path: root/indra/llkdu/llimagej2ckdu.cpp

/**
 * @file llimagej2ckdu.cpp
 * @brief This is an implementation of JPEG2000 encode/decode using Kakadu
 *
 * $LicenseInfo:firstyear=2010&license=viewerlgpl$
 * Second Life Viewer Source Code
 * Copyright (C) 2010, Linden Research, Inc.
 *
 * This library is free software; you can redistribute it and/or
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation;
 * version 2.1 of the License only.
 *
 * This library 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
 * Lesser General Public License for more details.
 *
 * You should have received a copy of the GNU Lesser General Public
 * License along with this library; if not, write to the Free Software
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA
 *
 * Linden Research, Inc., 945 Battery Street, San Francisco, CA  94111  USA
 * $/LicenseInfo$
 */

#include "linden_common.h"

#include "llimagej2ckdu.h"

#include "lltimer.h"
#include "llpointer.h"
#include "llmath.h"
#include "llkdumem.h"

#define kdu_xxxx "kdu_block_coding.h"
#include "include_kdu_xxxx.h"

// Avoid ubiquitous necessity of kdu_core:: qualification
using namespace kdu_core;

#include "llexception.h"
#include <boost/exception/diagnostic_information.hpp>
#include <sstream>
#include <iomanip>

// Turns out this must NOT be in the anonymous namespace!
namespace kdu_core
{
// stream kdu_dims to std::ostream
inline
std::ostream& operator<<(std::ostream& out, const kdu_dims& dims)
{
    return out << "(" << dims.pos.x << "," << dims.pos.y << "),"
                  "[" << dims.size.x << "x" << dims.size.y << "]";
}
} // namespace kdu_core

// operator<<(std::ostream&, const kdu_dims&) must precede #include "stringize.h"
#include "stringize.h"

namespace {
// Failure to load an image shouldn't crash the whole viewer.
struct KDUError: public LLContinueError
{
    KDUError(const std::string& msg): LLContinueError(msg) {}
};

// KDU defines int error codes as hex values, so we should log them in hex
// so we can grep KDU headers for the hex. However those hex values
// generally "happen" to encode big-endian multibyte character sequences,
// e.g. KDU_ERROR_EXCEPTION is 0x6b647545: 'kduE'
// But beware because KDU_NULL_EXCEPTION is simply 0 -- which doesn't
// preclude somebody from throwing it.
std::string report_kdu_exception(kdu_exception mb)
{
    std::ostringstream out;
    // always report mb in hex
    out << "kdu_exception " << std::hex << mb;

    // Also display as many chars as are encoded in the kdu_exception
    // value. Make a char array; reserve 1 extra byte for nul terminator.
    char bytes[sizeof(kdu_exception) + 1];
    // Back up through 'bytes'
    char *bptr = bytes + sizeof(bytes);
    *(--bptr) = '\0';
    while (mb)
    {
        // store low-order byte of mb in next-left char
        *(--bptr) = char(mb & 0xFF);
        // then shift mb right by one byte
        mb >>= 8;
    }
    // did that produce any characters?
    if (*bptr)
    {
        out << " (" << bptr << ')';
    }

    return out.str();
}
} // anonymous namespace


class kdc_flow_control {

public:
    kdc_flow_control(kdu_supp::kdu_image_in_base *img_in, kdu_codestream codestream);
    ~kdc_flow_control();
    bool advance_components();
    void process_components();

private:

    struct kdc_component_flow_control {
    public:
        kdu_supp::kdu_image_in_base *reader;
        int vert_subsampling;
        int ratio_counter;  /*  Initialized to 0, decremented by `count_delta';
                                when < 0, a new line must be processed, after
                                which it is incremented by `vert_subsampling'.  */
        int initial_lines;
        int remaining_lines;
        kdu_line_buf *line;
    };

    kdu_codestream codestream;
    kdu_dims valid_tile_indices;
    kdu_coords tile_idx;
    kdu_tile tile;
    int num_components;
    kdc_component_flow_control *components;
    int count_delta; // Holds the minimum of the `vert_subsampling' fields
    kdu_multi_analysis engine;
    kdu_long max_buffer_memory;
};

//
// Kakadu specific implementation
//
void set_default_colour_weights(kdu_params *siz);

// Factory function: see declaration in llimagej2c.cpp
LLImageJ2CImpl* fallbackCreateLLImageJ2CImpl()
{
    return new LLImageJ2CKDU();
}

std::string LLImageJ2CKDU::getEngineInfo() const
{
    return llformat("KDU %s", KDU_CORE_VERSION);
}

class LLKDUDecodeState
{
public:
    LLKDUDecodeState(kdu_tile tile, kdu_byte *buf, S32 row_gap,
                     kdu_codestream* codestreamp);
    ~LLKDUDecodeState();
    bool processTileDecode(F32 decode_time, bool limit_time = true);

private:
    S32 mNumComponents;
    bool mUseYCC;
    kdu_dims mDims;
    kdu_sample_allocator mAllocator;
    kdu_tile_comp mComps[4];
    kdu_line_buf mLines[4];
    kdu_pull_ifc mEngines[4];
    bool mReversible[4]; // Some components may be reversible and others not
    int mBitDepths[4];   // Original bit-depth may be quite different from 8

    kdu_tile mTile;
    kdu_byte *mBuf;
    S32 mRowGap;
};

// Stuff for new kdu error handling
class LLKDUMessage: public kdu_message
{
public:
    LLKDUMessage(const std::string& type):
        mType(type)
    {}

    virtual void put_text(const char *s)
    {
        LL_INFOS() << "KDU " << mType << ": " << s << LL_ENDL;
    }

    virtual void put_text(const kdu_uint16 *s)
    {
        // The previous implementation simply streamed 's' to the log. So
        // either this put_text() override was never called -- or it produced
        // some baffling log messages -- because I assert that streaming a
        // const kdu_uint16* to a std::ostream will display only the hex value
        // of the pointer.
        LL_INFOS() << "KDU " << mType << ": "
                   << utf16str_to_utf8str(llutf16string(s)) << LL_ENDL;
    }

private:
    std::string mType;
};

struct LLKDUMessageWarning : public LLKDUMessage
{
    LLKDUMessageWarning():
        LLKDUMessage("Warning")
    {
        kdu_customize_warnings(this);
    }
};
// Instantiating LLKDUMessageWarning calls kdu_customize_warnings() with the
// new instance. Make it static so this only happens once.
static LLKDUMessageWarning sWarningHandler;

