On: Transcoding Latin 1 strings to UTF-8 strings at 12 GB/s using AVX-512

[xiphias2]

Another option would be a vector of 256 16 bit entries and keeping the pointer advance vector as you suggested.

[jojobas → xiphias2]

That would work, the only problem in either solution is double-writing 8 of the 16 bits on sequential ASCII characters. Might give it a try.

[xiphias2 → jojobas]

That shouldn't be a problem, they all go to the store buffer of the CPU, which should be able to handle 1 write/cycle.