multiply A by 2, and use it as an offset to a memory read.

Basically a uint16_t aligned read.  Would be single instruction
on many architectures. Not copyrightable.