If you use the compiler option --64bit_doubles, you can force the compiler to use 64-bit doubles. The data type float is always represented using 32 bits.