXJSAMPLE; i++) {
+    rgb_ycc_tab[i+R_Y_OFF] = FIX(0.29900) * i;
+    rgb_ycc_tab[i+G_Y_OFF] = FIX(0.58700) * i;
+    rgb_ycc_tab[i+B_Y_OFF] = FIX(0.11400) * i     + ONE_HALF;
+    rgb_ycc_tab[i+R_CB_OFF] = (-FIX(0.16874)) * i;
+    rgb_ycc_tab[i+G_CB_OFF] = (-FIX(0.33126)) * i;
+    /* We use a rounding fudge-factor of 0.5-epsilon for Cb and Cr.
+     * This ensures that the maximum output will round to MAXJSAMPLE
+     * not MAXJSAMPLE+1, and thus that we don't have to range-limit.
+     */
+    rgb_ycc_tab[i+B_CB_OFF] = FIX(0.50000) * i    + CBCR_OFFSET + ONE_HALF-1;
+/*  B=>Cb and R=>Cr tables are the same
+    rgb_ycc_tab[i+R_CR_OFF] = FIX(0.50000) * i    + CBCR_OFFSET + ONE_HALF-1;
+*/
+    rgb_ycc_tab[i+G_CR_OFF] = (-FIX(0.41869)) * i;
+    rgb_ycc_tab[i+B_CR_OFF] = (-FIX(0.08131)) * i;
+  }
+}
+
+
+/*
+ * Convert some rows of samples to the JPEG colorspace.
+ */
+
+METHODDEF(void)
+rgb_ycc_convert (j_compress_ptr cinfo,
+		 JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+		 JDIMENSION output_row, int num_rows)
+{
+  switch (cinfo->in_color_space) {
+    case JCS_EXT_RGB:
+      extrgb_ycc_convert_internal(cinfo, input_buf, output_buf, output_row,
+                                  num_rows);
+      break;
+    case JCS_EXT_RGBX:
+    case JCS_EXT_RGBA:
+      extrgbx_ycc_convert_internal(cinfo, input_buf, output_buf, output_row,
+                                   num_rows);
+      break;
+    case JCS_EXT_BGR:
+      extbgr_ycc_convert_internal(cinfo, input_buf, output_buf, output_row,
+                                  num_rows);
+      break;
+    case JCS_EXT_BGRX:
+    case JCS_EXT_BGRA:
+      extbgrx_ycc_convert_internal(cinfo, input_buf, output_buf, output_row,
+                                   num_rows);
+      break;
+    case JCS_EXT_XBGR:
+    case JCS_EXT_ABGR:
+      extxbgr_ycc_convert_internal(cinfo, input_buf, output_buf, output_row,
+                                   num_rows);
+      break;
+    case JCS_EXT_XRGB:
+    case JCS_EXT_ARGB:
+      extxrgb_ycc_convert_internal(cinfo, input_buf, output_buf, output_row,
+                                   num_rows);
+      break;
+    default:
+      rgb_ycc_convert_internal(cinfo, input_buf, output_buf, output_row,
+                               num_rows);
+      break;
+  }
+}
+
+
+/**************** Cases other than RGB -> YCbCr **************/
+
+
+/*
+ * Convert some rows of samples to the JPEG colorspace.
+ */
+
+METHODDEF(void)
+rgb_gray_convert (j_compress_ptr cinfo,
+		  JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+		  JDIMENSION output_row, int num_rows)
+{
+  switch (cinfo->in_color_space) {
+    case JCS_EXT_RGB:
+      extrgb_gray_convert_internal(cinfo, input_buf, output_buf, output_row,
+                                   num_rows);
+      break;
+    case JCS_EXT_RGBX:
+    case JCS_EXT_RGBA:
+      extrgbx_gray_convert_internal(cinfo, input_buf, output_buf, output_row,
+                                    num_rows);
+      break;
+    case JCS_EXT_BGR:
+      extbgr_gray_convert_internal(cinfo, input_buf, output_buf, output_row,
+                                   num_rows);
+      break;
+    case JCS_EXT_BGRX:
+    case JCS_EXT_BGRA:
+      extbgrx_gray_convert_internal(cinfo, input_buf, output_buf, output_row,
+                                    num_rows);
+      break;
+    case JCS_EXT_XBGR:
+    case JCS_EXT_ABGR:
+      extxbgr_gray_convert_internal(cinfo, input_buf, output_buf, output_row,
+                                    num_rows);
+      break;
+    case JCS_EXT_XRGB:
+    case JCS_EXT_ARGB:
+      extxrgb_gray_convert_internal(cinfo, input_buf, output_buf, output_row,
+                                    num_rows);
+      break;
+    default:
+      rgb_gray_convert_internal(cinfo, input_buf, output_buf, output_row,
+                                num_rows);
+      break;
+  }
+}
+
+
+/*
+ * Convert some rows of samples to the JPEG colorspace.
+ * This version handles Adobe-style CMYK->YCCK conversion,
+ * where we convert R=1-C, G=1-M, and B=1-Y to YCbCr using the same
+ * conversion as above, while passing K (black) unchanged.
+ * We assume rgb_ycc_start has been called.
+ */
+
+METHODDEF(void)
+cmyk_ycck_convert (j_compress_ptr cinfo,
+		   JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+		   JDIMENSION output_row, int num_rows)
+{
+  my_cconvert_ptr cconvert = (my_cconvert_ptr) cinfo->cconvert;
+  register int r, g, b;
+  register INT32 * ctab = cconvert->rgb_ycc_tab;
+  register JSAMPROW inptr;
+  register JSAMPROW outptr0, outptr1, outptr2, outptr3;
+  register JDIMENSION col;
+  JDIMENSION num_cols = cinfo->image_width;
+
+  while (--num_rows >= 0) {
+    inptr = *input_buf++;
+    outptr0 = output_buf[0][output_row];
+    outptr1 = output_buf[1][output_row];
+    outptr2 = output_buf[2][output_row];
+    outptr3 = output_buf[3][output_row];
+    output_row++;
+    for (col = 0; col < num_cols; col++) {
+      r = MAXJSAMPLE - GETJSAMPLE(inptr[0]);
+      g = MAXJSAMPLE - GETJSAMPLE(inptr[1]);
+      b = MAXJSAMPLE - GETJSAMPLE(inptr[2]);
+      /* K passes through as-is */
+      outptr3[col] = inptr[3];	/* don't need GETJSAMPLE here */
+      inptr += 4;
+      /* If the inputs are 0..MAXJSAMPLE, the outputs of these equations
+       * must be too; we do not need an explicit range-limiting operation.
+       * Hence the value being shifted is never negative, and we don't
+       * need the general RIGHT_SHIFT macro.
