utf8 case folding and/or comparisons?

[liquidaty]

Hi,

Thank you for your work on this. Would it be possible to use this (or any other SIMD-based) approach for fast case folding and/or case-insensitive comparison?

[clausecker]

Case-folding and case-insensitive comparisons are complex and locale dependent. Will be very tricky to vectorise, even if normalised input can be assumed.

[WojciechMula]

At https://github.com/SnellerInc/sneller we have Unicode support and even for upper/lower-case we ended up with huge lookup tables. Here we might consider using gathers, and check the outcome. Although I'm quite sure performance will be mainly bounded by cache misses penalties.

[lin72h]

@WojciechMula I didn't know you are one of the sneller developer, very impressive project BTW!

[WojciechMula]

@WojciechMula I didn't know you are one of the sneller developer, very impressive project BTW!

Yeah, a lot of great stuff is there. :) Getting back to the main topic, there was an attempt to express lookup table as a huge vectorized if-ladder and it was very slow.