struct LLKDUMessageError : public LLKDUMessage
{
    LLKDUMessageError():
        LLKDUMessage("Error")
    {
        kdu_customize_errors(this);
    }

    virtual void flush(bool end_of_message = false)
    {
        // According to the documentation nat found:
        // http://pirlwww.lpl.arizona.edu/resources/guide/software/Kakadu/html_pages/globals__kdu$mize_errors.html
        // "If a kdu_error object is destroyed, handler→flush will be called with
        // an end_of_message argument equal to true and the process will
        // subsequently be terminated through exit. The termination may be
        // avoided, however, by throwing an exception from within the message
        // terminating handler→flush call."
        // So throwing an exception here isn't arbitrary: we MUST throw an
        // exception if we want to recover from a KDU error.
        // Because this confused me: the above quote specifically refers to
        // the kdu_error class, which is constructed internally within KDU at
        // the point where a fatal error is discovered and reported. It is NOT
        // talking about the kdu_message subclass passed to
        // kdu_customize_errors(). Destroying this static object at program
        // shutdown will NOT engage the behavior described above.
        if (end_of_message)
        {
            LLTHROW(KDUError("LLKDUMessageError::flush()"));
        }
    }
};
// Instantiating LLKDUMessageError calls kdu_customize_errors() with the new
// instance. Make it static so this only happens once.
static LLKDUMessageError sErrorHandler;

LLImageJ2CKDU::LLImageJ2CKDU() : LLImageJ2CImpl(),
    mInputp(),
    mCodeStreamp(),
    mTPosp(),
    mTileIndicesp(),
    mRawImagep(NULL),
    mDecodeState(),
    mBlocksSize(-1),
    mPrecinctsSize(-1),
    mLevels(0)
{
}

LLImageJ2CKDU::~LLImageJ2CKDU()
{
    cleanupCodeStream(); // in case destroyed before decode completed
}

// Stuff for new simple decode
void transfer_bytes(kdu_byte *dest, kdu_line_buf &src, int gap, int precision);

// This is called by the real (private) initDecode() (keep_codestream true)
// and getMetadata() methods (keep_codestream false). As far as nat can tell,
// mode is always MODE_FAST. It was called by findDiscardLevelsBoundaries()
// as well, when that still existed, with keep_codestream true and MODE_FAST.
void LLImageJ2CKDU::setupCodeStream(LLImageJ2C &base, bool keep_codestream, ECodeStreamMode mode)
{
    LLImageDataLock lock(&base);

    S32 data_size = base.getDataSize();
    S32 max_bytes = (base.getMaxBytes() ? base.getMaxBytes() : data_size);

    //
    //  Initialization
    //
    mCodeStreamp.reset();

    // It's not clear to nat under what circumstances we would reuse a
    // pre-existing LLKDUMemSource instance. As of 2016-08-05, it consists of
    // two U32s and a pointer, so it's not as if it would be a huge overhead
    // to allocate a new one every time.
    // Also -- why is base.getData() tested specifically here? If that returns
    // NULL, shouldn't we bail out of the whole method?
    if (!mInputp && base.getData())
    {
        // The compressed data has been loaded
        // Setup the source for the codestream
        mInputp.reset(new LLKDUMemSource(base.getData(), data_size));
    }

    if (mInputp)
    {
        // This is LLKDUMemSource::reset(), not boost::scoped_ptr::reset().
        mInputp->reset();
    }

    mCodeStreamp->create(mInputp.get());

    // Set the maximum number of bytes to use from the codestream
    // *TODO: This seems to be wrong. The base class should have no idea of
    // how j2c compression works so no good way of computing what's the byte
    // range to be used.
    mCodeStreamp->set_max_bytes(max_bytes,true);

    //  If you want to flip or rotate the image for some reason, change
    // the resolution, or identify a restricted region of interest, this is
    // the place to do it.  You may use "kdu_codestream::change_appearance"
    // and "kdu_codestream::apply_input_restrictions" for this purpose.
    //  If you wish to truncate the code-stream prior to decompression, you
    // may use "kdu_codestream::set_max_bytes".
    //  If you wish to retain all compressed data so that the material
    // can be decompressed multiple times, possibly with different appearance
    // parameters, you should call "kdu_codestream::set_persistent" here.
    //  There are a variety of other features which must be enabled at
    // this point if you want to take advantage of them.  See the
    // descriptions appearing with the "kdu_codestream" interface functions
    // in "kdu_compressed.h" for an itemized account of these capabilities.

    switch (mode)
    {
    case MODE_FAST:
        mCodeStreamp->set_fast();
        break;
    case MODE_RESILIENT:
        mCodeStreamp->set_resilient();
        break;
    case MODE_FUSSY:
        mCodeStreamp->set_fussy();
        break;
    default:
        llassert(0);
        mCodeStreamp->set_fast();
    }

    kdu_dims dims;
    mCodeStreamp->get_dims(0,dims);

    S32 components = mCodeStreamp->get_num_components();

    // Check that components have consistent dimensions (for PPM file)
    for (int idx = 1; idx < components; ++idx)
    {
        kdu_dims other_dims;
        mCodeStreamp->get_dims(idx, other_dims);
        if (other_dims != dims)
        {
            // This method is only called from methods that catch KDUError.
            // We want to fail the image load, not crash the viewer.
            LLTHROW(KDUError(STRINGIZE("Component " << idx << " dimensions "
                                       << stringize(other_dims)
                                       << " do not match component 0 dimensions "
                                       << stringize(dims) << "!")));
        }
    }

    // Get the number of resolution levels in that image
    mLevels = mCodeStreamp->get_min_dwt_levels();

    // Set the base dimensions
    base.setSize(dims.size.x, dims.size.y, components);
    base.setLevels(mLevels);

    if (!keep_codestream)
    {
        mCodeStreamp.reset();
        mInputp.reset();
    }
}

void LLImageJ2CKDU::cleanupCodeStream()
{
    LL_PROFILE_ZONE_SCOPED_CATEGORY_TEXTURE;
    mInputp.reset();
    mDecodeState.reset();
    mCodeStreamp.reset();
    mTPosp.reset();
    mTileIndicesp.reset();
}

