[patch, so_7] : search_n optimization

Paolo Carlini pcarlini@suse.de
Fri Sep 2 23:22:00 GMT 2005


random@bubblescope.net wrote:

>After quite a long time, an optimized version of search_n :)
>  
>
Great!

>Blame me for any mistakes in the formatting and delegation, Dimitris for
>any mistakes in the optimised random_access algorithm :) Already plenty
>of testcases for this, hence the lack of testcase :) This algorithm can
>be very slightly (~5%) slower in some cases, but is almost always much
>faster.
>  
>
Agreed. And we already have a performance testcase, I will double check
the numbers I get on a couple of modern machines.

>This patch is against so_7. If it's too late in the day to merge the
>large-scale changes to stl_algo.h and friends to mainline, but there is
>still time to submit this patch, I'll knock up a mainline version to
>(which would just require duplicating the code for the non-predicated
>version as well).
>  
>
Thanks. Let's apply the patch first to v7-branch. At this point, I don't
think the move semantics stuff and the large scale changes to stl_algo.h
can go in in time for 4.1. Anyway, beginning of next week I will come to
your last move semantics patch and we can figure out the details.
Roughly, I'm confident that this patch and some other small improvements
can be in 4.1, then, as soon as 4.1 branches, we can apply first the
largely mechanical changes to algo.h, then proceed with move semantics:
ideally, it would be nice if we could cover consistently all the containers.

Right now, I'm trying to figure out what's going wrong with 23667. I'm
really nervous because of it!

Paolo.



More information about the Libstdc++ mailing list