On 22/04/2021 19:54, Bernhard Voelker wrote:
On 4/21/21 8:13 PM, Assaf Gordon wrote:
I'm attaching the patch (copied from the Github's pull-request),
hopefully we can continue the discussion here on the mailing list.
Phew, so I see it's about getting 0-5x better performance versus decreased
maintainability
due having yet another implementation of the counting in wc: we already have
different cases
for multi-byte, for !count_chars etc.
I'm not sure whether it's worth complicating the tool, and - as Rasmus
mentioned - the
bottleneck is probably somewhere else.
Yes a fair point,
but the gains are worth it IMHO.
The awkward maintenance issue with separated flags in separate libs
is already developed and in place.
Going forward I think there will be a less homogenous landscape of CPUs,
so believe this will be of future benefit.
I plan to work on wc-avx2 this weekend.
cheers,
Pádraig