I liked UTF-8 more before I started to think hard about efficient RISC-V BitManip code for encoding/decoding it.
-
How is it with edge cases? overbyte: c0 80 e0 80 80 f0 80 80 80 f8 80 80 80 80 fc 80 80 80 80 80 cesu-8: ed a0 bd ed b1 86 >10FFFF: f4 90 80 80 5 byte nonsense: f9 84 80 80 80 split surrogates: ed a0 80 ed bf bf overbyte split surrogates: f0 8d a0 80 f0 8d bf bf
This is just a benchmark example to see if the hot loop can be written branch-free using RISC-V BitManip instructions. Edge cases can be handled in branches that are __builtin_expected() to be not taken.