http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50160
--- Comment #12 from Jonathan Schmidt-Dominé 2011-08-23
19:31:37 UTC ---
“Any particular reason?”
No particular one, but a small specialisation would be nicer than changing
everything or doing bitfiddling.
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50160
--- Comment #11 from Marc Glisse 2011-08-23
19:16:40 UTC ---
(In reply to comment #8)
> For my application I should simply use an unordered_map and all the overhead
> has gone. :D
Great.
> “I haven't thought about the potential drawbacks of imp
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50160
--- Comment #10 from Jonathan Schmidt-Dominé 2011-08-23
19:15:20 UTC ---
There seem to be a lot of tricks to achieve that:
http://graphics.stanford.edu/~seander/bithacks.html#ReverseByteWith64BitsDiv
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50160
--- Comment #9 from Jonathan Schmidt-Dominé 2011-08-23
19:10:06 UTC ---
Hmm, reversing really is not that nice.
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50160
--- Comment #8 from Jonathan Schmidt-Dominé 2011-08-23
19:04:25 UTC ---
For my application I should simply use an unordered_map and all the overhead
has gone. :D But operator< simply should not be that slow.
“I haven't thought about the potentia
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50160
--- Comment #7 from Marc Glisse 2011-08-23
18:35:57 UTC ---
If I understand correctly, operator< is supposed to give a lexicographic order,
and vector stores {true,false,false} as 1 and {false,false,true} as 4, so
we can't just make operator< com
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50160
--- Comment #6 from Jonathan Schmidt-Dominé 2011-08-23
18:29:19 UTC ---
Ok, I can write a patch and test if it is actually faster (it
should). Is there any coding-guide line (style, variables etc.)?
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50160
--- Comment #5 from Jonathan Schmidt-Dominé 2011-08-23
18:26:48 UTC ---
In libc++ it seems to be the same, specialised hashing, but no specialised
comparison. Ok, I can write a patch and test if it is actually faster (it
should). Is there any cod
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50160
--- Comment #4 from Jonathan Schmidt-Dominé 2011-08-23
18:21:36 UTC ---
In libc++ it is the same, specialised hash, but no specialised operator<.
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50160
--- Comment #3 from Paolo Carlini 2011-08-23
16:31:40 UTC ---
I meant: which improvement do you expect, in practice? I supposed you are aware
if other implementations of the library already including the optimization. I'm
looking for some concret
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50160
--- Comment #2 from Jonathan Schmidt-Dominé 2011-08-23
13:50:09 UTC ---
Well, it made my specific application very slow, when using vector as
key_type for (ordered) map. What kind of numbers would be usefull? I think it
should simply have a spec
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50160
--- Comment #1 from Paolo Carlini 2011-08-23
08:14:06 UTC ---
Do you have some numbers? Of course we can make progress on this, but we'd like
to know when it's good enough.
12 matches
Mail list logo