// This is the protected virtual method called by LLImageJ2C::initDecode().
// However, as far as nat can tell, LLImageJ2C::initDecode() is called only by
// llimage_libtest.cpp's load_image() function. No detectable production use.
bool LLImageJ2CKDU::initDecode(LLImageJ2C &base, LLImageRaw &raw_image, int discard_level, int* region)
{
    return initDecode(base,raw_image,0.0f,MODE_FAST,0,4,discard_level,region);
}

bool LLImageJ2CKDU::initEncode(LLImageJ2C &base, LLImageRaw &raw_image, int blocks_size, int precincts_size, int levels)
{
    mPrecinctsSize = precincts_size;
    if (mPrecinctsSize != -1)
    {
        mPrecinctsSize = get_lower_power_two(mPrecinctsSize,MAX_PRECINCT_SIZE);
        mPrecinctsSize = llmax(mPrecinctsSize,MIN_PRECINCT_SIZE);
    }
    mBlocksSize = blocks_size;
    if (mBlocksSize != -1)
    {
        mBlocksSize = get_lower_power_two(mBlocksSize,MAX_BLOCK_SIZE);
        mBlocksSize = llmax(mBlocksSize,MIN_BLOCK_SIZE);
        if (mPrecinctsSize != -1)
        {
            mBlocksSize = llmin(mBlocksSize,mPrecinctsSize);    // blocks *must* be smaller than precincts
        }
    }
    mLevels = levels;
    if (mLevels != 0)
    {
        mLevels = llclamp(mLevels,MIN_DECOMPOSITION_LEVELS,MAX_DECOMPOSITION_LEVELS);
        base.setLevels(mLevels);
    }
    return true;
}

// This is the real (private) initDecode() called both by the protected
// initDecode() method and by decodeImpl(). As far as nat can tell, only the
// decodeImpl() usage matters for production.
bool LLImageJ2CKDU::initDecode(LLImageJ2C &base, LLImageRaw &raw_image, F32 decode_time, ECodeStreamMode mode, S32 first_channel, S32 max_channel_count, int discard_level, int* region)
{
    LL_PROFILE_ZONE_SCOPED_CATEGORY_TEXTURE;
    base.resetLastError();

    // *FIX: kdu calls our callback function if there's an error, and then bombs.
    // To regain control, we throw an exception, and catch it here.
    try
    {
        // Merov : Test!! DO NOT COMMIT!!
        //findDiscardLevelsBoundaries(base);

        base.updateRawDiscardLevel();
        setupCodeStream(base, true, mode);

        mRawImagep = &raw_image;
        mCodeStreamp->change_appearance(false, true, false);

        // Apply loading discard level and cropping if required
        kdu_dims* region_kdu = NULL;
        if (region != NULL)
        {
            region_kdu = new kdu_dims;
            region_kdu->pos.x  = region[0];
            region_kdu->pos.y  = region[1];
            region_kdu->size.x = region[2] - region[0];
            region_kdu->size.y = region[3] - region[1];
        }
        int discard = (discard_level != -1 ? discard_level : base.getRawDiscardLevel());
        //LL_INFOS() << "Merov debug : initDecode, discard used = " << discard << ", asked = " << discard_level << LL_ENDL;
        // Apply loading restrictions
        mCodeStreamp->apply_input_restrictions( first_channel, max_channel_count, discard, 0, region_kdu);

        // Clean-up
        if (region_kdu)
        {
            delete region_kdu;
            region_kdu = NULL;
        }

        // Resize raw_image according to the image to be decoded
        kdu_dims dims; mCodeStreamp->get_dims(0,dims);
        S32 channels = base.getComponents() - first_channel;
        channels = llmin(channels,max_channel_count);
        raw_image.resize(dims.size.x, dims.size.y, channels);

        if (!mTileIndicesp)
        {
            mTileIndicesp.reset(new kdu_dims);
        }
        mCodeStreamp->get_valid_tiles(*mTileIndicesp);
        if (!mTPosp)
        {
            mTPosp.reset(new kdu_coords);
            mTPosp->y = 0;
            mTPosp->x = 0;
        }
    }
    catch (const KDUError& msg)
    {
        base.setLastError(msg.what());
        return false;
    }
    catch (kdu_exception kdu_value)
    {
        // KDU internally throws kdu_exception. It's possible that such an
        // exception might leak out into our code. Catch kdu_exception
        // specially because boost::current_exception_diagnostic_information()
        // could do nothing with it.
        base.setLastError(report_kdu_exception(kdu_value));
        return false;
    }
    catch (...)
    {
        base.setLastError("Unknown J2C error: " +
                          boost::current_exception_diagnostic_information());
        return false;
    }

    return true;
}


// Returns true to mean done, whether successful or not.
bool LLImageJ2CKDU::decodeImpl(LLImageJ2C &base, LLImageRaw &raw_image, F32 decode_time, S32 first_channel, S32 max_channel_count)
{
    LL_PROFILE_ZONE_SCOPED_CATEGORY_TEXTURE;

    LLImageDataLock lockIn(&base);
    LLImageDataLock lockOut(&raw_image);

    ECodeStreamMode mode = MODE_FAST;

    bool limit_time = decode_time > 0.0f;
    LLTimer decode_timer;

    if (!mCodeStreamp->exists())
    {
        if (!initDecode(base, raw_image, decode_time, mode, first_channel, max_channel_count))
        {
            // Initializing the J2C decode failed, bail out.
            cleanupCodeStream();
            return true; // done
        }
    }

    // These can probably be grabbed from what's saved in the class.
    kdu_dims dims;
    mCodeStreamp->get_dims(0, dims);

    // Now we are ready to walk through the tiles processing them one-by-one.
    kdu_byte *buffer = raw_image.getData();
    if (!buffer)
    {
        base.setLastError("Memory error");
        base.decodeFailed();
        cleanupCodeStream();
        return true; // done
    }

    while (mTPosp->y < mTileIndicesp->size.y)
    {
        while (mTPosp->x < mTileIndicesp->size.x)
        {
            try
            {
                if (!mDecodeState)
                {
                    kdu_tile tile = mCodeStreamp->open_tile(*(mTPosp)+mTileIndicesp->pos);

