Use 16bits arithmetic for F_4

[vrancurel]

Instead of using 32bits arithmetic we can use 16 bits arithmetic.

[vrancurel]

Here is a simulation: https://github.com/vrancurel/f4mul

[vrancurel]

Lâm, can you give it a try ?

[vrancurel]

We can probably further optimize the loop we do to adjust with SIMD operations themselves, e.g. with cmpeq_epu16(), etc

Bitfields operation are in fact unlikely so we can process them only if the bitmap != 0.