Try replacing 'aset-byte' with 'aset' and hinting all literals (change 1 to (int 1), etc.) as per David's suggestions. That should reduce the gap even further.
On Jan 28, 11:31 am, Robert McIntyre <r...@mit.edu> wrote: > And the plot thickens: > > This: > > (defn convert-image [#^bytes cpuArray] > (let [unchecked-add clojure.core/unchecked-add > len (int (count cpuArray))] > (loop [i (int 0)] (if (< i len) > (let [ > b (byte (aget cpuArray i)) > g (byte (aget cpuArray (unchecked-add 1 > i))) > r (byte (aget cpuArray (unchecked-add 2 > i))) > a (byte (aget cpuArray (unchecked-add 3 > i)))] > (aset-byte cpuArray i a) > (aset-byte cpuArray (unchecked-add 1 i) b) > (aset-byte cpuArray (unchecked-add 2 i) g) > (aset-byte cpuArray (unchecked-add 3 i) r) > (recur (int (unchecked-add i 4)))))))) > > vs this. > > (defn convert-image [#^bytes cpuArray] > (let [len (java.lang.reflect.Array/getLength cpuArray)] > (loop [i (int 0)] (if (< i len) > (let [i2 (unchecked-add 1 i) > i3 (unchecked-add 2 i) > i4 (unchecked-add 3 i) > b (byte (aget cpuArray i)) > g (byte (aget cpuArray i2)) > r (byte (aget cpuArray i3)) > a (byte (aget cpuArray i4))] > (aset-byte cpuArray i a) > (aset-byte cpuArray i2 b) > (aset-byte cpuArray i3 g) > (aset-byte cpuArray i4 r) > (recur (unchecked-add i 4))))))) > > The first function takes forever; the second MUCH faster. > Upon disassembling the byte-code of the two compiled functions, it > does seem like the + was not being inlined. > Since the method of reassignment doesn't preserve the metadata, this > makes sense. > > However, my new, modified function is still around 20 times slower > than the java version :( > > I still don't understand what's slowing me down, but I'm much happier > that I can get within 20x of java instead of 20000x > > thanks everyone. > > sincerely, > --Robert McIntyre > > > > > > > > On Fri, Jan 28, 2011 at 1:07 PM, Ken Wesson <kwess...@gmail.com> wrote: > > On Fri, Jan 28, 2011 at 12:55 PM, Benny Tsai <benny.t...@gmail.com> wrote: > >> It seems that 'unchecked-add' returns a primitive (note that in the > >> first version, 'recur' happily accepts the return from 'unchecked-add' > >> without coercion), but when 'unchecked-add' is bound to a new name, > >> the return gets boxed. > > >> Is this the correct interpretation, or am I missing something? > > > It's correct. Clojure's compiler inlines certain functions, including > > clojure.core/unchecked-foo and, when the arity is 2, clojure.core/+ -- > > but not a bare, un-qualified +, which might (or might not) at runtime > > refer to clojure.core/+ or clojure.core/unchecked-add or whatever. > > > -- > > 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 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
