http://dev.clojure.org/jira This file is an XML representation of an issue en-us 4.4 649 25-07-2011 http://dev.clojure.org/jira/browse/CLJ-1000 Clojure <p>It seems there is a 30-40% performance degradation of PersistentHashMap.valAt(...) in Clojure 1.4.<br/> Possibly due to references to new CPU-hungry implementation of Util.hasheq(...).</p> <p>I have created a demo project with more details and some profiling information here:<br/> <a href="https://github.com/oshyshko/clj-perf">https://github.com/oshyshko/clj-perf</a></p> Java(TM) SE Runtime Environment (build 1.7.0_04-b21) <br/> Java HotSpot(TM) 64-Bit Server VM (build 23.0-b21, mixed mode) CLJ-1000

Performance drop in PersistentHashMap.valAt(...) in v.1.4 -- Util.hasheq(...) ?

Enhancement Major Closed Completed Timothy Baldridge Oleksandr Shyshko performance Mon, 21 May 2012 22:06:54 -0500 Fri, 1 Mar 2013 09:49:21 -0600 Tue, 11 Dec 2012 11:14:49 -0600 Release 1.4 Release 1.5 1 2 <p>I added a patch consisting of three commits:</p> <ul> <li>one which adds caching to seqs, sets, maps, vectors and queues</li> <li>one that aligns the shape of Util/hasheq on the one Util/equiv (to have a consistent behavior from the JIT compiler: without that deoptimization was more penalizing for hasheq than for equiv)</li> <li>one that fix hasheq on records (which was non consistent with its equiv impl.) &#8211; and this commit relies on a static method introduced in the "caching hasheq" commit</li> </ul> <p>In the process of screening this, I'm not seeing much of a performance difference after applying the patch. </p> <p>before patch:<br/> user=&gt; (-main)</p> <p>Version: 1.5.0-master-SNAPSHOT<br/> "Elapsed time: 6373.752 msecs"<br/> "Elapsed time: 6578.037 msecs"<br/> "Elapsed time: 6476.399 msecs"</p> <p>after patch:<br/> user=&gt; (-main)</p> <p>Version: 1.5.0-master-SNAPSHOT<br/> "Elapsed time: 6182.699 msecs"<br/> "Elapsed time: 6548.086 msecs"<br/> "Elapsed time: 6496.711 msecs"</p> <p>clojure 1.4:<br/> user=&gt; (-main)</p> <p>Version: 1.4.0<br/> "Elapsed time: 6484.234 msecs"<br/> "Elapsed time: 6243.672 msecs"<br/> "Elapsed time: 6248.898 msecs"</p> <p>clojure 1.3<br/> user=&gt; (-main)</p> <p>Version: 1.3.0<br/> "Elapsed time: 3584.966 msecs"<br/> "Elapsed time: 3618.189 msecs"<br/> "Elapsed time: 3372.979 msecs"</p> <p>I blew away my local clojure repo and re-applied the patch just to make sure, but the results are the same. Does this fix not optimize the case given in the original test project? </p> <p>For reference I'm running this code:</p> <p>(defn -main<br/> <span class="error">&#91;&amp; args&#93;</span></p> <p> (println)<br/> (println "Version: " (clojure-version))</p> <p> (def mm 10000)</p> <p> (def str-keys (map str (range mm)))<br/> (def m (zipmap str-keys (range mm)))<br/> (time (dotimes <span class="error">&#91;i mm&#93;</span> (doseq <span class="error">&#91;k str-keys&#93;</span> (m k))))</p> <p> (def kw-keys (map #(keyword (str %)) (range mm)))<br/> (def m (zipmap kw-keys (range mm)))<br/> (time (dotimes <span class="error">&#91;i mm&#93;</span> (doseq <span class="error">&#91;k kw-keys&#93;</span> (m k))))</p> <p> (def sym-keys (map #(symbol (str %)) (range mm)))<br/> (def m (zipmap sym-keys (range mm)))<br/> (time (dotimes <span class="error">&#91;i mm&#93;</span> (doseq <span class="error">&#91;k sym-keys&#93;</span> (m k))))</p> <p> (println))</p> <p>Sorry, I was too quick to react on the ML (someone said it was related to hasheq caching and since I had the patch almost ready: on a project I noticed too much time spent computing hasheq on vectors).<br/> So no, my patch doesn't improve anything with kws, syms or strs as keys. However when keys are collections, it fares better.</p> <p>In 1.4, for a "regular" object, it must fails two instanceof tests before calling .hashCode().<br/> If we make keywords and symbols implement IHashEq and reverse the test order (first IHashEq, then Number) it should improve the performance of the above benchmark &#8211; except for Strings.