[LLVMdev] On LLD performance

Davide Italiano davide at freebsd.org
Tue Mar 17 22:30:46 PDT 2015


On Tue, Mar 17, 2015 at 10:20 PM, Bruce Hoult <bruce at hoult.org> wrote:
> With a 4 second difference between best and worst runs in both cases, and
> only 0.2 second difference between the best for the two different cases
> (user time), I don't think you can make any conclusion that one is better
> than the other! They are very very similar.
>

I wasn't (unfortunately) able to control that variance :(
Reading the results again, I tend to agree though that the difference
might be noise. I also think that the change per se doesn't introduce
any benefit so i'm less inclined to keep going for this route. I don't
exclude this might have some utility though, that's why I attached the
patch to the mail, in case somebody wants to play with it a bit more.

--
Davide



More information about the llvm-dev mailing list