Configuration with 32-bit Int and Float types.

[katyo]

Many ARM and MIPS-based 32-bit embedded platforms works natively faster with single precision floats and int32_t and much slower with 64-bit variants. I assume that it must be configurable depending of platform.