+       */
+      /* Y */
+      outptr0[col] = (JSAMPLE)
+		((ctab[r+R_Y_OFF] + ctab[g+G_Y_OFF] + ctab[b+B_Y_OFF])
+		 >> SCALEBITS);
+      /* Cb */
+      outptr1[col] = (JSAMPLE)
+		((ctab[r+R_CB_OFF] + ctab[g+G_CB_OFF] + ctab[b+B_CB_OFF])
+		 >> SCALEBITS);
+      /* Cr */
+      outptr2[col] = (JSAMPLE)
+		((ctab[r+R_CR_OFF] + ctab[g+G_CR_OFF] + ctab[b+B_CR_OFF])
+		 >> SCALEBITS);
+    }
+  }
+}
+
+
+/*
+ * Convert some rows of samples to the JPEG colorspace.
+ * This version handles grayscale output with no conversion.
+ * The source can be either plain grayscale or YCbCr (since Y == gray).
+ */
+
+METHODDEF(void)
+grayscale_convert (j_compress_ptr cinfo,
+		   JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+		   JDIMENSION output_row, int num_rows)
+{
+  register JSAMPROW inptr;
+  register JSAMPROW outptr;
+  register JDIMENSION col;
+  JDIMENSION num_cols = cinfo->image_width;
+  int instride = cinfo->input_components;
+
+  while (--num_rows >= 0) {
+    inptr = *input_buf++;
+    outptr = output_buf[0][output_row];
+    output_row++;
+    for (col = 0; col < num_cols; col++) {
+      outptr[col] = inptr[0];	/* don't need GETJSAMPLE() here */
+      inptr += instride;
+    }
+  }
+}
+
+
+/*
+ * Convert some rows of samples to the JPEG colorspace.
+ * This version handles multi-component colorspaces without conversion.
+ * We assume input_components == num_components.
+ */
+
+METHODDEF(void)
+null_convert (j_compress_ptr cinfo,
+	      JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+	      JDIMENSION output_row, int num_rows)
+{
+  register JSAMPROW inptr;
+  register JSAMPROW outptr;
+  register JDIMENSION col;
+  register int ci;
+  int nc = cinfo->num_components;
+  JDIMENSION num_cols = cinfo->image_width;
+
+  while (--num_rows >= 0) {
+    /* It seems fastest to make a separate pass for each component. */
+    for (ci = 0; ci < nc; ci++) {
+      inptr = *input_buf;
+      outptr = output_buf[ci][output_row];
+      for (col = 0; col < num_cols; col++) {
+	outptr[col] = inptr[ci]; /* don't need GETJSAMPLE() here */
+	inptr += nc;
+      }
+    }
+    input_buf++;
+    output_row++;
+  }
+}
+
+
+/*
+ * Empty method for start_pass.
+ */
+
+METHODDEF(void)
+null_method (j_compress_ptr cinfo)
+{
+  /* no work needed */
+}
+
+
+/*
+ * Module initialization routine for input colorspace conversion.
+ */
+
+GLOBAL(void)
+jinit_color_converter (j_compress_ptr cinfo)
+{
+  my_cconvert_ptr cconvert;
+
+  cconvert = (my_cconvert_ptr)
+    (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
+				SIZEOF(my_color_converter));
+  cinfo->cconvert = (struct jpeg_color_converter *) cconvert;
+  /* set start_pass to null method until we find out differently */
+  cconvert->pub.start_pass = null_method;
+
+  /* Make sure input_components agrees with in_color_space */
+  switch (cinfo->in_color_space) {
+  case JCS_GRAYSCALE:
+    if (cinfo->input_components != 1)
+      ERREXIT(cinfo, JERR_BAD_IN_COLORSPACE);
+    break;
+
+  case JCS_RGB:
+  case JCS_EXT_RGB:
+  case JCS_EXT_RGBX:
+  case JCS_EXT_BGR:
+  case JCS_EXT_BGRX:
+  case JCS_EXT_XBGR:
+  case JCS_EXT_XRGB:
+  case JCS_EXT_RGBA:
+  case JCS_EXT_BGRA:
+  case JCS_EXT_ABGR:
+  case JCS_EXT_ARGB:
+    if (cinfo->input_components != rgb_pixelsize[cinfo->in_color_space])
+      ERREXIT(cinfo, JERR_BAD_IN_COLORSPACE);
+    break;
+
+  case JCS_YCbCr:
+    if (cinfo->input_components != 3)
+      ERREXIT(cinfo, JERR_BAD_IN_COLORSPACE);
+    break;
+
+  case JCS_CMYK:
+  case JCS_YCCK:
+    if (cinfo->input_components != 4)
+      ERREXIT(cinfo, JERR_BAD_IN_COLORSPACE);
+    break;
+
+  default:			/* JCS_UNKNOWN can be anything */
+    if (cinfo->input_components < 1)
+      ERREXIT(cinfo, JERR_BAD_IN_COLORSPACE);
+    break;
+  }
+
+  /* Check num_components, set conversion method based on requested space */
+  switch (cinfo->jpeg_color_space) {
+  case JCS_GRAYSCALE:
+    if (cinfo->num_components != 1)
+      ERREXIT(cinfo, JERR_BAD_J_COLORSPACE);
+    if (cinfo->in_color_space == JCS_GRAYSCALE)
+      cconvert->pub.color_convert = grayscale_convert;
+    else if (cinfo->in_color_space == JCS_RGB ||
+             cinfo->in_color_space == JCS_EXT_RGB ||
+             cinfo->in_color_space == JCS_EXT_RGBX ||
+             cinfo->in_color_space == JCS_EXT_BGR ||
+             cinfo->in_color_space == JCS_EXT_BGRX ||
+             cinfo->in_color_space == JCS_EXT_XBGR ||
+             cinfo->in_color_space == JCS_EXT_XRGB ||
+             cinfo->in_color_space == JCS_EXT_RGBA ||
+             cinfo->in_color_space == JCS_EXT_BGRA ||
+             cinfo->in_color_space == JCS_EXT_ABGR ||
+             cinfo->in_color_space == JCS_EXT_ARGB) {
+      if (jsimd_can_rgb_gray())
+        cconvert->pub.color_convert = jsimd_rgb_gray_convert;
+      else {
+        cconvert->pub.start_pass = rgb_ycc_start;
+        cconvert->pub.color_convert = rgb_gray_convert;
+      }
+    } else if (cinfo->in_color_space == JCS_YCbCr)
+      cconvert->pub.color_convert = grayscale_convert;
+    else
+      ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL);
+    break;
+
+  case JCS_RGB:
+  case JCS_EXT_RGB:
+  case JCS_EXT_RGBX:
+  case JCS_EXT_BGR:
+  case JCS_EXT_BGRX:
+  case JCS_EXT_XBGR:
+  case JCS_EXT_XRGB:
+  case JCS_EXT_RGBA:
+  case JCS_EXT_BGRA:
+  case JCS_EXT_ABGR:
+  case JCS_EXT_ARGB:
+    if (cinfo->num_components != 3)
+      ERREXIT(cinfo, JERR_BAD_J_COLORSPACE);
+    if (cinfo->in_color_space == cinfo->jpeg_color_space &&
