Hi Marshall, the megamorphic call site hypothesis does sound plausible but I'm not sure where the following test fits in. If I understand correctly we believe that it's the fact that the base case (an PersistentList$EmptyList instance) and the normal case (an PersistsentList instance) have different types and when run in paralell the interleaving invocations are causing problems for the JIT but when we use a vector it's a single type so we see a better speedup.
I was toying with the idea of replacing the EmptyList class with a PersistsentList instance to mitigate the problem in at least one common case, however it doesn't seem to help. If I replace the reverse call in burn with the following code: #(reduce conj (list nil) %) I get the same slowdown as we see if reverse (equivalent to #(reduce conj '() %)) where: (class '()) => clojure.lang.PersistentList$EmptyList (class (list nil)) => clojure.lang.PersistentList I'm not convinced were at the bottom of it yet, perhaps the earlier post on memory contention by Carlos might yield some clues. If the issue was related to memory access it might help explain why the impact differed significantly between intel and amd hardware. Cameron. -- 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