                    // Find the region of the buffer occupied by this
                    // tile.  Note that we have no control over
                    // sub-sampling factors which might have been used
                    // during compression and so it can happen that tiles
                    // (at the image component level) actually have
                    // different dimensions.  For this reason, we cannot
                    // figure out the buffer region occupied by a tile
                    // directly from the tile indices.  Instead, we query
                    // the highest resolution of the first tile-component
                    // concerning its location and size on the canvas --
                    // the `dims' object already holds the location and
                    // size of the entire image component on the same
                    // canvas coordinate system.  Comparing the two tells
                    // us where the current tile is in the buffer.
                    S32 channels = base.getComponents() - first_channel;
                    if (channels > max_channel_count)
                    {
                        channels = max_channel_count;
                    }
                    kdu_resolution res = tile.access_component(0).access_resolution();
                    kdu_dims tile_dims; res.get_dims(tile_dims);
                    kdu_coords offset = tile_dims.pos - dims.pos;
                    int row_gap = channels*dims.size.x; // inter-row separation
                    kdu_byte *buf = buffer + offset.y*row_gap + offset.x*channels;
                    mDecodeState.reset(new LLKDUDecodeState(tile, buf, row_gap,
                                                            mCodeStreamp.get()));
                }
                // Do the actual processing
                F32 remaining_time = limit_time ? decode_time - decode_timer.getElapsedTimeF32().value() : 0.0f;
                // This is where we do the actual decode.  If we run out of time, return false.
                if (mDecodeState->processTileDecode(remaining_time, limit_time))
                {
                    mDecodeState.reset();
                }
                else
                {
                    // Not finished decoding yet.
                    base.setLastError("Ran out of time while decoding");
                    base.decodeFailed();
                    cleanupCodeStream();
                    return false;
                }
            }
            catch (const KDUError& msg)
            {
                base.setLastError(msg.what());
                base.decodeFailed();
                cleanupCodeStream();
                return true; // done
            }
            catch (kdu_exception kdu_value)
            {
                // KDU internally throws kdu_exception. It's possible that such an
                // exception might leak out into our code. Catch kdu_exception
                // specially because boost::current_exception_diagnostic_information()
                // could do nothing with it.
                base.setLastError(report_kdu_exception(kdu_value));
                base.decodeFailed();
                cleanupCodeStream();
                return true; // done
            }
            catch (...)
            {
                base.setLastError("Unknown J2C error: " +
                                  boost::current_exception_diagnostic_information());
                base.decodeFailed();
                cleanupCodeStream();
                return true; // done
            }


            mTPosp->x++;
        }
        mTPosp->y++;
        mTPosp->x = 0;
    }

    cleanupCodeStream();

    return true;
}


bool LLImageJ2CKDU::encodeImpl(LLImageJ2C &base, const LLImageRaw &raw_image, const char* comment_text, F32 encode_time, bool reversible)
{
    // Declare and set simple arguments
    bool transpose = false;
    bool vflip = true;
    bool hflip = false;

    try
    {
        // Set up input image files
        siz_params siz;

        // Should set rate someplace here
        LLKDUMemIn mem_in(raw_image.getData(),
            raw_image.getDataSize(),
            raw_image.getWidth(),
            raw_image.getHeight(),
            raw_image.getComponents(),
            &siz);

        base.setSize(raw_image.getWidth(), raw_image.getHeight(), raw_image.getComponents());

        int num_components = raw_image.getComponents();

        siz.set(Scomponents,0,0,num_components);
        siz.set(Sdims,0,0,base.getHeight());  // Height of first image component
        siz.set(Sdims,0,1,base.getWidth());   // Width of first image component
        siz.set(Sprecision,0,0,8);  // Image samples have original bit-depth of 8
        siz.set(Ssigned,0,0,false); // Image samples are originally unsigned

        kdu_params *siz_ref = &siz;
        siz_ref->finalize();
        siz_params transformed_siz; // Use this one to construct code-stream
        transformed_siz.copy_from(&siz,-1,-1,-1,0,transpose,false,false);

        // Construct the `kdu_codestream' object and parse all remaining arguments
        U32 max_output_size = base.getWidth()*base.getHeight()*base.getComponents();
        max_output_size = (max_output_size < 1000 ? 1000 : max_output_size);
        U8 *output_buffer = new U8[max_output_size];
        U32 output_size = 0; // Address updated by LLKDUMemTarget to give the final compressed buffer size
        LLKDUMemTarget output(output_buffer, output_size, max_output_size);

        kdu_codestream codestream;
        codestream.create(&transformed_siz,&output);

        if (comment_text)
        {
            // Set the comments for the codestream
            kdu_codestream_comment comment = codestream.add_comment();
            comment.put_text(comment_text);
        }

        if (num_components >= 3)
        {
            // Note that we always use YCC and not YUV
            // *TODO: Verify this doesn't screws up reversible textures (like sculpties) as YCC is not reversible but YUV is...
            set_default_colour_weights(codestream.access_siz());
        }

        // Set codestream options
        int nb_layers = 0;
        kdu_long layer_bytes[MAX_NB_LAYERS];
        U32 max_bytes = (U32)(base.getWidth() * base.getHeight() * base.getComponents());

        // Rate is the argument passed into the LLImageJ2C which specifies the target compression rate. The default is 8:1.
        // *TODO: mRate is actually always 8:1 in the viewer. Test different values.
        llassert (base.mRate > 0.f);
        max_bytes = (U32)((F32)(max_bytes) * base.mRate);

        // This code is where we specify the target number of bytes for each quality layer.
        // We're using a logarithmic spacing rule that fits with our way of fetching texture data.
        // Note: For more info on this layers business, read kdu_codestream::flush() doc in kdu_compressed.h
        layer_bytes[nb_layers++] = FIRST_PACKET_SIZE;
        U32 i = MIN_LAYER_SIZE;
        while ((i < max_bytes) && (nb_layers < (MAX_NB_LAYERS-1)))
        {
            layer_bytes[nb_layers++] = i;
            i *= 4;
        }
        // Note: for small images, we can have (max_bytes < FIRST_PACKET_SIZE), hence the test
        if (layer_bytes[nb_layers-1] < max_bytes)
        {
            // Set the last quality layer so to fit the preset compression ratio
            layer_bytes[nb_layers++] = max_bytes;
        }

        if (reversible)
        {
            // Use 0 for a last quality layer for reversible images so all remaining code blocks will be flushed
            // Hack: KDU encoding for reversible images has a bug for small images that leads to j2c images that
            // cannot be open or are very blurry. Avoiding that last layer prevents the problem to happen.
            if ((base.getWidth() >= 32) || (base.getHeight() >= 32))
            {
                layer_bytes[nb_layers++] = 0;
            }
            codestream.access_siz()->parse_string("Creversible=yes");
            // *TODO: we should use yuv in reversible mode
            // Don't turn this on now though as it creates problems on decoding for the moment
            //codestream.access_siz()->parse_string("Cycc=no");
        }

        std::string layer_string = llformat("Clayers=%d",nb_layers);
        codestream.access_siz()->parse_string(layer_string.c_str());