+      rgb_pixelsize[cinfo->in_color_space] == 3)
+      cconvert->pub.color_convert = null_convert;
+    else
+      ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL);
+    break;
+
+  case JCS_YCbCr:
+    if (cinfo->num_components != 3)
+      ERREXIT(cinfo, JERR_BAD_J_COLORSPACE);
+    if (cinfo->in_color_space == JCS_RGB ||
+        cinfo->in_color_space == JCS_EXT_RGB ||
+        cinfo->in_color_space == JCS_EXT_RGBX ||
+        cinfo->in_color_space == JCS_EXT_BGR ||
+        cinfo->in_color_space == JCS_EXT_BGRX ||
+        cinfo->in_color_space == JCS_EXT_XBGR ||
+        cinfo->in_color_space == JCS_EXT_XRGB ||
+        cinfo->in_color_space == JCS_EXT_RGBA ||
+        cinfo->in_color_space == JCS_EXT_BGRA ||
+        cinfo->in_color_space == JCS_EXT_ABGR ||
+        cinfo->in_color_space == JCS_EXT_ARGB) {
+      if (jsimd_can_rgb_ycc())
+        cconvert->pub.color_convert = jsimd_rgb_ycc_convert;
+      else {
+        cconvert->pub.start_pass = rgb_ycc_start;
+        cconvert->pub.color_convert = rgb_ycc_convert;
+      }
+    } else if (cinfo->in_color_space == JCS_YCbCr)
+      cconvert->pub.color_convert = null_convert;
+    else
+      ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL);
+    break;
+
+  case JCS_CMYK:
+    if (cinfo->num_components != 4)
+      ERREXIT(cinfo, JERR_BAD_J_COLORSPACE);
+    if (cinfo->in_color_space == JCS_CMYK)
+      cconvert->pub.color_convert = null_convert;
+    else
+      ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL);
+    break;
+
+  case JCS_YCCK:
+    if (cinfo->num_components != 4)
+      ERREXIT(cinfo, JERR_BAD_J_COLORSPACE);
+    if (cinfo->in_color_space == JCS_CMYK) {
+      cconvert->pub.start_pass = rgb_ycc_start;
+      cconvert->pub.color_convert = cmyk_ycck_convert;
+    } else if (cinfo->in_color_space == JCS_YCCK)
+      cconvert->pub.color_convert = null_convert;
+    else
+      ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL);
+    break;
+
+  default:			/* allow null conversion of JCS_UNKNOWN */
+    if (cinfo->jpeg_color_space != cinfo->in_color_space ||
+	cinfo->num_components != cinfo->input_components)
+      ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL);
+    cconvert->pub.color_convert = null_convert;
+    break;
+  }
+}

=== added file 'src/libjpeg-turbo/jcdctmgr.c'
--- src/libjpeg-turbo/jcdctmgr.c	1970-01-01 00:00:00 +0000
+++ src/libjpeg-turbo/jcdctmgr.c	2012-06-27 16:20:24 +0000
@@ -0,0 +1,642 @@
+/*
+ * jcdctmgr.c
+ *
+ * Copyright (C) 1994-1996, Thomas G. Lane.
+ * Copyright (C) 1999-2006, MIYASAKA Masaru.
+ * Copyright 2009 Pierre Ossman <[e-mail]> for Cendio AB
+ * Copyright (C) 2011 D. R. Commander
+ * This file is part of the Independent JPEG Group's software.
+ * For conditions of distribution and use, see the accompanying README file.
+ *
+ * This file contains the forward-DCT management logic.
+ * This code selects a particular DCT implementation to be used,
+ * and it performs related housekeeping chores including coefficient
+ * quantization.
+ */
+
+#define JPEG_INTERNALS
+#include "jinclude.h"
+#include "jpeglib.h"
+#include "jdct.h"		/* Private declarations for DCT subsystem */
+#include "jsimddct.h"
+
+
+/* Private subobject for this module */
+
+typedef JMETHOD(void, forward_DCT_method_ptr, (DCTELEM * data));
+typedef JMETHOD(void, float_DCT_method_ptr, (FAST_FLOAT * data));
+
+typedef JMETHOD(void, convsamp_method_ptr,
+                (JSAMPARRAY sample_data, JDIMENSION start_col,
+                 DCTELEM * workspace));
+typedef JMETHOD(void, float_convsamp_method_ptr,
+                (JSAMPARRAY sample_data, JDIMENSION start_col,
+                 FAST_FLOAT *workspace));
+
+typedef JMETHOD(void, quantize_method_ptr,
+                (JCOEFPTR coef_block, DCTELEM * divisors,
+                 DCTELEM * workspace));
+typedef JMETHOD(void, float_quantize_method_ptr,
+                (JCOEFPTR coef_block, FAST_FLOAT * divisors,
+                 FAST_FLOAT * workspace));
+
+METHODDEF(void) quantize (JCOEFPTR, DCTELEM *, DCTELEM *);
+
+typedef struct {
+  struct jpeg_forward_dct pub;	/* public fields */
+
+  /* Pointer to the DCT routine actually in use */
+  forward_DCT_method_ptr dct;
+  convsamp_method_ptr convsamp;
+  quantize_method_ptr quantize;
+
+  /* The actual post-DCT divisors --- not identical to the quant table
+   * entries, because of scaling (especially for an unnormalized DCT).
+   * Each table is given in normal array order.
+   */
+  DCTELEM * divisors[NUM_QUANT_TBLS];
+
+  /* work area for FDCT subroutine */
+  DCTELEM * workspace;
+
+#ifdef DCT_FLOAT_SUPPORTED
+  /* Same as above for the floating-point case. */
+  float_DCT_method_ptr float_dct;
+  float_convsamp_method_ptr float_convsamp;
+  float_quantize_method_ptr float_quantize;
+  FAST_FLOAT * float_divisors[NUM_QUANT_TBLS];
+  FAST_FLOAT * float_workspace;
+#endif
+} my_fdct_controller;
+
+typedef my_fdct_controller * my_fdct_ptr;
+
+
+/*
+ * Find the highest bit in an integer through binary search.
+ */
+LOCAL(int)
+flss (UINT16 val)
+{
+  int bit;
+
+  bit = 16;
+
+  if (!val)
+    return 0;
+
+  if (!(val & 0xff00)) {
+    bit -= 8;
+    val <<= 8;
+  }
+  if (!(val & 0xf000)) {
+    bit -= 4;
+    val <<= 4;
+  }
+  if (!(val & 0xc000)) {
+    bit -= 2;
+    val <<= 2;
+  }
+  if (!(val & 0x8000)) {
+    bit -= 1;
+    val <<= 1;
+  }
+
+  return bit;
+}
+
+/*
+ * Compute values to do a division using reciprocal.
