Improve performance of mbfl_name2encoding() by using perfect hashing

[nielsdos]

mbfl_name2encoding() uses a linear loop through the encodings, comparing the name one by one, which is very slow. For the benchmark [1] just looking up the name takes about 50% of run-time.

By using perfect hashing instead, we no longer have to loop over the list, and the number of string comparisons is reduced to just a single one. The perfect hashing table is generated using GNU gperf and amended manually to fit in with mbstring and manually changed to reduce the cache size.

Previously the linked benchmark took 2.39s on my system, now it takes 1.50s. This is a nice improvement that can be felt when processing large amounts of data.

[1] #12684 (comment)

[alexdowad]

Nice job!!

I haven't looked at this code for a while, but I seem to recall that it cached the last looked-up encoding and checked if the next requested one was the same using something like strcmp. Or did it just use a pointer equality comparison on the string (so the optimization would only work if the encoding name was interned)? Don't remember. Or perhaps I am thinking about something completely different?

[nielsdos]

Thanks for the review Alex!
The caching seems to be for calls to php_mb_get_encoding, and it only caches the last used encoding.
I'm not sure if the benchmark actually calls php_mb_get_encoding, but even if it does, it is alternating between different encodings so it doesn't benefit from the caching.