int24 - 24 bit integral datatype

Question

Hi,
is there a 24Bit primitive integral datatype in C++?

If there is none, would it be possible to create a class int24 (, uint24 ) ?

it's purpose could be:
* manipulating soundfiles in 24 bit format
* manipulating bitmapdata without alphachannel

many thanks in advance

Oops

Answer 1

I wrote this to help me with audio manipulation. Its not the fastest but it works for me :)

const int INT24_MAX = 8388607;

class Int24
{
protected:
    unsigned char m_Internal[3];
public:
    Int24()
    {
    }

    Int24( const int val )
    {
        *this   = val;
    }

    Int24( const Int24& val )
    {
        *this   = val;
    }

    operator int() const
    {
        if ( m_Internal[2] & 0x80 ) // Is this a negative?  Then we need to siingn extend.
        {
            return (0xff << 24) | (m_Internal[2] << 16) | (m_Internal[1] << 8) | (m_Internal[0] << 0);
        }
        else
        {
            return (m_Internal[2] << 16) | (m_Internal[1] << 8) | (m_Internal[0] << 0);
        }
    }

    operator float() const
    {
        return (float)this->operator int();
    }

    Int24& operator =( const Int24& input )
    {
        m_Internal[0]   = input.m_Internal[0];
        m_Internal[1]   = input.m_Internal[1];
        m_Internal[2]   = input.m_Internal[2];

        return *this;
    }

    Int24& operator =( const int input )
    {
        m_Internal[0]   = ((unsigned char*)&input)[0];
        m_Internal[1]   = ((unsigned char*)&input)[1];
        m_Internal[2]   = ((unsigned char*)&input)[2];

        return *this;
    }

    /***********************************************/

    Int24 operator +( const Int24& val ) const
    {
        return Int24( (int)*this + (int)val );
    }

    Int24 operator -( const Int24& val ) const
    {
        return Int24( (int)*this - (int)val );
    }

    Int24 operator *( const Int24& val ) const
    {
        return Int24( (int)*this * (int)val );
    }

    Int24 operator /( const Int24& val ) const
    {
        return Int24( (int)*this / (int)val );
    }

    /***********************************************/

    Int24 operator +( const int val ) const
    {
        return Int24( (int)*this + val );
    }

    Int24 operator -( const int val ) const
    {
        return Int24( (int)*this - val );
    }

    Int24 operator *( const int val ) const
    {
        return Int24( (int)*this * val );
    }

    Int24 operator /( const int val ) const
    {
        return Int24( (int)*this / val );
    }

    /***********************************************/
    /***********************************************/


    Int24& operator +=( const Int24& val )
    {
        *this   = *this + val;
        return *this;
    }

    Int24& operator -=( const Int24& val )
    {
        *this   = *this - val;
        return *this;
    }

    Int24& operator *=( const Int24& val )
    {
        *this   = *this * val;
        return *this;
    }

    Int24& operator /=( const Int24& val )
    {
        *this   = *this / val;
        return *this;
    }

    /***********************************************/

    Int24& operator +=( const int val )
    {
        *this   = *this + val;
        return *this;
    }

    Int24& operator -=( const int val )
    {
        *this   = *this - val;
        return *this;
    }

    Int24& operator *=( const int val )
    {
        *this   = *this * val;
        return *this;
    }

    Int24& operator /=( const int val )
    {
        *this   = *this / val;
        return *this;
    }

    /***********************************************/
    /***********************************************/

    Int24 operator >>( const int val ) const
    {
        return Int24( (int)*this >> val );
    }

    Int24 operator <<( const int val ) const
    {
        return Int24( (int)*this << val );
    }

    /***********************************************/

    Int24& operator >>=( const int val )
    {
        *this = *this >> val;
        return *this;
    }

    Int24& operator <<=( const int val )
    {
        *this = *this << val;
        return *this;
    }

    /***********************************************/
    /***********************************************/

    operator bool() const
    {
        return (int)*this != 0;
    }

    bool operator !() const
    {
        return !((int)*this);
    }

    Int24 operator -()
    {
        return Int24( -(int)*this );
    }

    /***********************************************/
    /***********************************************/

    bool operator ==( const Int24& val ) const
    {
        return (int)*this == (int)val;
    }

    bool operator !=( const Int24& val ) const
    {
        return (int)*this != (int)val;
    }

    bool operator >=( const Int24& val ) const
    {
        return (int)*this >= (int)val;
    }

    bool operator <=( const Int24& val ) const
    {
        return (int)*this <= (int)val;
    }

    bool operator >( const Int24& val ) const
    {
        return (int)*this > (int)val;
    }

    bool operator <( const Int24& val ) const
    {
        return (int)*this < (int)val;
    }

    /***********************************************/

    bool operator ==( const int val ) const
    {
        return (int)*this == val;
    }

    bool operator !=( const int val ) const
    {
        return (int)*this != val;
    }

    bool operator >=( const int val ) const
    {
        return (int)*this >= val;
    }

    bool operator <=( const int val ) const
    {
        return (int)*this <= val;
    }

    bool operator >( const int val ) const
    {
        return ((int)*this) > val;
    }

    bool operator <( const int val ) const
    {
        return (int)*this < val;
    }

    /***********************************************/
    /***********************************************/
};

Answer 2

No - all you can really do is:

typedef int32_t int24_t;

which helps to make code/intent more readable/obvious, but doesn't impose any limits on range or storage space.

Answer 3

Depending on the requirements I'd use a bitfield for it.

struct int24{
    unsigned int data : 24;
};

Or, if a separation is easier, just use 3 bytes (chars).

Btw, both use cases you mention in the question generally use 32bit integers. In the case of audio processing you'll generally convert to 32 bit ints (or floats, preferably, to prevent overflow situations you'd get with fixed point or integer math) when loading in chunks of audio because you're not going to have the entire file in memory at once.

For image data, people just tend to use 32 bit integers and ignore the alpha 8 alpha bits all together, or if you're dealing with a tightly packed format you're probably better of just manipulating them as char-pointers anyway because you'll have all channels separate. It's going to be a performance/memory trade-off anyway because writing one int is generally faster than three chars separately; however it will take 25% more memory.

Packing structs like this is compiler specific. However, in Visual Studio you'd do the following to make the struct exactly 24 bits.

#pragma pack(push, 1)
struct int24{
    unsigned int data : 24;
}
#pragma pack(pop)

Answer 4

Working with anything smaller than an integer (32 or 64 bit depending on your architecture) is not ideal. All CPU operations of the smaller data types (short, etc) are done using integer arithmetic. Conversion to and from the CPU has to be done, slowing your application down (even if it is just a tad).

My advice: Store them as 32 (or 64 bit) integers to improve your overall speed. When it comes time to do I/O, then you'll have to do the conversion yourself.

As far as manipulating audio data, there are many libraries available that take care of the I/O for you - unless you want to start learning how PCM, etc are stored - as well as other DSP functions. I would suggest using one of the many libraries out there.