+ *
+ * This implementation is based on an algorithm described in
+ *   "How to optimize for the Pentium family of microprocessors"
+ *   (http://www.agner.org/assem/).
+ * More information about the basic algorithm can be found in
+ * the paper "Integer Division Using Reciprocals" by Robert Alverson.
+ *
+ * The basic idea is to replace x/d by x * d^-1. In order to store
+ * d^-1 with enough precision we shift it left a few places. It turns
+ * out that this algoright gives just enough precision, and also fits
+ * into DCTELEM:
+ *
+ *   b = (the number of significant bits in divisor) - 1
+ *   r = (word size) + b
+ *   f = 2^r / divisor
+ *
+ * f will not be an integer for most cases, so we need to compensate
+ * for the rounding error introduced:
+ *
+ *   no fractional part:
+ *
+ *       result = input >> r
+ *
+ *   fractional part of f < 0.5:
+ *
+ *       round f down to nearest integer
+ *       result = ((input + 1) * f) >> r
+ *
+ *   fractional part of f > 0.5:
+ *
+ *       round f up to nearest integer
+ *       result = (input * f) >> r
+ *
+ * This is the original algorithm that gives truncated results. But we
+ * want properly rounded results, so we replace "input" with
+ * "input + divisor/2".
+ *
+ * In order to allow SIMD implementations we also tweak the values to
+ * allow the same calculation to be made at all times:
+ * 
+ *   dctbl[0] = f rounded to nearest integer
+ *   dctbl[1] = divisor / 2 (+ 1 if fractional part of f < 0.5)
+ *   dctbl[2] = 1 << ((word size) * 2 - r)
+ *   dctbl[3] = r - (word size)
+ *
+ * dctbl[2] is for stupid instruction sets where the shift operation
+ * isn't member wise (e.g. MMX).
+ *
+ * The reason dctbl[2] and dctbl[3] reduce the shift with (word size)
+ * is that most SIMD implementations have a "multiply and store top
+ * half" operation.
+ *
+ * Lastly, we store each of the values in their own table instead
+ * of in a consecutive manner, yet again in order to allow SIMD
+ * routines.
+ */
+LOCAL(int)
+compute_reciprocal (UINT16 divisor, DCTELEM * dtbl)
+{
+  UDCTELEM2 fq, fr;
+  UDCTELEM c;
+  int b, r;
+
+  b = flss(divisor) - 1;
+  r  = sizeof(DCTELEM) * 8 + b;
+
+  fq = ((UDCTELEM2)1 << r) / divisor;
+  fr = ((UDCTELEM2)1 << r) % divisor;
+
+  c = divisor / 2; /* for rounding */
+
+  if (fr == 0) { /* divisor is power of two */
+    /* fq will be one bit too large to fit in DCTELEM, so adjust */
+    fq >>= 1;
+    r--;
+  } else if (fr <= (divisor / 2U)) { /* fractional part is < 0.5 */
+    c++;
+  } else { /* fractional part is > 0.5 */
+    fq++;
+  }
+
+  dtbl[DCTSIZE2 * 0] = (DCTELEM) fq;      /* reciprocal */
+  dtbl[DCTSIZE2 * 1] = (DCTELEM) c;       /* correction + roundfactor */
+  dtbl[DCTSIZE2 * 2] = (DCTELEM) (1 << (sizeof(DCTELEM)*8*2 - r));  /* scale */
+  dtbl[DCTSIZE2 * 3] = (DCTELEM) r - sizeof(DCTELEM)*8; /* shift */
+
+  if(r <= 16) return 0;
+  else return 1;
+}
+
+/*
+ * Initialize for a processing pass.
+ * Verify that all referenced Q-tables are present, and set up
+ * the divisor table for each one.
+ * In the current implementation, DCT of all components is done during
+ * the first pass, even if only some components will be output in the
+ * first scan.  Hence all components should be examined here.
+ */
+
+METHODDEF(void)
+start_pass_fdctmgr (j_compress_ptr cinfo)
+{
+  my_fdct_ptr fdct = (my_fdct_ptr) cinfo->fdct;
+  int ci, qtblno, i;
+  jpeg_component_info *compptr;
+  JQUANT_TBL * qtbl;
+  DCTELEM * dtbl;
+
+  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
+       ci++, compptr++) {
+    qtblno = compptr->quant_tbl_no;
+    /* Make sure specified quantization table is present */
+    if (qtblno < 0 || qtblno >= NUM_QUANT_TBLS ||
+	cinfo->quant_tbl_ptrs[qtblno] == NULL)
+      ERREXIT1(cinfo, JERR_NO_QUANT_TABLE, qtblno);
+    qtbl = cinfo->quant_tbl_ptrs[qtblno];
+    /* Compute divisors for this quant table */
+    /* We may do this more than once for same table, but it's not a big deal */
+    switch (cinfo->dct_method) {
+#ifdef DCT_ISLOW_SUPPORTED
+    case JDCT_ISLOW:
+      /* For LL&M IDCT method, divisors are equal to raw quantization
+       * coefficients multiplied by 8 (to counteract scaling).
+       */
+      if (fdct->divisors[qtblno] == NULL) {
+	fdct->divisors[qtblno] = (DCTELEM *)
+	  (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
+				      (DCTSIZE2 * 4) * SIZEOF(DCTELEM));
+      }
+      dtbl = fdct->divisors[qtblno];
+      for (i = 0; i < DCTSIZE2; i++) {
+	if(!compute_reciprocal(qtbl->quantval[i] << 3, &dtbl[i])
+	  && fdct->quantize == jsimd_quantize)
+	  fdct->quantize = quantize;
+      }
+      break;
+#endif
+#ifdef DCT_IFAST_SUPPORTED
+    case JDCT_IFAST:
+      {
+	/* For AA&N IDCT method, divisors are equal to quantization
+	 * coefficients scaled by scalefactor[row]*scalefactor[col], where
+	 *   scalefactor[0] = 1
+	 *   scalefactor[k] = cos(k*PI/16) * sqrt(2)    for k=1..7
+	 * We apply a further scale factor of 8.
