奇怪的aget优化行为 [英] Strange aget optimisation behavior

问题描述

关于 aget性能的问题的后续跟进

似乎有一些非常奇怪的优化进行。我们知道以下是真实的：

There seems to be something very strange going on optimisation wise. We knew the following was true:

=> (def xa (int-array (range 100000)))
#'user/xa

=> (set! *warn-on-reflection* true)
true

=> (time (reduce + (for [x xa] (aget ^ints xa x))))
"Elapsed time: 42.80174 msecs"
4999950000

=> (time (reduce + (for [x xa] (aget xa x))))
"Elapsed time: 2067.673859 msecs"
4999950000
Reflection warning, NO_SOURCE_PATH:1 - call to aget can't be resolved.
Reflection warning, NO_SOURCE_PATH:1 - call to aget can't be resolved.

但是，一些进一步的试验真的让我很奇怪：

However, some further experimenting really weirded me out:

=> (for [f [get nth aget]] (time (reduce + (for [x xa] (f xa x)))))
("Elapsed time: 71.898128 msecs"
"Elapsed time: 62.080851 msecs"
"Elapsed time: 46.721892 msecs"
4999950000 4999950000 4999950000)

无反射警告，无需提示。通过将aget绑定到根var或在let中可以看到相同的行为。

No reflection warnings, no hints needed. Same behavior is seen by binding aget to a root var or in a let.

=> (let [f aget] (time (reduce + (for [x xa] (f xa x)))))
"Elapsed time: 43.912129 msecs"
4999950000

任何想法为什么绑定aget似乎知道如何优化，核心功能不在哪里？

Any idea why a bound aget seems to 'know' how to optimise, where the core function doesn't ?

推荐答案

它与：inline 指令 / code>，它扩展为（。clojure.lang.RT（aget〜a（int〜i）），而正常函数调用涉及 Reflector 。尝试这些：

It has to do with the :inline directive on aget, which expands to (. clojure.lang.RT (aget ~a (int ~i)), whereas the normal function call involves the Reflector. Try these:

user> (time (reduce + (map #(clojure.lang.Reflector/prepRet 
       (.getComponentType (class xa)) (. java.lang.reflect.Array (get xa %))) xa)))
"Elapsed time: 63.484 msecs"
4999950000
user> (time (reduce + (map #(. clojure.lang.RT (aget xa (int %))) xa)))
Reflection warning, NO_SOURCE_FILE:1 - call to aget can't be resolved.
"Elapsed time: 2390.977 msecs"
4999950000

那么，你可能会想知道内联的意义。好，看看这些结果：

You might wonder what's the point of inlining, then. Well, check out these results:

user> (def xa (int-array (range 1000000))) ;; going to one million elements
#'user/xa
user> (let [f aget] (time (dotimes [n 1000000] (f xa n))))
"Elapsed time: 187.219 msecs"
user> (time (dotimes [n 1000000] (aget ^ints xa n)))
"Elapsed time: 8.562 msecs"

事实证明，在你的例子中，一旦你遇到反射警告，你的新瓶颈是 reduce + 部分而不是数组访问。此示例消除了这一点，并显示了类型提示的内联 aget 的有序数量级优势。

It turns out that in your example, as soon as you get past reflection warnings, your new bottleneck is the reduce + part and not array access. This example eliminates that and shows an order-of-magnitude advantage of the type-hinted, inlined aget.

这篇关于奇怪的aget优化行为的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！