        // Set up data ordering, markers, etc... if precincts or blocks specified
        if ((mBlocksSize != -1) || (mPrecinctsSize != -1))
        {
            if (mPrecinctsSize != -1)
            {
                std::string precincts_string = llformat("Cprecincts={%d,%d}",mPrecinctsSize,mPrecinctsSize);
                codestream.access_siz()->parse_string(precincts_string.c_str());
            }
            if (mBlocksSize != -1)
            {
                std::string blocks_string = llformat("Cblk={%d,%d}",mBlocksSize,mBlocksSize);
                codestream.access_siz()->parse_string(blocks_string.c_str());
            }
            std::string ordering_string = llformat("Corder=LRCP");
            codestream.access_siz()->parse_string(ordering_string.c_str());
            std::string PLT_string = llformat("ORGgen_plt=yes");
            codestream.access_siz()->parse_string(PLT_string.c_str());
            std::string Parts_string = llformat("ORGtparts=R");
            codestream.access_siz()->parse_string(Parts_string.c_str());
        }

        // Set the number of wavelets subresolutions (aka levels)
        if (mLevels != 0)
        {
            std::string levels_string = llformat("Clevels=%d",mLevels);
            codestream.access_siz()->parse_string(levels_string.c_str());
        }

        // Complete the encode settings
        codestream.access_siz()->finalize_all();
        codestream.change_appearance(transpose,vflip,hflip);

        // Now we are ready for sample data processing
        kdc_flow_control *tile = new kdc_flow_control(&mem_in,codestream);
        bool done = false;
        while (!done)
        {
            // Process line by line
            if (tile->advance_components())
            {
                tile->process_components();
            }
            else
            {
                done = true;
            }
        }

        // Produce the compressed output
        codestream.flush(layer_bytes,nb_layers);

        // Cleanup
        delete tile;
        codestream.destroy();

        // Now that we're done encoding, create the new data buffer for the compressed
        // image and stick it there.
        base.copyData(output_buffer, output_size);
        base.updateData(); // set width, height
        delete[] output_buffer;
    }
    catch(const KDUError& msg)
    {
        base.setLastError(msg.what());
        return false;
    }
    catch (kdu_exception kdu_value)
    {
        // KDU internally throws kdu_exception. It's possible that such an
        // exception might leak out into our code. Catch kdu_exception
        // specially because boost::current_exception_diagnostic_information()
        // could do nothing with it.
        base.setLastError(report_kdu_exception(kdu_value));
        return false;
    }
    catch( ... )
    {
        base.setLastError("Unknown J2C error: " +
                          boost::current_exception_diagnostic_information());
        return false;
    }

    return true;
}

bool LLImageJ2CKDU::getMetadata(LLImageJ2C &base)
{
    // *FIX: kdu calls our callback function if there's an error, and
    // then bombs. To regain control, we throw an exception, and
    // catch it here.
    try
    {
        setupCodeStream(base, false, MODE_FAST);
        return true;
    }
    catch (const KDUError& msg)
    {
        base.setLastError(msg.what());
        return false;
    }
    catch (kdu_exception kdu_value)
    {
        // KDU internally throws kdu_exception. It's possible that such an
        // exception might leak out into our code. Catch kdu_exception
        // specially because boost::current_exception_diagnostic_information()
        // could do nothing with it.
        base.setLastError(report_kdu_exception(kdu_value));
        return false;
    }
    catch (...)
    {
        base.setLastError("Unknown J2C error: " +
                          boost::current_exception_diagnostic_information());
        return false;
    }
}

/*****************************************************************************/
/* STATIC                        copy_block                                  */
/*****************************************************************************/

/*==========================================================================*|
// Only called by copy_tile(), which is itself commented out
static void copy_block(kdu_block *in, kdu_block *out)
{
    if (in->K_max_prime != out->K_max_prime)
    {
        std::cout << "Cannot copy blocks belonging to subbands with different quantization parameters." << std::endl;
        return;
    }
    if ((in->size.x != out->size.x) || (in->size.y != out->size.y))
    {
        std::cout << "Cannot copy code-blocks with different dimensions." << std::endl;
        return;
    }
    out->missing_msbs = in->missing_msbs;
    if (out->max_passes < (in->num_passes+2))        // Gives us enough to round up
        out->set_max_passes(in->num_passes+2,false); // to the next whole bit-plane
    out->num_passes = in->num_passes;
    int num_bytes = 0;
    for (int z=0; z < in->num_passes; z++)
    {
        num_bytes += (out->pass_lengths[z] = in->pass_lengths[z]);
        out->pass_slopes[z] = in->pass_slopes[z];
    }

    // Just copy compressed code-bytes. Block transcoding not supported.
    if (out->max_bytes < num_bytes)
        out->set_max_bytes(num_bytes,false);
    memcpy(out->byte_buffer,in->byte_buffer,(size_t) num_bytes);
}
|*==========================================================================*/

/*****************************************************************************/
/* STATIC                        copy_tile                                   */
/*****************************************************************************/

/*==========================================================================*|
// Only called by findDiscardLevelsBoundaries(), which is itself commented out
static void
copy_tile(kdu_tile tile_in, kdu_tile tile_out, int tnum_in, int tnum_out,
          kdu_params *siz_in, kdu_params *siz_out, int skip_components,
          int &num_blocks)
{
    int num_components = tile_out.get_num_components();
    int new_tpart=0, next_tpart = 1;

    for (int c=0; c < num_components; c++)
    {
        kdu_tile_comp comp_in, comp_out;
        comp_in = tile_in.access_component(c);
        comp_out = tile_out.access_component(c);
        int num_resolutions = comp_out.get_num_resolutions();
        //std::cout << "    Copying tile : num_resolutions = " << num_resolutions << std::endl;
        for (int r=0; r < num_resolutions; r++)
        {
            kdu_resolution res_in;  res_in = comp_in.access_resolution(r);
            kdu_resolution res_out; res_out = comp_out.access_resolution(r);
            int b, min_band;
            int num_bands = res_in.get_valid_band_indices(min_band);
            std::cout << "        Copying tile : num_bands = " << num_bands << std::endl;
            for (b=min_band; num_bands > 0; num_bands--, b++)
            {
                kdu_subband band_in;  band_in = res_in.access_subband(b);
                kdu_subband band_out; band_out = res_out.access_subband(b);
                kdu_dims blocks_in;  band_in.get_valid_blocks(blocks_in);
                kdu_dims blocks_out; band_out.get_valid_blocks(blocks_out);
                if ((blocks_in.size.x != blocks_out.size.x) ||
                    (blocks_in.size.y != blocks_out.size.y))
                {
                    std::cout << "Transcoding operation cannot proceed: Code-block partitions for the input and output code-streams do not agree." << std::endl;
                    return;
                }
                kdu_coords idx;
                //std::cout << "            Copying tile : block indices, x = " << blocks_out.size.x << " and y = " << blocks_out.size.y << std::endl;
                for (idx.y=0; idx.y < blocks_out.size.y; idx.y++)
                {
                    for (idx.x=0; idx.x < blocks_out.size.x; idx.x++)
                    {
                        kdu_block *in =
                        band_in.open_block(idx+blocks_in.pos,&new_tpart);
                        for (; next_tpart <= new_tpart; next_tpart++)
                            siz_out->copy_from(siz_in,tnum_in,tnum_out,next_tpart,
                                               skip_components);
                        kdu_block *out = band_out.open_block(idx+blocks_out.pos);
                        copy_block(in,out);
                        band_in.close_block(in);
                        band_out.close_block(out);
                        num_blocks++;
                    }
                }
            }
        }
    }
}
|*==========================================================================*/