+	 */
+#define CONST_BITS 14
+	static const INT16 aanscales[DCTSIZE2] = {
+	  /* precomputed values scaled up by 14 bits */
+	  16384, 22725, 21407, 19266, 16384, 12873,  8867,  4520,
+	  22725, 31521, 29692, 26722, 22725, 17855, 12299,  6270,
+	  21407, 29692, 27969, 25172, 21407, 16819, 11585,  5906,
+	  19266, 26722, 25172, 22654, 19266, 15137, 10426,  5315,
+	  16384, 22725, 21407, 19266, 16384, 12873,  8867,  4520,
+	  12873, 17855, 16819, 15137, 12873, 10114,  6967,  3552,
+	   8867, 12299, 11585, 10426,  8867,  6967,  4799,  2446,
+	   4520,  6270,  5906,  5315,  4520,  3552,  2446,  1247
+	};
+	SHIFT_TEMPS
+
+	if (fdct->divisors[qtblno] == NULL) {
+	  fdct->divisors[qtblno] = (DCTELEM *)
+	    (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
+					(DCTSIZE2 * 4) * SIZEOF(DCTELEM));
+	}
+	dtbl = fdct->divisors[qtblno];
+	for (i = 0; i < DCTSIZE2; i++) {
+	  if(!compute_reciprocal(
+	    DESCALE(MULTIPLY16V16((INT32) qtbl->quantval[i],
+				  (INT32) aanscales[i]),
+		    CONST_BITS-3), &dtbl[i])
+	    && fdct->quantize == jsimd_quantize)
+	    fdct->quantize = quantize;
+	}
+      }
+      break;
+#endif
+#ifdef DCT_FLOAT_SUPPORTED
+    case JDCT_FLOAT:
+      {
+	/* For float AA&N IDCT method, divisors are equal to quantization
+	 * coefficients scaled by scalefactor[row]*scalefactor[col], where
+	 *   scalefactor[0] = 1
+	 *   scalefactor[k] = cos(k*PI/16) * sqrt(2)    for k=1..7
+	 * We apply a further scale factor of 8.
+	 * What's actually stored is 1/divisor so that the inner loop can
+	 * use a multiplication rather than a division.
+	 */
+	FAST_FLOAT * fdtbl;
+	int row, col;
+	static const double aanscalefactor[DCTSIZE] = {
+	  1.0, 1.387039845, 1.306562965, 1.175875602,
+	  1.0, 0.785694958, 0.541196100, 0.275899379
+	};
+
+	if (fdct->float_divisors[qtblno] == NULL) {
+	  fdct->float_divisors[qtblno] = (FAST_FLOAT *)
+	    (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
+					DCTSIZE2 * SIZEOF(FAST_FLOAT));
+	}
+	fdtbl = fdct->float_divisors[qtblno];
+	i = 0;
+	for (row = 0; row < DCTSIZE; row++) {
+	  for (col = 0; col < DCTSIZE; col++) {
+	    fdtbl[i] = (FAST_FLOAT)
+	      (1.0 / (((double) qtbl->quantval[i] *
+		       aanscalefactor[row] * aanscalefactor[col] * 8.0)));
+	    i++;
+	  }
+	}
+      }
+      break;
+#endif
+    default:
+      ERREXIT(cinfo, JERR_NOT_COMPILED);
+      break;
+    }
+  }
+}
+
+
+/*
+ * Load data into workspace, applying unsigned->signed conversion.
+ */
+
+METHODDEF(void)
+convsamp (JSAMPARRAY sample_data, JDIMENSION start_col, DCTELEM * workspace)
+{
+  register DCTELEM *workspaceptr;
+  register JSAMPROW elemptr;
+  register int elemr;
+
+  workspaceptr = workspace;
+  for (elemr = 0; elemr < DCTSIZE; elemr++) {
+    elemptr = sample_data[elemr] + start_col;
+
+#if DCTSIZE == 8		/* unroll the inner loop */
+    *workspaceptr++ = GETJSAMPLE(*elemptr++) - CENTERJSAMPLE;
+    *workspaceptr++ = GETJSAMPLE(*elemptr++) - CENTERJSAMPLE;
+    *workspaceptr++ = GETJSAMPLE(*elemptr++) - CENTERJSAMPLE;
+    *workspaceptr++ = GETJSAMPLE(*elemptr++) - CENTERJSAMPLE;
+    *workspaceptr++ = GETJSAMPLE(*elemptr++) - CENTERJSAMPLE;
+    *workspaceptr++ = GETJSAMPLE(*elemptr++) - CENTERJSAMPLE;
+    *workspaceptr++ = GETJSAMPLE(*elemptr++) - CENTERJSAMPLE;
+    *workspaceptr++ = GETJSAMPLE(*elemptr++) - CENTERJSAMPLE;
+#else
+    {
+      register int elemc;
+      for (elemc = DCTSIZE; elemc > 0; elemc--)
+        *workspaceptr++ = GETJSAMPLE(*elemptr++) - CENTERJSAMPLE;
+    }
+#endif
+  }
+}
+
+
+/*
+ * Quantize/descale the coefficients, and store into coef_blocks[].
+ */
+
+METHODDEF(void)
+quantize (JCOEFPTR coef_block, DCTELEM * divisors, DCTELEM * workspace)
+{
+  int i;
+  DCTELEM temp;
+  UDCTELEM recip, corr, shift;
+  UDCTELEM2 product;
+  JCOEFPTR output_ptr = coef_block;
+
+  for (i = 0; i < DCTSIZE2; i++) {
+    temp = workspace[i];
+    recip = divisors[i + DCTSIZE2 * 0];
+    corr =  divisors[i + DCTSIZE2 * 1];
+    shift = divisors[i + DCTSIZE2 * 3];
+
+    if (temp < 0) {
+      temp = -temp;
+      product = (UDCTELEM2)(temp + corr) * recip;
+      product >>= shift + sizeof(DCTELEM)*8;
+      temp = product;
+      temp = -temp;
+    } else {
+      product = (UDCTELEM2)(temp + corr) * recip;
+      product >>= shift + sizeof(DCTELEM)*8;
+      temp = product;
+    }
+
+    output_ptr[i] = (JCOEF) temp;
+  }
+}
+
+
+/*
+ * Perform forward DCT on one or more blocks of a component.
+ *
+ * The input samples are taken from the sample_data[] array starting at
+ * position start_row/start_col, and moving to the right for any additional
+ * blocks. The quantized coefficients are returned in coef_blocks[].
