Using getComponentType, it appears to be handling different primitive array
types ok:
(defn fconcat [& arrays]
(let [sizes (map count arrays)
sizes_r (vec (reductions + sizes))
offsets (cons 0 (drop-last sizes_r))
total (last sizes_r)
out (make-array (.getComponentType (class (first arrays))) total)]
(dorun (map #(System/arraycopy %2 0 out %1 %3) offsets arrays sizes))
out))
On Sunday, July 21, 2013 12:26:26 PM UTC-7, Brian Craft wrote:
>
> (make-array (.getComponentType (class arr)) n) seems to work.
>
> On Sunday, July 21, 2013 12:22:41 PM UTC-7, Brian Craft wrote:
>>
>> Is there a way to create an array with the type of another array? (type
>> arr) returns the array type, but make-array wants the element type not the
>> array type, so
>>
>> (make-array (type arr) n)
>>
>> doesn't work as one might hope.
>>
>>
>> On Sunday, July 21, 2013 8:36:22 AM UTC-7, Alex Fowler wrote:
>>>
>>> Java's System.arraycopy is the fastest you can get, since it delegates
>>> execution to a function implemented in C inside JVM. Simply, this is the
>>> fastest that your computer hardware can get. All in all Java arrays meet
>>> the same difficulties and implications as C arrays and that is why
>>> concationation of raw arrays is so "complex", in contrast to higher-level
>>> collections which use objects and pointers (e.g. LinkedList). In other
>>> words, difficulties you experience are natural outcome of how computer's
>>> memory management is made and there is no way around them. You get the most
>>> of the speed from arrays because they are solid (not fragmented) chunks of
>>> bytes allocated in memory in the moment of their creation. For that very
>>> reason you cannot extend an existing array (the size cannot be changed
>>> after creation) and you can't concatenate it with another array since first
>>> it would have to be concatenated.
>>>
>>> The natural outcome also is that only arrays of same types can be
>>> concatenated with System.arraycopy since only array pointers store type
>>> data, and the contents are simply untyped bytes. And this is why it is
>>> byte-level and no type-checks are ever done besiedes the initial
>>> type-check. Again, higher-level pointer-based data structures like
>>> LinkedList or Queue can introduce boxed typed values, but that'd be waaay
>>> slower. Considering that only arrays of same type are concatenateable,
>>> creating a polymorphic function is easy - simply check the argument type
>>> like:
>>>
>>> ; first save types to use them later
>>> (def arr-type-int (class (ints 3)))
>>> ; ... same for other primitives...
>>>
>>> ; then in your func:
>>> (cond
>>> (= (class arr) arr-type-int) (do-int-concat)
>>> ...)
>>>
>>> For more reference:
>>> http://docs.oracle.com/javase/tutorial/java/nutsandbolts/arrays.html
>>> http://docstore.mik.ua/orelly/java-ent/jnut/ch02_09.htm
>>>
>>> As an alternative, try looking into Java NIO buffers - they too are fast
>>> and too have some limits. But maybe you could make good of them, depends on
>>> your use case.
>>>
>>> Although somewhat in another vein, but still relating fast data
>>> management is
>>> https://groups.google.com/forum/?hl=en#!topic/clojure/BayfuaqMzvs which
>>> brings in C-like structs in.
>>>
>>> On Sunday, July 21, 2013 2:39:38 AM UTC+4, Brian Craft wrote:
>>>>
>>>> Here are some experiments that aren't polymorphic. The System/arraycopy
>>>> version is fastest, by far. Is there any good way to make the other
>>>> versions faster, or make them handle any array type?
>>>>
>>>> (defn bconcat [& arrays]
>>>> (let [sizes (map count arrays)
>>>> sizes_r (vec (reductions + sizes))
>>>> offsets (cons 0 (drop-last sizes_r))
>>>> total (last sizes_r)
>>>> out (float-array total)]
>>>> (dorun (map #(System/arraycopy %2 0 out %1 %3) offsets arrays sizes))
>>>> out))
>>>>
>>>> (defn cconcat [& arrays]
>>>> (let [vs (map vec arrays)
>>>> cc (apply concat vs)]
>>>> (float-array cc)))
>>>>
>>>> (defn dconcat [& arrays]
>>>> (let [vs (map vec arrays)
>>>> cc (reduce into [] vs)]
>>>> (float-array cc)))
>>>>
>>>> (defn econcat [& arrays]
>>>> (let [cc (reduce into [] arrays)]
>>>> (float-array cc)))
>>>>
>>>>
>>>> On Saturday, July 20, 2013 2:24:14 PM UTC-7, Brian Craft wrote:
>>>>>
>>>>> Is there an easy, fast way to concat primitive arrays? I was hoping
>>>>> java arrays had some common interface for this, but I haven't found much
>>>>> of
>>>>> use. I mostly see code like this:
>>>>>
>>>>> byte[] c = new byte[a.length + b.length];
>>>>> System.arraycopy(a, 0, c, 0, a.length);
>>>>> System.arraycopy(b, 0, c, a.length, b.length);
>>>>>
>>>>> which only works for bytes (in this case).
>>>>>
>>>>
--
--
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/groups/opt_out.
