On Monday, 26 November 2012 at 07:43:30 UTC, monarch_dodra wrote:
Also, druntime has a *radically* different approach to striding
UTF-8. I'll try to see which approach is faster.
Well, my (quick) benches showed that unless the input is
something like 90% multibyte unicode, std.utf's stride
implementation beats the crap out of rc.util.utf's.
Depending on compile options (-inline/-noboundscheck):
Given an ASCII string, then phobo's implementation is anywhere
from 100% to 500% faster.
Given a unicode only string, then UTF's is actually 30% faster.
If anybody else wants to have a go at it...
I'll go ahead and update rc.util.utf's implementation to match
phobo's (minus @jmdavis' support for ranges).
