This has been suggested (and rejected) many times, AFAIK one of the first was this one: #2382
(a portable approach of 5 conditional shifts instead of that ugly loop)

But I have tested before, which will improve the performance 10 times on X86.

improve performance of what?
performance of this function when tested in a loop? or some actual workload of redis?
the reason it was rejected by the previous maintainer was that the performance of this function isn't important for redis since is very rarely used.
the other argument against is was that it make the code less readable. this is something i didn't agree with.