+ */
+
+METHODDEF(void)
+forward_DCT (j_compress_ptr cinfo, jpeg_component_info * compptr,
+	     JSAMPARRAY sample_data, JBLOCKROW coef_blocks,
+	     JDIMENSION start_row, JDIMENSION start_col,
+	     JDIMENSION num_blocks)
+/* This version is used for integer DCT implementations. */
+{
+  /* This routine is heavily used, so it's worth coding it tightly. */
+  my_fdct_ptr fdct = (my_fdct_ptr) cinfo->fdct;
+  DCTELEM * divisors = fdct->divisors[compptr->quant_tbl_no];
+  DCTELEM * workspace;
+  JDIMENSION bi;
+
+  /* Make sure the compiler doesn't look up these every pass */
+  forward_DCT_method_ptr do_dct = fdct->dct;
+  convsamp_method_ptr do_convsamp = fdct->convsamp;
+  quantize_method_ptr do_quantize = fdct->quantize;
+  workspace = fdct->workspace;
+
+  sample_data += start_row;	/* fold in the vertical offset once */
+
+  for (bi = 0; bi < num_blocks; bi++, start_col += DCTSIZE) {
+    /* Load data into workspace, applying unsigned->signed conversion */
+    (*do_convsamp) (sample_data, start_col, workspace);
+
+    /* Perform the DCT */
+    (*do_dct) (workspace);
+
+    /* Quantize/descale the coefficients, and store into coef_blocks[] */
+    (*do_quantize) (coef_blocks[bi], divisors, workspace);
+  }
+}
+
+
+#ifdef DCT_FLOAT_SUPPORTED
+
+
+METHODDEF(void)
+convsamp_float (JSAMPARRAY sample_data, JDIMENSION start_col, FAST_FLOAT * workspace)
+{
+  register FAST_FLOAT *workspaceptr;
+  register JSAMPROW elemptr;
+  register int elemr;
+
+  workspaceptr = workspace;
+  for (elemr = 0; elemr < DCTSIZE; elemr++) {
+    elemptr = sample_data[elemr] + start_col;
+#if DCTSIZE == 8		/* unroll the inner loop */
+    *workspaceptr++ = (FAST_FLOAT)(GETJSAMPLE(*elemptr++) - CENTERJSAMPLE);
+    *workspaceptr++ = (FAST_FLOAT)(GETJSAMPLE(*elemptr++) - CENTERJSAMPLE);
+    *workspaceptr++ = (FAST_FLOAT)(GETJSAMPLE(*elemptr++) - CENTERJSAMPLE);
+    *workspaceptr++ = (FAST_FLOAT)(GETJSAMPLE(*elemptr++) - CENTERJSAMPLE);
+    *workspaceptr++ = (FAST_FLOAT)(GETJSAMPLE(*elemptr++) - CENTERJSAMPLE);
+    *workspaceptr++ = (FAST_FLOAT)(GETJSAMPLE(*elemptr++) - CENTERJSAMPLE);
+    *workspaceptr++ = (FAST_FLOAT)(GETJSAMPLE(*elemptr++) - CENTERJSAMPLE);
+    *workspaceptr++ = (FAST_FLOAT)(GETJSAMPLE(*elemptr++) - CENTERJSAMPLE);
+#else
+    {
+      register int elemc;
+      for (elemc = DCTSIZE; elemc > 0; elemc--)
+        *workspaceptr++ = (FAST_FLOAT)
+                          (GETJSAMPLE(*elemptr++) - CENTERJSAMPLE);
+    }
+#endif
+  }
+}
+
+
+METHODDEF(void)
+quantize_float (JCOEFPTR coef_block, FAST_FLOAT * divisors, FAST_FLOAT * workspace)
+{
+  register FAST_FLOAT temp;
+  register int i;
+  register JCOEFPTR output_ptr = coef_block;
+
+  for (i = 0; i < DCTSIZE2; i++) {
+    /* Apply the quantization and scaling factor */
+    temp = workspace[i] * divisors[i];
+
+    /* Round to nearest integer.
+     * Since C does not specify the direction of rounding for negative
+     * quotients, we have to force the dividend positive for portability.
+     * The maximum coefficient size is +-16K (for 12-bit data), so this
+     * code should work for either 16-bit or 32-bit ints.
+     */
+    output_ptr[i] = (JCOEF) ((int) (temp + (FAST_FLOAT) 16384.5) - 16384);
+  }
+}
+
+
+METHODDEF(void)
+forward_DCT_float (j_compress_ptr cinfo, jpeg_component_info * compptr,
+		   JSAMPARRAY sample_data, JBLOCKROW coef_blocks,
+		   JDIMENSION start_row, JDIMENSION start_col,
+		   JDIMENSION num_blocks)
+/* This version is used for floating-point DCT implementations. */
+{
+  /* This routine is heavily used, so it's worth coding it tightly. */
+  my_fdct_ptr fdct = (my_fdct_ptr) cinfo->fdct;
+  FAST_FLOAT * divisors = fdct->float_divisors[compptr->quant_tbl_no];
+  FAST_FLOAT * workspace;
+  JDIMENSION bi;
+
+
+  /* Make sure the compiler doesn't look up these every pass */
+  float_DCT_method_ptr do_dct = fdct->float_dct;
+  float_convsamp_method_ptr do_convsamp = fdct->float_convsamp;
+  float_quantize_method_ptr do_quantize = fdct->float_quantize;
+  workspace = fdct->float_workspace;
+
+  sample_data += start_row;	/* fold in the vertical offset once */
+
+  for (bi = 0; bi < num_blocks; bi++, start_col += DCTSIZE) {
+    /* Load data into workspace, applying unsigned->signed conversion */
+    (*do_convsamp) (sample_data, start_col, workspace);
+
+    /* Perform the DCT */
+    (*do_dct) (workspace);
+
+    /* Quantize/descale the coefficients, and store into coef_blocks[] */
+    (*do_quantize) (coef_blocks[bi], divisors, workspace);
+  }
+}
+
+#endif /* DCT_FLOAT_SUPPORTED */
+
+
+/*
+ * Initialize FDCT manager.