// Find the block boundary for each discard level in the input image.
// We parse the input blocks and copy them in a temporary output stream.
// For the moment, we do nothing more that parsing the raw list of blocks and outputing result.
/*==========================================================================*|
// See comments in header file for why this is commented out.
void LLImageJ2CKDU::findDiscardLevelsBoundaries(LLImageJ2C &base)
{
    // We need the number of levels in that image before starting.
    getMetadata(base);

    for (int discard_level = 0; discard_level < mLevels; discard_level++)
    {
        //std::cout << "Parsing discard level = " << discard_level << std::endl;
        // Create the input codestream object.
        setupCodeStream(base, true, MODE_FAST);
        mCodeStreamp->apply_input_restrictions(0, 4, discard_level, 0, NULL);
        mCodeStreamp->set_max_bytes(KDU_LONG_MAX,true);
        siz_params *siz_in = mCodeStreamp->access_siz();

        // Create the output codestream object.
        siz_params siz;
        siz.copy_from(siz_in,-1,-1,-1,0,discard_level,false,false,false);
        siz.set(Scomponents,0,0,mCodeStreamp->get_num_components());

        U32 max_output_size = base.getWidth()*base.getHeight()*base.getComponents();
        max_output_size = (max_output_size < 1000 ? 1000 : max_output_size);
        U8 *output_buffer = new U8[max_output_size];
        U32 output_size = 0; // Address updated by LLKDUMemTarget to give the final compressed buffer size
        LLKDUMemTarget output(output_buffer, output_size, max_output_size);
        kdu_codestream codestream_out;
        codestream_out.create(&siz,&output);
        //codestream_out.share_buffering(*mCodeStreamp);
        siz_params *siz_out = codestream_out.access_siz();
        siz_out->copy_from(siz_in,-1,-1,-1,0,discard_level,false,false,false);
        codestream_out.access_siz()->finalize_all(-1);

        // Set up rate control variables
        kdu_long max_bytes = KDU_LONG_MAX;
        kdu_params *cod = siz_out->access_cluster(COD_params);
        int total_layers;  cod->get(Clayers,0,0,total_layers);
        kdu_long *layer_bytes = new kdu_long[total_layers];
        int nel, non_empty_layers = 0;

        // Now ready to perform the transfer of compressed data between streams
        int flush_counter = INT_MAX;
        kdu_dims tile_indices_in;
        mCodeStreamp->get_valid_tiles(tile_indices_in);
        kdu_dims tile_indices_out;
        codestream_out.get_valid_tiles(tile_indices_out);
        assert((tile_indices_in.size.x == tile_indices_out.size.x) &&
               (tile_indices_in.size.y == tile_indices_out.size.y));
        int num_blocks=0;

        kdu_coords idx;
        //std::cout << "Parsing tiles : x = " << tile_indices_out.size.x << " to y = " << tile_indices_out.size.y << std::endl;
        for (idx.y=0; idx.y < tile_indices_out.size.y; idx.y++)
        {
            for (idx.x=0; idx.x < tile_indices_out.size.x; idx.x++)
            {
                kdu_tile tile_in = mCodeStreamp->open_tile(idx+tile_indices_in.pos);
                int tnum_in = tile_in.get_tnum();
                int tnum_out = idx.x + idx.y*tile_indices_out.size.x;
                siz_out->copy_from(siz_in,tnum_in,tnum_out,0,0,discard_level,false,false,false);
                siz_out->finalize_all(tnum_out);
                // Note: do not open the output tile without first copying any tile-specific code-stream parameters
                kdu_tile tile_out = codestream_out.open_tile(idx+tile_indices_out.pos);
                assert(tnum_out == tile_out.get_tnum());
                copy_tile(tile_in,tile_out,tnum_in,tnum_out,siz_in,siz_out,0,num_blocks);
                tile_in.close();
                tile_out.close();
                flush_counter--;
                if ((flush_counter <= 0) && codestream_out.ready_for_flush())
                {
                    flush_counter = INT_MAX;
                    nel = codestream_out.trans_out(max_bytes,layer_bytes,total_layers);
                    non_empty_layers = (nel > non_empty_layers)?nel:non_empty_layers;
                }
            }
        }

        // Generate the output code-stream
        if (codestream_out.ready_for_flush())
        {
            nel = codestream_out.trans_out(max_bytes,layer_bytes,total_layers);
            non_empty_layers = (nel > non_empty_layers)?nel:non_empty_layers;
        }
        if (non_empty_layers > total_layers)
            non_empty_layers = total_layers; // Can happen if a tile has more layers

        // Print out stats
        std::cout << "Code stream parsing for discard level = " << discard_level << std::endl;
        std::cout << "    Total compressed memory in  = " << mCodeStreamp->get_compressed_data_memory() << " bytes" << std::endl;
        std::cout << "    Total compressed memory out = " << codestream_out.get_compressed_data_memory() << " bytes" << std::endl;
        //std::cout << "    Output contains " << total_layers << " quality layers" << std::endl;
        std::cout << "    Transferred " << num_blocks << " code-blocks from in to out" << std::endl;
        //std::cout << "    Read " << mCodeStreamp->get_num_tparts() << " tile-part(s) from a total of " << (int) tile_indices_in.area() << " tile(s)" << std::endl;
        std::cout << "    Total bytes read = " << mCodeStreamp->get_total_bytes() << std::endl;
        //std::cout << "    Wrote " << codestream_out.get_num_tparts() << " tile-part(s) in a total of " << (int) tile_indices_out.area() << " tile(s)" << std::endl;
        std::cout << "    Total bytes written = " << codestream_out.get_total_bytes() << std::endl;
        std::cout << "-------------" << std::endl;

