This is the mail archive of the
libstdc++@gcc.gnu.org
mailing list for the libstdc++ project.
Re: tr1::hashtable::operator[]
Peter Doerfler wrote:
The performance of your patch is the same as with mine for my testcase
(which I find surprising, I thought there would be a slight benefit
from the insert with hint).
Well, it depends on a lot of factors. For example, for the default
allocator we are spending a big fraction of the total time allocating
memory. Also, it depends on the load factor: when it's low the time
spent in find_node is smaller (thus the expected advantage smaller).
I'm attaching a small artificial testcase that shows the effect. It's
more drastic than my real use as the old implementation takes over a
minute to complete against below 3 seconds for either patch.
Thanks a lot for the testcase and for your feedback!
Paolo.