Problems using jpeg_mem_dest

[Anonymous]

Hi,
I use libjpeg-turbo in my project and noticed some problems with using
jpeg_mem_dest. Am I doing something wrong here? THX.

Small image 640x480 compression in loop as examples used in 4 different ways.

example 1: works but leaks

while( 1 )
{
    jpeg_compress_struct m_sCompress = {0};
    jpeg_error_mgr m_sError = {0};

    m_sCompress.err = jpeg_std_error(&m_sError);

    m_sCompress.image_width = pBMPInfoHeader->biWidth;
    m_sCompress.image_height = pBMPInfoHeader->biHeight;
    m_sCompress.input_components = pBMPInfoHeader->biBitCount / 8;
    m_sCompress.in_color_space = JCS_EXT_BGR;

    unsigned char* pCompressed = 0;
    unsigned long lSize = 0;

    jpeg_mem_dest( &m_sCompress, &pCompressed, &lSize );

    jpeg_set_defaults( &m_sCompress );

    jpeg_set_quality( &m_sCompress, 100, FALSE );

    jpeg_start_compress( &m_sCompress, TRUE );

    JSAMPROW pRow[1];
    DWORD dwRowSize = img_size( m_sCompress.image_width, 1, m_sCompress.input_components, 4 );

    while (m_sCompress.next_scanline < m_sCompress.image_height) 
    {
        pRow[0] = & pData[ ( m_sCompress.image_height - m_sCompress.next_scanline - 1 ) * dwRowSize];

        jpeg_write_scanlines(&m_sCompress, pRow, 1);
    }

    jpeg_finish_compress(&m_sCompress);

        SaveCompressed( pCompressed, lSize);

    jpeg_destroy_compress(&m_sCompress);
} //while(1)

example 2: works but leaks

jpeg_compress_struct m_sCompress = {0};
jpeg_error_mgr m_sError = {0};

m_sCompress.err = jpeg_std_error(&m_sError);

m_sCompress.image_width = pBMPInfoHeader->biWidth;
m_sCompress.image_height = pBMPInfoHeader->biHeight;
m_sCompress.input_components = pBMPInfoHeader->biBitCount / 8;
m_sCompress.in_color_space = JCS_EXT_BGR;

unsigned char* pCompressed = 0;
unsigned long lSize = 0;

while( 1 )
{
    jpeg_mem_dest( &m_sCompress, &pCompressed, &lSize );

    jpeg_set_defaults( &m_sCompress );

    jpeg_set_quality( &m_sCompress, 100, FALSE );

    jpeg_start_compress( &m_sCompress, TRUE );

    JSAMPROW pRow[1];
    DWORD dwRowSize = img_size( m_sCompress.image_width, 1, m_sCompress.input_components, 4 );

    while (m_sCompress.next_scanline < m_sCompress.image_height) 
    {
        pRow[0] = & pData[ ( m_sCompress.image_height - m_sCompress.next_scanline - 1 ) * dwRowSize];

        jpeg_write_scanlines(&m_sCompress, pRow, 1);
    }

    jpeg_finish_compress(&m_sCompress);

        SaveCompressed( pCompressed, lSize);
} //while(1)

jpeg_destroy_compress(&m_sCompress);

example 3: works like append -> BAD

jpeg_compress_struct m_sCompress = {0};
jpeg_error_mgr m_sError = {0};

m_sCompress.err = jpeg_std_error(&m_sError);

m_sCompress.image_width = pBMPInfoHeader->biWidth;
m_sCompress.image_height = pBMPInfoHeader->biHeight;
m_sCompress.input_components = pBMPInfoHeader->biBitCount / 8;
m_sCompress.in_color_space = JCS_EXT_BGR;

unsigned char* pCompressed = 0;
unsigned long lSize = 0;

jpeg_mem_dest( &m_sCompress, &pCompressed, &lSize );

jpeg_set_defaults( &m_sCompress );

jpeg_set_quality( &m_sCompress, 100, FALSE );

while( 1 )
{
    jpeg_start_compress( &m_sCompress, TRUE );

    JSAMPROW pRow[1];
    DWORD dwRowSize = img_size( m_sCompress.image_width, 1, m_sCompress.input_components, 4 );

    while (m_sCompress.next_scanline < m_sCompress.image_height) 
    {
        pRow[0] = & pData[ ( m_sCompress.image_height - m_sCompress.next_scanline - 1 ) * dwRowSize];

        jpeg_write_scanlines(&m_sCompress, pRow, 1);
    }

    jpeg_finish_compress(&m_sCompress);

        SaveCompressed( pCompressed, lSize);
} //while(1)

jpeg_destroy_compress(&m_sCompress);

example 4: seems OK, no leak

jpeg_compress_struct m_sCompress = {0};
jpeg_error_mgr m_sError = {0};

m_sCompress.err = jpeg_std_error(&m_sError);

m_sCompress.image_width = pBMPInfoHeader->biWidth;
m_sCompress.image_height = pBMPInfoHeader->biHeight;
m_sCompress.input_components = pBMPInfoHeader->biBitCount / 8;
m_sCompress.in_color_space = JCS_EXT_BGR;

unsigned char* pCompressed = new BYTE[1000000];
unsigned long lSize = 1000000;

while( 1 )
{
    lSize = 1000000;
    jpeg_mem_dest( &m_sCompress, &pCompressed, &lSize );

    jpeg_set_defaults( &m_sCompress );

    jpeg_set_quality( &m_sCompress, 100, FALSE );

    jpeg_start_compress( &m_sCompress, TRUE );

    JSAMPROW pRow[1];
    DWORD dwRowSize = img_size( m_sCompress.image_width, 1, m_sCompress.input_components, 4 );

    while (m_sCompress.next_scanline < m_sCompress.image_height) 
    {
        pRow[0] = & pData[ ( m_sCompress.image_height - m_sCompress.next_scanline - 1 ) * dwRowSize];

        jpeg_write_scanlines(&m_sCompress, pRow, 1);
    }

    jpeg_finish_compress(&m_sCompress);

        SaveCompressed( pCompressed, lSize);
} //while(1)

jpeg_destroy_compress(&m_sCompress);

[DRC]

We borrow tj_mem_dest() from jpeg-8 and don't make any local modifications to
it, so I suspect this is not a library error, but I don't have time to pore
through your code and figure out what it might be. valgrind is your friend.

[Anonymous]

I case that someone is interested, my working solution is:
per frame:

...
//reset output memdest
m_sCompress.dest->free_in_buffer += m_sCompress.dest->next_output_byte - m_pCompressed;
m_sCompress.dest->next_output_byte = m_pCompressed;

//start frame compression
jpeg_start_compress( &m_sCompress, TRUE );
...

m_pCompressed was initially set only once during initialization

...
m_pCompressed = NULL;
m_lSize = 0;
jpeg_mem_dest( &m_sCompress, &m_pCompressed, &m_lSize );
...