Besides reliable, what requirements do you have? If you want something generic use `trunc` (<https://nim-lang.org/docs/math.html#trunc%2Cfloat32)>.
If you want something fast, extract the mantissa bits from the float and shift by the exponent bits. Example: * <https://github.com/numforge/laser/blob/d1e6ae61/laser/primitives/simd_math/exp_log_fallback.nim#L32-L38> * And float32 bit position: <https://github.com/numforge/laser/blob/d1e6ae61/laser/primitives/simd_math/exp_log_common.nim#L26-L38> Good luck to support big-endian machine in that case though, and make sure to test with denormal, NaN, -Inf, +Inf
