Is the mmap interface a possibility? You could bulkget an array for each thread. I guess that would probably not gain much against just copying an array for each thread.
I couldn't find a way to create a subvector of a java array. Chris On Jan 21, 12:20 pm, "Mark H." <mark.hoem...@gmail.com> wrote: > On Jan 21, 8:56 am, Perry Trolard <trol...@gmail.com> wrote: > > > If I understand what you did correctly, the reason it worked for you > > is that Rich committed a fix to SVN in the meantime! > > Awesome, real-time fix! > > > > The last (count result-p) call takes a few seconds (probably > > > because there's a one-to-one mapping of tasks to sequence > > > elements) but it finishes. > > > That's right, just for the record, using pmap to map inc across a > > collection of integers -- or to do any similarly low-cost computation > > -- is a *bad* idea. Map will always be faster -- in my casual tests > > with inc, pmap is slower by an order of magnitude. (But pmap should > > still compute the value, & inc was an easy way to make a reduction for > > bug reporting...) > > *nods* if you want parallelism for a low-cost computation with pmap, > then it might be better to divide a Java vector into subvectors and > map over chunks of ranges. However, it's not immediately obvious how > to do this. For example, on my fairly modern dual-core machine: > > * I have a function make-test-array that produces a filled-in array of > doubles > * areduce-part is like areduce, but works on a range of vector > indices, rather than a vector > > user> (let [v (make-test-array 1000000)] (time (+ (areduce-part v 0 > 500000) (areduce-part v 500000 > 1000000)))) > "Elapsed time: 6.454977 > msecs" > 4.999995E11 > user> (let [v (make-test-array 1000000)] (time (reduce + (pmap (fn > [[start end]] (areduce-part v start end)) [[0 500000] [500000 > 1000000]])))) > "Elapsed time: 9.638913 > msecs" > 4.999995E11 > > I know that "time" isn't counting test array creation time: > > (let [v (. Thread sleep 1000)] (time > 42)) > "Elapsed time: 0.020304 > msecs" > 42 > > It could be that the two threads are contending over the Array object > reference (Java arrays aren't pointers). Is there a nice way to > create "subvector" objects that only reference the underlying memory > and not the parent array object? > > mfh --~--~---------~--~----~------------~-------~--~----~ 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 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 -~----------~----~----~----~------~----~------~--~---
