Just done some of my own quick benchmarking on this. (= :foo :bar) ;; 21.7 ns (.equals :foo :bar) ;; 3.0 ns (identical? :foo :bar) ;; 3.0 ns
(.equals "foo" "bar") ;; 5.6 ns (= "foo" "bar") ;; 24.4 ns This seems to support my suspicion that the overhead is in the extra overhead in clojure.core/=, rather than an intrinsic issue with keyword equality performance. On Saturday, 11 October 2014 18:41:12 UTC+8, Mikera wrote: > > I believe this is a symptom of the fact that the Clojure compiler isn't > very type aware, and = inserts a bunch of redundant runtime type checks > (see clojure.lang.Util.equiv(Object, Object) ) > > In particular, it always checks for whether the arguments are instances of > Number or IPersistentCollection first, because these get special handling, > before calling Keyword.equals(). Obviously, these checks are pointless if > you already know one or both arguments is a keyword, but the compiler > doesn't make that inference at compile time so we have to pay the runtime > overhead. These type checks are pretty cheap, but certainly not free. It's > also possible that the JVM isn't inlining as much as it could. > > On Saturday, 11 October 2014 06:23:10 UTC+8, Jony Hudson wrote: >> >> Hi All, >> >> I've been optimising a piece of code lately, and have come to wonder >> about the performance of keyword comparison. Specifically, I'm not sure >> whether the performance I'm seeing is what is expected. The data structures >> page on clojure.org [1] indicates that keywords "provide very fast >> equality tests". If I micro-benchmark with criterium, then I find the >> following: >> >> As a baseline, comparing integers with `(= 0 1)` takes around 4ns. >> >> Comparing keywords with `(= :plus :minus)` takes around 30ns. >> >> This is about the same amount of time it takes to compare strings, `(= >> "plus" "minus")`, which comes in at about 25ns. >> >> This surprised me, as I would have guessed that "fast" would have been >> closer to the integer performance than the string performance. It's worth >> saying that I don't know a lot about benchmarking, but I do have some >> "real" code that's performance depends heavily on comparisons, and it seems >> to line up performance-wise with these micro-benchmarks. >> >> So, am I doing something silly (like I don't know about the fast = for >> keywords)? Or, are my expectations wrong, and this is about how long "fast" >> should be? Or is there a performance bug lurking? >> >> I'm using Clojure 1.6.0 (but have tried 1.5.0 and 1.7.0-alpha1 with >> similar results). >> x86_64 Mac OS X 10.9.5 4 cpu(s) >> >> Java HotSpot(TM) 64-Bit Server VM 25.5-b02 >> >> Thanks in advance for any input, >> >> >> Jony >> >> [1] http://clojure.org/data%5Fstructures >> > -- You received this message because you are subscribed to the Google Groups "Clojure" group. To post to this group, send email to clojure@googlegroups.com Note that posts from new members are moderated - please be patient with your first post. To unsubscribe from this group, send email to clojure+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/clojure?hl=en --- You received this message because you are subscribed to the Google Groups "Clojure" group. To unsubscribe from this group and stop receiving emails from it, send an email to clojure+unsubscr...@googlegroups.com. For more options, visit https://groups.google.com/d/optout.
