Isn't it possible to give a "coefficients" vector with large enough values so that computations with 64 bits would overflow too? I don't doubt the code works in practice, because the range of values are reasonable. However this is not explicit, just looking at the code. I'd prefer if the actual domains were being made obvious. Right now, 12, 32 and 64 feel like magic numbers in the code.
Isn't it possible to give a "coefficients" vector with large enough values so that computations with 64 bits would overflow too? I don't doubt the code works in practice, because the range of values are reasonable. However this is not explicit, just looking at the code. I'd prefer if the actual domains were being made obvious. Right now, 12, 32 and 64 feel like magic numbers in the code.