This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: [PATCH v2] Improve performance of strstr
On Thu, Apr 18, 2019 at 04:37:41PM +0000, Wilco Dijkstra wrote:
> Hi,
>
> > 40x happens on an input where twoway is fast and new algo is slow,
> > there are examples the other way around.
>
> On the worst cases I tested Two-way is 78 times slower than my new
> algorithm. It's trivial to find even worse factors, but it's not a useful
> comparison given each algorithm will have different worst cases, and
> they differ with each microarchitecture.
Unless you have concrete suggestions to improve the two-way
implementation you want to discuss, I'm done with this thread for now.
It's an endless time-sink/bikeshed to "fix" a non-problem with a
fundamentally bad algorithm with bad performance properties.
If glibc folks want to adopt this bad code, that's on you guys. I've
made it clear what directions I'd like to see pursued instead, if
there's an actual problem that needs to be fixed. But if there's not,
this is all a colossal waste of time.
Rich