'blocks_for_bits' implementation.

[Parakleta]

I don't know how the function as it exists is currently optimised, but in principle you can much more simply use (without risk of over/underflow):

if bits == 0 {
    0
} else {
    (bits-1) / B::bits() + 1
}

If bits were signed (and you have a signed right shift operated that was not UB) you can just use (bits-1) >> B::bitsbits() + 1 where bitsbits() is the number of bits required to store the index into a word (i.e. 5 for 32 bit words). This is probably not practical though.