https://bugs.kde.org/show_bug.cgi?id=397083
--- Comment #8 from Paul Floyd <pjfl...@wanadoo.fr> ---
It was just a copy and quick adaptation of the memchr representation. There are
other issues - the byte count isn't right either.
There's a lot less benefit to using a loop of UWord steps since UWord is only
2x the size of wchar_t (whilst for memchr it is 8x the size of char).
Looking at a couple of libc implementations, FreeBSD keeps it very simple, no
real effort to optimize.
wmemcmp(const wchar_t *s1, const wchar_t *s2, size_t n)
{
size_t i;
for (i = 0; i < n; i++) {
if (*s1 != *s2) {
/* wchar might be unsigned */
return *s1 > *s2 ? 1 : -1;
}
s1++;
s2++;
}
return 0;
}
GNU libc does try harder - though that also has to work on Windows where
wchar_t is 16bits where I expect there will be more benefit.
Code is a bit longer so here's the link. In short, an unrolled loop doing 4
whar_ts at a time followed by 3 ifs to handle the remainder.
https://elixir.bootlin.com/glibc/glibc-2.37/source/wcsmbs/wmemcmp.c#L25
--
You are receiving this mail because:
You are watching all bug changes.