        // Clean-up
        cleanupCodeStream();
        codestream_out.destroy();
        delete[] output_buffer;
    }
    return;
}
|*==========================================================================*/

void set_default_colour_weights(kdu_params *siz)
{
    kdu_params *cod = siz->access_cluster(COD_params);
    assert(cod != NULL);

    bool can_use_ycc = true;
    bool rev0 = false;
    int depth0 = 0, sub_x0 = 1, sub_y0 = 1;
    for (int c = 0; c < 3; c++)
    {
        int depth = 0; siz->get(Sprecision,c,0,depth);
        int sub_y = 1; siz->get(Ssampling,c,0,sub_y);
        int sub_x = 1; siz->get(Ssampling,c,1,sub_x);
        kdu_params *coc = cod->access_relation(-1,c);
        bool rev = false; coc->get(Creversible,0,0,rev);
        if (c == 0)
        {
            rev0 = rev; depth0 = depth; sub_x0 = sub_x; sub_y0 = sub_y;
        }
        else if ((rev != rev0) || (depth != depth0) ||
                 (sub_x != sub_x0) || (sub_y != sub_y0))
        {
            can_use_ycc = false;
        }
    }
    if (!can_use_ycc)
    {
        return;
    }

    bool use_ycc;
    if (!cod->get(Cycc,0,0,use_ycc))
    {
        cod->set(Cycc,0,0,use_ycc=true);
    }
    if (!use_ycc)
    {
        return;
    }
    float weight;
    if (cod->get(Clev_weights,0,0,weight) || cod->get(Cband_weights,0,0,weight))
    {
        // Weights already specified explicitly -> nothing to do
        return;
    }

    // These example weights are adapted from numbers generated by Marcus Nadenau
    // at EPFL, for a viewing distance of 15 cm and a display resolution of
    // 300 DPI.

    cod->parse_string("Cband_weights:C0="
        "{0.0901},{0.2758},{0.2758},"
        "{0.7018},{0.8378},{0.8378},{1}");
    cod->parse_string("Cband_weights:C1="
        "{0.0263},{0.0863},{0.0863},"
        "{0.1362},{0.2564},{0.2564},"
        "{0.3346},{0.4691},{0.4691},"
        "{0.5444},{0.6523},{0.6523},"
        "{0.7078},{0.7797},{0.7797},{1}");
    cod->parse_string("Cband_weights:C2="
        "{0.0773},{0.1835},{0.1835},"
        "{0.2598},{0.4130},{0.4130},"
        "{0.5040},{0.6464},{0.6464},"
        "{0.7220},{0.8254},{0.8254},"
        "{0.8769},{0.9424},{0.9424},{1}");
}

/******************************************************************************/
/*                              transfer_bytes                                */
/******************************************************************************/

void transfer_bytes(kdu_byte *dest, kdu_line_buf &src, int gap, int precision)
/* Transfers source samples from the supplied line buffer into the output
byte buffer, spacing successive output samples apart by `gap' bytes
(to allow for interleaving of colour components).  The function performs
all necessary level shifting, type conversion, rounding and truncation. */
{
    int width = src.get_width();
    if (src.get_buf32() != NULL)
    { // Decompressed samples have a 32-bit representation (integer or float)
        assert(precision >= 8); // Else would have used 16 bit representation
        kdu_sample32 *sp = src.get_buf32();
        if (!src.is_absolute())
        { // Transferring normalized floating point data.
            float scale16 = (float)(1<<16);
            kdu_int32 val;

            for (; width > 0; width--, sp++, dest+=gap)
            {
                val = (kdu_int32)(sp->fval*scale16);
                val = (val+128)>>8; // May be faster than true rounding
                val += 128;
                if (val & ((0xffffffffU)<<8))
                {
                    val = (val < 0 ? 0 : 255);
                }
                *dest = (kdu_byte) val;
            }
        }
        else
        { // Transferring 32-bit absolute integers.
            kdu_int32 val;
            kdu_int32 downshift = precision-8;
            kdu_int32 offset = (1<<downshift)>>1;

            for (; width > 0; width--, sp++, dest+=gap)
            {
                val = sp->ival;
                val = (val+offset)>>downshift;
                val += 128;
                if (val & ((0xffffffffU)<<8))
                {
                    val = (val < 0 ? 0 : 255);
                }
                *dest = (kdu_byte) val;
            }
        }
    }
    else
    { // Source data is 16 bits.
        kdu_sample16 *sp = src.get_buf16();
        if (!src.is_absolute())
        { // Transferring 16-bit fixed point quantities
            kdu_int16 val;

            if (precision >= 8)
            { // Can essentially ignore the bit-depth.
                for (; width > 0; width--, sp++, dest+=gap)
                {
                    val = sp->ival;
                    val += (1<<(KDU_FIX_POINT-8))>>1;
                    val >>= (KDU_FIX_POINT-8);
                    val += 128;
                    if (val & ((0xffffffffU)<<8))
                    {
                        val = (val < 0 ? 0 : 255);
                    }
                    *dest = (kdu_byte) val;
                }
            }
            else
            { // Need to force zeros into one or more least significant bits.
                kdu_int16 downshift = KDU_FIX_POINT-precision;
                kdu_int16 upshift = 8-precision;
                kdu_int16 offset = 1<<(downshift-1);

                for (; width > 0; width--, sp++, dest+=gap)
                {
                    val = sp->ival;
                    val = (val+offset)>>downshift;
                    val <<= upshift;
                    val += 128;
                    if (val & ((0xffffffffU)<<8))
                    {
                        val = (val < 0 ? 0 : 256 - (1<<upshift));
                    }
                    *dest = (kdu_byte) val;
                }
            }
        }
        else
        { // Transferring 16-bit absolute integers.
            kdu_int16 val;

            if (precision >= 8)
            {
                kdu_int16 downshift = precision-8;
                kdu_int16 offset = (1<<downshift)>>1;

                for (; width > 0; width--, sp++, dest+=gap)
                {
                    val = sp->ival;
                    val = (val+offset)>>downshift;
                    val += 128;
                    if (val & ((0xffffffffU)<<8))
                    {
                        val = (val < 0 ? 0 : 255);
                    }
                    *dest = (kdu_byte) val;
                }
            }
            else
            {
                kdu_int16 upshift = 8-precision;

                for (; width > 0; width--, sp++, dest+=gap)
                {
                    val = sp->ival;
                    val <<= upshift;
                    val += 128;
                    if (val & ((0xffffffffU)<<8))
                    {
                        val = (val < 0 ? 0 : 256 - (1<<upshift));
                    }
                    *dest = (kdu_byte) val;
                }
            }
        }
    }
}

