Optimized computation for block_len

[kampersanda]

This PR implemented computations for block_len in BuildHelper by using bit shift.

[vbkaisetsu]

Hmm...
I think the benefit of this change is too small for increasing complexity of the struct.

First, multiplication is as efficient as shifting, or just a little slower.
Second, division is significantly slower, but it is a small difference compared to the overall processing.

[vbkaisetsu]

@kampersanda btw, it would make more sense to store the number of blocks rather than the number of elements, wouldn't it?
The number of elements is always the multiply of the number of blocks, and multiplication is faster than division.

[kampersanda]

@vbkaisetsu Makes sense. Your solution will allow for faster computation without increasing complexity.
Thanks for your suggestion. I'll implement it in this PR.

[kampersanda]

I implemented 0544cbc.
@vbkaisetsu Could you review it again?

[vbkaisetsu]

Need #52 before merging this