+ */
+
+GLOBAL(void)
+jinit_forward_dct (j_compress_ptr cinfo)
+{
+  my_fdct_ptr fdct;
+  int i;
+
+  fdct = (my_fdct_ptr)
+    (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
+				SIZEOF(my_fdct_controller));
+  cinfo->fdct = (struct jpeg_forward_dct *) fdct;
+  fdct->pub.start_pass = start_pass_fdctmgr;
+
+  /* First determine the DCT... */
+  switch (cinfo->dct_method) {
+#ifdef DCT_ISLOW_SUPPORTED
+  case JDCT_ISLOW:
+    fdct->pub.forward_DCT = forward_DCT;
+    if (jsimd_can_fdct_islow())
+      fdct->dct = jsimd_fdct_islow;
+    else
+      fdct->dct = jpeg_fdct_islow;
+    break;
+#endif
+#ifdef DCT_IFAST_SUPPORTED
+  case JDCT_IFAST:
+    fdct->pub.forward_DCT = forward_DCT;
+    if (jsimd_can_fdct_ifast())
+      fdct->dct = jsimd_fdct_ifast;
+    else
+      fdct->dct = jpeg_fdct_ifast;
+    break;
+#endif
+#ifdef DCT_FLOAT_SUPPORTED
+  case JDCT_FLOAT:
+    fdct->pub.forward_DCT = forward_DCT_float;
+    if (jsimd_can_fdct_float())
+      fdct->float_dct = jsimd_fdct_float;
+    else
+      fdct->float_dct = jpeg_fdct_float;
+    break;
+#endif
+  default:
+    ERREXIT(cinfo, JERR_NOT_COMPILED);
+    break;
+  }
+
+  /* ...then the supporting stages. */
+  switch (cinfo->dct_method) {
+#ifdef DCT_ISLOW_SUPPORTED
+  case JDCT_ISLOW:
+#endif
+#ifdef DCT_IFAST_SUPPORTED
+  case JDCT_IFAST:
+#endif
+#if defined(DCT_ISLOW_SUPPORTED) || defined(DCT_IFAST_SUPPORTED)
+    if (jsimd_can_convsamp())
+      fdct->convsamp = jsimd_convsamp;
+    else
+      fdct->convsamp = convsamp;
+    if (jsimd_can_quantize())
+      fdct->quantize = jsimd_quantize;
+    else
+      fdct->quantize = quantize;
+    break;
+#endif
+#ifdef DCT_FLOAT_SUPPORTED
+  case JDCT_FLOAT:
+    if (jsimd_can_convsamp_float())
+      fdct->float_convsamp = jsimd_convsamp_float;
+    else
+      fdct->float_convsamp = convsamp_float;
+    if (jsimd_can_quantize_float())
+      fdct->float_quantize = jsimd_quantize_float;
+    else
+      fdct->float_quantize = quantize_float;
+    break;
+#endif
+  default:
+    ERREXIT(cinfo, JERR_NOT_COMPILED);
+    break;
+  }
+
+  /* Allocate workspace memory */
+#ifdef DCT_FLOAT_SUPPORTED
+  if (cinfo->dct_method == JDCT_FLOAT)
+    fdct->float_workspace = (FAST_FLOAT *)
+      (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
+				  SIZEOF(FAST_FLOAT) * DCTSIZE2);
+  else
+#endif
+    fdct->workspace = (DCTELEM *)
+      (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
+				  SIZEOF(DCTELEM) * DCTSIZE2);
+
+  /* Mark divisor tables unallocated */
+  for (i = 0; i < NUM_QUANT_TBLS; i++) {
+    fdct->divisors[i] = NULL;
+#ifdef DCT_FLOAT_SUPPORTED
+    fdct->float_divisors[i] = NULL;
+#endif
+  }
+}

=== added file 'src/libjpeg-turbo/jchuff.c'
--- src/libjpeg-turbo/jchuff.c	1970-01-01 00:00:00 +0000
+++ src/libjpeg-turbo/jchuff.c	2012-06-27 16:20:24 +0000
@@ -0,0 +1,1026 @@
+/*
+ * jchuff.c
+ *
+ * Copyright (C) 1991-1997, Thomas G. Lane.
+ * Copyright (C) 2009-2011, D. R. Commander.
+ * This file is part of the Independent JPEG Group's software.
+ * For conditions of distribution and use, see the accompanying README file.
+ *
+ * This file contains Huffman entropy encoding routines.
+ *
+ * Much of the complexity here has to do with supporting output suspension.
+ * If the data destination module demands suspension, we want to be able to
+ * back up to the start of the current MCU.  To do this, we copy state
+ * variables into local working storage, and update them back to the
+ * permanent JPEG objects only upon successful completion of an MCU.
+ */
+
+#define JPEG_INTERNALS
+#include "jinclude.h"
+#include "jpeglib.h"
+#include "jchuff.h"		/* Declarations shared with jcphuff.c */
+#include <limits.h>
+
+static unsigned char jpeg_nbits_table[65536];
+static int jpeg_nbits_table_init = 0;
+
+#ifndef min
+ #define min(a,b) ((a)<(b)?(a):(b))
+#endif
+
+
+/* Expanded entropy encoder object for Huffman encoding.
+ *
+ * The savable_state subrecord contains fields that change within an MCU,
+ * but must not be updated permanently until we complete the MCU.
+ */
+
+typedef struct {
+  size_t put_buffer;		/* current bit-accumulation buffer */
+  int put_bits;			/* # of bits now in it */
+  int last_dc_val[MAX_COMPS_IN_SCAN]; /* last DC coef for each component */
+} savable_state;
+
+/* This macro is to work around compilers with missing or broken
+ * structure assignment.  You'll need to fix this code if you have
+ * such a compiler and you change MAX_COMPS_IN_SCAN.
+ */
+
+#ifndef NO_STRUCT_ASSIGN
+#define ASSIGN_STATE(dest,src)  ((dest) = (src))
+#else
+#if MAX_COMPS_IN_SCAN == 4
+#define ASSIGN_STATE(dest,src)  \
+	((dest).put_buffer = (src).put_buffer, \
+	 (dest).put_bits = (src).put_bits, \
+	 (dest).last_dc_val[0] = (src).last_dc_val[0], \
+	 (dest).last_dc_val[1] = (src).last_dc_val[1], \
+	 (dest).last_dc_val[2] = (src).last_dc_val[2], \
+	 (dest).last_dc_val[3] = (src).last_dc_val[3])
+#endif
+#endif
+
+
+typedef struct {
+  struct jpeg_entropy_encoder pub; /* public fields */
+
+  savable_state saved;		/* Bit buffer & DC state at start of MCU */
+
+  /* These fields are NOT loaded into local working state. */
+  unsigned int restarts_to_go;	/* MCUs left in this restart interval */
+  int next_restart_num;		/* next restart number to write (0-7) */
+
+  /* Pointers to derived tables (these workspaces have image lifespan) */
+  c_derived_tbl * dc_derived_tbls[NUM_HUFF_TBLS];
+  c_derived_tbl * ac_derived_tbls[NUM_HUFF_TBLS];
+
+#ifdef ENTROPY_OPT_SUPPORTED	/* Statistics tables for optimization */
+  long * dc_count_ptrs[NUM_HUFF_TBLS];
+  long * ac_count_ptrs[NUM_HUFF_TBLS];
+#endif
+} huff_entropy_encoder;
+
+typedef huff_entropy_encoder * huff_entropy_ptr;
+
+/* Working state while writing an MCU.
+ * This struct contains all the fields that are needed by subroutines.
+ */
+
+typedef struct {
+  JOCTET * next_output_byte;	/* => next byte to write in buffer */
+  size_t free_in_buffer;	/* # of byte spaces remaining in buffer */
+  savable_state cur;		/* Current bit buffer & DC state */
+  j_compress_ptr cinfo;		/* dump_buffer needs access to this */
+} working_state;
+
+
+/* Forward declarations */
+METHODDEF(boolean) encode_mcu_huff JPP((j_compress_ptr cinfo,
+					JBLOCKROW *MCU_data));
+METHODDEF(void) finish_pass_huff JPP((j_compress_ptr cinfo));
+#ifdef ENTROPY_OPT_SUPPORTED
+METHODDEF(boolean) encode_mcu_gather JPP((j_compress_ptr cinfo,
+					  JBLOCKROW *MCU_data));
+METHODDEF(void) finish_pass_gather JPP((j_compress_ptr cinfo));
+#endif
+
+
+/*
+ * Initialize for a Huffman-compressed scan.
