// We need to use shifts here in order to mantain the sign bit for signed
// integers.  The compiler should optimize this to (x & 0x00FFFFFF) for
// unsigned integers.
#define CONVERT_TO_24BIT(x) (((x) << 8) >> 8)

_CLC_OVERLOAD _CLC_DEF __CLC_GENTYPE mul24(__CLC_GENTYPE x, __CLC_GENTYPE y){
  return CONVERT_TO_24BIT(x) * CONVERT_TO_24BIT(y);
}

#undef CONVERT_TO_24BIT