On Monday, February 27, 2012 23:55:39 Dmitry Olshansky wrote: > On 27.02.2012 23:18, Jonathan M Davis wrote: > > On Monday, February 27, 2012 10:55:12 Ali Çehreli wrote: > >> On 02/27/2012 08:29 AM, Tyler Jameson Little wrote: > >>> I didn't want to do subclassing, because my parser is a state-machine > >> > >> style > >> > >>> parser, so it's in a big switch. Pretty gross, but I would like it to > >> > >> be as > >> > >>> fast as possible. That's why I thought this model would be so cool, > >> > >> because > >> > >>> I could remove conditions from the generated code, and get rid of a > >> > >> lot of > >> > >>> the conditionals. > >> > >> I am pretty sure switch statements boil down to a sequence conditionals > >> consisting of equality comparisons. > >> > >> I know that some compilers use optimizations where the comparisons are > >> converted to a single lookup, but last I checked, dmd does not apply > >> that optimization. Perhaps because it's not implemented yet, or because > >> using a table lookup might be slower because of reaching outside of the > >> cpu cache. (Or another reason that I am not aware of. :)) > > > > Yeah. In theory, switch statements can be optimized better than if-else > > chains, and eventually I'd fully expect that dmd would do that, but right > > now, I don't think that they really are. You'd have to look at the > > generated assembly though to be sure though. > > Please stop spreading the misconception that they are going to be > optimized "probably and eventually". Switch statements over integers are > much better then if/else chains and were for something like 20 years. > They do produce nice improvements over a sequence of if/else because > values are known in advance and are bounded. > So it goes like this: [snip] > And for the record I seen all of that in assembly listings produced by > dmd, when optimizing VM dispatch in std.regex. I had to revisit opcodes > layout a bit so that dmd outputs jump table and not a binary-searching > like stuff in a tight loop.
I'm not saying that dmd doesn't ever optimize switch statements. I'm just saying that as I understand it, it doesn't always do so (its performance with ranged case statements isn't great for instance). Odds are that it _does_ optimize straight integer case statements fairly well, because that's the classic C stuff (and if you've verified that, all the better - I haven't). It's the new stuff that's less likely to be well-optimized (like ranged case statements or case statements with strings). And those probably _will_ be better optimized eventually, but as I understand it, they aren't now. So, if the reason for using a switch statement over and an if-else chain is purely for performance (particularly if doing something else than just using plain case statements with integers like in C), then you should probably profile your code and/or look at the assembly to be sure that you're getting the performance gain that you think that you're getting. - Jonathan M Davis