+ * If gather_statistics is TRUE, we do not output anything during the scan,
+ * just count the Huffman symbols used and generate Huffman code tables.
+ */
+
+METHODDEF(void)
+start_pass_huff (j_compress_ptr cinfo, boolean gather_statistics)
+{
+  huff_entropy_ptr entropy = (huff_entropy_ptr) cinfo->entropy;
+  int ci, dctbl, actbl;
+  jpeg_component_info * compptr;
+
+  if (gather_statistics) {
+#ifdef ENTROPY_OPT_SUPPORTED
+    entropy->pub.encode_mcu = encode_mcu_gather;
+    entropy->pub.finish_pass = finish_pass_gather;
+#else
+    ERREXIT(cinfo, JERR_NOT_COMPILED);
+#endif
+  } else {
+    entropy->pub.encode_mcu = encode_mcu_huff;
+    entropy->pub.finish_pass = finish_pass_huff;
+  }
+
+  for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
+    compptr = cinfo->cur_comp_info[ci];
+    dctbl = compptr->dc_tbl_no;
+    actbl = compptr->ac_tbl_no;
+    if (gather_statistics) {
+#ifdef ENTROPY_OPT_SUPPORTED
+      /* Check for invalid table indexes */
+      /* (make_c_derived_tbl does this in the other path) */
+      if (dctbl < 0 || dctbl >= NUM_HUFF_TBLS)
+	ERREXIT1(cinfo, JERR_NO_HUFF_TABLE, dctbl);
+      if (actbl < 0 || actbl >= NUM_HUFF_TBLS)
+	ERREXIT1(cinfo, JERR_NO_HUFF_TABLE, actbl);
+      /* Allocate and zero the statistics tables */
+      /* Note that jpeg_gen_optimal_table expects 257 entries in each table! */
+      if (entropy->dc_count_ptrs[dctbl] == NULL)
+	entropy->dc_count_ptrs[dctbl] = (long *)
+	  (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
+				      257 * SIZEOF(long));
+      MEMZERO(entropy->dc_count_ptrs[dctbl], 257 * SIZEOF(long));
+      if (entropy->ac_count_ptrs[actbl] == NULL)
+	entropy->ac_count_ptrs[actbl] = (long *)
+	  (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
+				      257 * SIZEOF(long));
+      MEMZERO(entropy->ac_count_ptrs[actbl], 257 * SIZEOF(long));
+#endif
+    } else {
+      /* Compute derived values for Huffman tables */
+      /* We may do this more than once for a table, but it's not expensive */
+      jpeg_make_c_derived_tbl(cinfo, TRUE, dctbl,
+			      & entropy->dc_derived_tbls[dctbl]);
+      jpeg_make_c_derived_tbl(cinfo, FALSE, actbl,
+			      & entropy->ac_derived_tbls[actbl]);
+    }
+    /* Initialize DC predictions to 0 */
+    entropy->saved.last_dc_val[ci] = 0;
+  }
+
+  /* Initialize bit buffer to empty */
+  entropy->saved.put_buffer = 0;
+  entropy->saved.put_bits = 0;
+
+  /* Initialize restart stuff */
+  entropy->restarts_to_go = cinfo->restart_interval;
+  entropy->next_restart_num = 0;
+}
+
+
+/*
+ * Compute the derived values for a Huffman table.
+ * This routine also performs some validation checks on the table.
+ *
+ * Note this is also used by jcphuff.c.
+ */
+
+GLOBAL(void)
+jpeg_make_c_derived_tbl (j_compress_ptr cinfo, boolean isDC, int tblno,
+			 c_derived_tbl ** pdtbl)
+{
+  JHUFF_TBL *htbl;
+  c_derived_tbl *dtbl;
+  int p, i, l, lastp, si, maxsymbol;
+  char huffsize[257];
+  unsigned int huffcode[257];
+  unsigned int code;
+
+  /* Note that huffsize[] and huffcode[] are filled in code-length order,
+   * paralleling the order of the symbols themselves in htbl->huffval[].
+   */
+
+  /* Find the input Huffman table */
+  if (tblno < 0 || tblno >= NUM_HUFF_TBLS)
+    ERREXIT1(cinfo, JERR_NO_HUFF_TABLE, tblno);
+  htbl =
+    isDC ? cinfo->dc_huff_tbl_ptrs[tblno] : cinfo->ac_huff_tbl_ptrs[tblno];
+  if (htbl == NULL)
+    ERREXIT1(cinfo, JERR_NO_HUFF_TABLE, tblno);
+
+  /* Allocate a workspace if we haven't already done so. */
+  if (*pdtbl == NULL)
+    *pdtbl = (c_derived_tbl *)
+      (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
+				  SIZEOF(c_derived_tbl));
+  dtbl = *pdtbl;
+  
+  /* Figure C.1: make table of Huffman code length for each symbol */
+
+  p = 0;
+  for (l = 1; l <= 16; l++) {
+    i = (int) htbl->bits[l];
+    if (i < 0 || p + i > 256)	/* protect against table overrun */
+      ERREXIT(cinfo, JERR_BAD_HUFF_TABLE);
+    while (i--)
+      huffsize[p++] = (char) l;
+  }
+  huffsize[p] = 0;
+  lastp = p;
+  
+  /* Figure C.2: generate the codes themselves */
+  /* We also validate that the counts represent a legal Huffman code tree. */
+
+  code = 0;
+  si = huffsize[0];
+  p = 0;
+  while (huffsize[p]) {
+    while (((int) huffsize[p]) == si) {
+      huffcode[p++] = code;
+      code++;
+    }
+    /* code is now 1 more than the last code used for codelength si; but
+     * it must still fit in si bits, since no code is allowed to be all ones.
+     */
+    if (((INT32) code) >= (((INT32) 1) << si))
+      ERREXIT(cinfo, JERR_BAD_HUFF_TABLE);
+    code <<= 1;
+    si++;
+  }
+  
+  /* Figure C.3: generate encoding tables */
+  /* These are code and size indexed by symbol value */
+
+  /* Set all codeless symbols to have code length 0;
+   * this lets us detect duplicate VAL entries here, and later
+   * allows emit_bits to detect any attempt to emit such symbols.
+   */
+  MEMZERO(dtbl->ehufsi, SIZEOF(dtbl->ehufsi));
+
+  /* This is also a convenient place to check for out-of-range
+   * and duplicated VAL entries.  We allow 0..255 for AC symbols
+   * but only 0..15 for DC.  (We could constrain them further
+   * based on data depth and mode, but this seems enough.)
+   */
+  maxsymbol = isDC ? 15 : 255;