</p> <p>Marking as incomplete, should we also delete the patch as it seems like it should be in a different ticket?</p> <p>In 1.3, #'hash was going through Object.hashCode and thus was simple and fast. Plus collections hashes were cached.<br/> In 1.4 and master, #'hash goes now through two instanceof test (Number and IHasheq in this order) before trying Object.hashCode in last resort. Plus collections hashes are not cached.<br/> As such I'm not sure Util.hasheq inherent slowness (compared to Util.hash) and hasheq caching should be separated in two issues.</p> <p>The caching-hasheq-v2.diff patchset reintroduces hashes caching for collections/hasheq and reorders the instanceof tests (to test for IHashEq before Number) and makes Keyword and Symbol implement IHashEq to branch fast in Util.hasheq.</p> <p>I recommend adding a collection test to the current benchmark:</p> <div class="code panel" style="border-width: 1px;"><div class="codeContent panelContent"> <pre class="code-java">(defn -main [&amp; args] (println) (println <span class="code-quote">"Version: "</span> (clojure-version)) (def mm 10000) (def str-keys (map str (range mm))) (def m (zipmap str-keys (range mm))) (time (dotimes [i mm] (doseq [k str-keys] (m k)))) (def kw-keys (map #(keyword (str %)) (range mm))) (def m (zipmap kw-keys (range mm))) (time (dotimes [i mm] (doseq [k kw-keys] (m k)))) (def sym-keys (map #(symbol (str %)) (range mm))) (def m (zipmap sym-keys (range mm))) (time (dotimes [i mm] (doseq [k sym-keys] (m k)))) (def vec-keys (map (comp (juxt keyword symbol identity) str) (range mm))) (def m (zipmap vec-keys (range mm))) (time (dotimes [i mm] (doseq [k vec-keys] (m k)))) (println))</pre> </div></div> <p>For some reason I can't get v2 to build against master. It applies cleanly, but fails to build.</p> <p>Timothy: I inadvertently deleted a "public" modifier before commiting... fixed in caching-hasheq-v3.diff</p> <p>I now get the following results:</p> <p>Version: 1.4.0<br/> "Elapsed time: 6281.345 msecs"<br/> "Elapsed time: 6344.321 msecs"<br/> "Elapsed time: 6108.55 msecs"<br/> "Elapsed time: 36172.135 msecs"</p> <p>Version: 1.5.0-master-SNAPSHOT (pre-patch)<br/> "Elapsed time: 6126.337 msecs"<br/> "Elapsed time: 6320.857 msecs"<br/> "Elapsed time: 6237.251 msecs"<br/> "Elapsed time: 18167.05 msecs"</p> <p>Version: 1.5.0-master-SNAPSHOT (post-patch)<br/> "Elapsed time: 6501.929 msecs"<br/> "Elapsed time: 3861.987 msecs"<br/> "Elapsed time: 3871.557 msecs"<br/> "Elapsed time: 5049.067 msecs"</p> <p>Marking as screened</p> <p>Please, could you add as a comment the bench result using 1.3 vs 1.5-master-post-patch?</p> <p>The performance with 1.5-master is now very close to 1.3 for 3/4 of the benchmark.</p> <p>However, this code is still showing 43% performance drop (3411 ms VS 6030 ms &#8211; 1.3 VS 1.5-master):</p> <p>(def str-keys (map str (range mm)))<br/> (def m (zipmap str-keys (range mm)))<br/> (time (dotimes <span class="error">&#91;i mm&#93;</span> (doseq <span class="error">&#91;k str-keys&#93;</span> (m k))))</p> <p>Version: 1.3.0<br/> "Elapsed time: 3411.353 msecs" &lt;---<br/> "Elapsed time: 3459.992 msecs"<br/> "Elapsed time: 3365.182 msecs"<br/> "Elapsed time: 3813.637 msecs"</p> <p>Version: 1.4.0<br/> "Elapsed time: 5710.073 msecs" &lt;---<br/> "Elapsed time: 5817.356 msecs"<br/> "Elapsed time: 5774.856 msecs"<br/> "Elapsed time: 18754.482 msecs"</p> <p>Version: 1.5.0-master-SNAPSHOT<br/> "Elapsed time: 6030.247 msecs" &lt;---<br/> "Elapsed time: 3372.637 msecs"<br/> "Elapsed time: 3267.481 msecs"<br/> "Elapsed time: 3852.927 msecs"</p> <p>To reproduce:<br/> $ git clone <a href="https://github.com/clojure/clojure.git">https://github.com/clojure/clojure.git</a><br/> $ cd clojure<br/> $ mvn install -Dmaven.test.skip=true</p> <p>$ cd ..</p> <p>$ git clone <a href="https://github.com/oshyshko/clj-perf.git">https://github.com/oshyshko/clj-perf.git</a><br/> $ cd clj-perf<br/> $ lein run-all</p> Approval Ok Global Rank Patch Code