Table of Contents
Library for playing around with low level Clojure code for performance reasons given some assumptions. Inspired by Naked Performance (with Clojure) – Tommi Reiman.
Some of the code is based on implementations in metosin's projects. Credit in code.
This repo serves a dual purpose:
- Providing faster implementations of Clojure's core functions as macros.
- Reference guide on the performance characteristics of different ways of using Clojure's data structures.
What makes it possible?
Plenty of Clojure's core functions are implemented to be generic (good) and to accept a variable number of arguments (also very good). The problem is that we pay for this in performance. Wherever we iterate over a sequence of input arguments or dispatch on the class, we lose on performance, especially when iterating on arguments and calling
Plenty of these behaviors are just forms of flow-control, and like
or, other forms of flow control can too be statically analyzed, under certain constraints, and replaced by faster code.
See results.md for experiments' detailed benchmark results.
Add in your
(require '[clj-fast.core :as fast])
entry-at: used like
find but doesn't dispatch and has inline definition. Works for
val-at: used like
get but doesn't dispatch and has inline definition. Works for
fast-assoc: Used like
assoc but doesn't take variable key-values, only one pair and has inline definition. Works on
fast-map-merge: Slightly faster version for
merge, takes only 2 maps.
rmerge!: merges a map into a transient map.
(require '[clj-fast.inline :as inline])
Like regular core functions but sequence arguments must be written explicitly for static analysis or
defed in advance (i.e.
(def ks [:a :b])
(inline/assoc m :a 1 :b 2)
(inline/dissoc m :a :b)
(inline/fast-assoc m :a 1 :b 2)
(inline/get-in m ks)
(inline/get-in m [:c :d])
(inline/get-some-in m [:c :d])
(inline/assoc-in m [:c :d] foo)
(inline/update-in m [:c :d] inc)
(inline/select-keys m [:a :b :c])
(inline/merge m1 m2 m3)
(def assoc* (inline/memoize-c 3 assoc))
- Merge analysis unrolls inline maps as well.
fast-assoc: inlines in the same manner of
assoc but uses
fast-map-merge: inlines in the same manner of
merge but uses
clj-fast.core/fast-map-merge instead (Metosin).
select-keys, but faster and dirtier, adds nil entries to the results map.
get-in at the expense of working only on callables (objects implementing
get-some-in but returns a map-entry in the end, like
memoize-c*: Alternative implementations for memoization using a nested Clojure hash map and a nested Java concurrent hash map respectively. Fall back to
core/memoize for large arities. Due to the cost of hashing objects in Clojure, it's recommended to use
memoize-c* for most use cases.
nth macros operate similarly to their respective functions with one notable difference: When provided with an appropriate type hint, they will dispatch to the underlying method at compile time instead of run time.
(def arr (long-array [1 2 3]))
(nth ^longs arr 0)
(def m (doto (java.util.HashMap.) (.put :a 1)))
(get ^Map m :a)
(require '[clj-fast.collections.hash-map :as hm])
get: wraps method call for
get. Has inline definition.
put: wraps method call for
put. Has inline definition.
(require '[clj-fast.collections.concurrent-hash-map :as chm])
concurrent-map?: instance check.
get?: get if is a concurrent hash map.
get-in?: like clojure core's get-in but for nested concurrent hash maps.
put-in!: like clojure core's assoc-in but for nested concurrent hash maps.
(require '[clj-fast.lens :as lens])
In typed functional programming, lenses are a generic way of getting and setting nested data structures (records).
In this context, the
lens namespace implements the basic code structure underlying Clojure's
update-in. They can be used in macros to expand to real code provided an appropriate 1-depth
put transformer, which takes arguments and returns an expression.
For example, the
get-some lens is used to define
(lens/get-some (fn [m k] `(~m ~k)) m ks))
[m ks v]
(fn [m k v] `(c/assoc ~m ~k ~v))
(fn [m k] `(c/get ~m ~k))
So be careful, these are not functional programming lenses, but metaprogramming lenses used for code generation.
clj-fast.clojure.core contains drop-in replacement functions and macros for Clojure's core.
It opportunistically replaces functions by their inlined implementations. It also includes binding macros (let, fn, loop, defn) which will use inlining versions of
nth when possible. (i.e. when type-hinted).
[get nth assoc get-in merge assoc-in update-in select-keys memoize destructure let fn loop defn defn-])
[clojure.core :as c]
[clj-fast.clojure.core :refer [get nth assoc get-in merge assoc-in update-in select-keys memoize destructure let fn loop defn defn-]]))
(Structural)[https://github.com/joinr/structural] is a small library by joinr (Tom) which provides more efficient destructuring macros with type hints.
(Stringer)[https://github.com/kumarshantanu/stringer] is a library by Shantanu Kumar for fast string operations. Of interest are the capabilities it provides in faster string building and formatting, also by "unrolling" the building operations where statically possible.
Copyright © 2019-2020 firstname.lastname@example.org
Copyright © Rich Hickey for the implementation in
This program and the accompanying materials are made available under the terms of the Eclipse Public License 2.0 which is available at http://www.eclipse.org/legal/epl-2.0.
This Source Code may also be made available under the following Secondary Licenses when the conditions for such availability set forth in the Eclipse Public License, v. 2.0 are satisfied: GNU General Public License as published by the Free Software Foundation, either version 2 of the License, or (at your option) any later version, with the GNU Classpath Exception which is available at https://www.gnu.org/software/classpath/license.html.
Credit to Metosin wherever noted in the code.
Rich Hickey for clojure.core ns.