LLKDUDecodeState::LLKDUDecodeState(kdu_tile tile, kdu_byte *buf, S32 row_gap,
                                   kdu_codestream* codestreamp)
{
    S32 c;

    mTile = tile;
    mBuf = buf;
    mRowGap = row_gap;

    mNumComponents = tile.get_num_components();

    llassert(mNumComponents <= 4);
    mUseYCC = tile.get_ycc();

    for (c = 0; c < 4; ++c)
    {
        mReversible[c] = false;
        mBitDepths[c] = 0;
    }

    // Open tile-components and create processing engines and resources
    for (c = 0; c < mNumComponents; c++)
    {
        mComps[c] = mTile.access_component(c);
        mReversible[c] = mComps[c].get_reversible();
        mBitDepths[c] = mComps[c].get_bit_depth();
        kdu_resolution res = mComps[c].access_resolution(); // Get top resolution
        kdu_dims comp_dims; res.get_dims(comp_dims);
        if (c == 0)
        {
            mDims = comp_dims;
        }
        else
        {
            llassert(mDims == comp_dims); // Safety check; the caller has ensured this
        }
        bool use_shorts = (mComps[c].get_bit_depth(true) <= 16);
        mLines[c].pre_create(&mAllocator,mDims.size.x,mReversible[c],use_shorts,0,0);
        if (res.which() == 0) // No DWT levels used
        {
            mEngines[c] = kdu_decoder(res.access_subband(LL_BAND),&mAllocator,use_shorts);
        }
        else
        {
            mEngines[c] = kdu_synthesis(res,&mAllocator,use_shorts);
        }
    }
    mAllocator.finalize(*codestreamp); // Actually creates buffering resources
    for (c = 0; c < mNumComponents; c++)
    {
        mLines[c].create(); // Grabs resources from the allocator.
    }
}

LLKDUDecodeState::~LLKDUDecodeState()
{
    // Cleanup
    for (S32 c = 0; c < mNumComponents; c++)
    {
        mEngines[c].destroy(); // engines are interfaces; no default destructors
    }
    mTile.close();
}

bool LLKDUDecodeState::processTileDecode(F32 decode_time, bool limit_time)
/* Decompresses a tile, writing the data into the supplied byte buffer.
The buffer contains interleaved image components, if there are any.
Although you may think of the buffer as belonging entirely to this tile,
the `buf' pointer may actually point into a larger buffer representing
multiple tiles.  For this reason, `row_gap' is needed to identify the
separation between consecutive rows in the real buffer. */
{
    LL_PROFILE_ZONE_SCOPED_CATEGORY_TEXTURE;
    S32 c;
    // Now walk through the lines of the buffer, recovering them from the
    // relevant tile-component processing engines.

    LLTimer decode_timer;
    while (mDims.size.y--)
    {
        {
            LL_PROFILE_ZONE_NAMED_CATEGORY_TEXTURE("kduptc - pull");
            for (c = 0; c < mNumComponents; c++)
            {
                mEngines[c].pull(mLines[c]);
            }
        }

        if ((mNumComponents >= 3) && mUseYCC)
        {
            LL_PROFILE_ZONE_NAMED_CATEGORY_TEXTURE("kduptc - convert");
            kdu_convert_ycc_to_rgb(mLines[0],mLines[1],mLines[2]);
        }

        {
            LL_PROFILE_ZONE_NAMED_CATEGORY_TEXTURE("kduptc - transfer");
            for (c = 0; c < mNumComponents; c++)
            {
                transfer_bytes(mBuf + c, mLines[c], mNumComponents, mBitDepths[c]);
            }
        }
        mBuf += mRowGap;
        if (mDims.size.y % 10)
        {
            if (limit_time && decode_timer.getElapsedTimeF32() > decode_time)
            {
                return false;
            }
        }
    }
    return true;
}

// kdc_flow_control

kdc_flow_control::kdc_flow_control (kdu_supp::kdu_image_in_base *img_in, kdu_codestream codestream)
{
    int n;

    this->codestream = codestream;
    codestream.get_valid_tiles(valid_tile_indices);
    tile_idx = valid_tile_indices.pos;
    tile = codestream.open_tile(tile_idx,NULL);

    // Set up the individual components
    num_components = codestream.get_num_components(true);
    components = new kdc_component_flow_control[num_components];
    count_delta = 0;
    kdc_component_flow_control *comp = components;
    for (n = 0; n < num_components; n++, comp++)
    {
        comp->line = NULL;
        comp->reader = img_in;
        kdu_coords subsampling;
        codestream.get_subsampling(n,subsampling,true);
        kdu_dims dims;
        codestream.get_tile_dims(tile_idx,n,dims,true);
        comp->vert_subsampling = subsampling.y;
        if ((n == 0) || (comp->vert_subsampling < count_delta))
        {
            count_delta = comp->vert_subsampling;
        }
        comp->ratio_counter = 0;
        comp->remaining_lines = comp->initial_lines = dims.size.y;
    }
    assert(num_components >= 0);

    tile.set_components_of_interest(num_components);
    max_buffer_memory = engine.create(codestream,tile,false,NULL,false,1,NULL,NULL,false);
}

kdc_flow_control::~kdc_flow_control()
{
    if (components != NULL)
    {
        delete[] components;
    }
    if (engine.exists())
    {
        engine.destroy();
    }
}

bool kdc_flow_control::advance_components()
{
    bool found_line = false;
    while (!found_line)
    {
        bool all_done = true;
        kdc_component_flow_control *comp = components;
        for (int n = 0; n < num_components; n++, comp++)
        {
            assert(comp->ratio_counter >= 0);
            if (comp->remaining_lines > 0)
            {
                all_done = false;
                comp->ratio_counter -= count_delta;
                if (comp->ratio_counter < 0)
                {
                    found_line = true;
                    comp->line = engine.exchange_line(n,NULL,NULL);
                    assert(comp->line != NULL);
                    if (comp->line->get_width())
                    {
                        comp->reader->get(n,*(comp->line),0);
                    }
                }
            }
        }
        if (all_done)
        {
            return false;
        }
    }
    return true;
}

void kdc_flow_control::process_components()
{
    kdc_component_flow_control *comp = components;
    for (int n = 0; n < num_components; n++, comp++)
    {
        if (comp->ratio_counter < 0)
        {
            comp->ratio_counter += comp->vert_subsampling;
            assert(comp->ratio_counter >= 0);
            assert(comp->remaining_lines > 0);
            comp->remaining_lines--;
            assert(comp->line != NULL);
            engine.exchange_line(n,comp->line,NULL);
            comp->line = NULL;
        }
    }
}