junkdog / Bitvector

BitVector

• Uncompressed, dynamically resizeable bitset, similar to java.util.BitSet
• Fast-ish enumeration of set bits (JS not yet profiled)
• Compatible with JavaScript and JVM/Android backends
  • 4 byte words, avoids Long emulation in js

The gist

Constructing

val bv: BitVector = bitsOf(1, 2, 56, 64, 128, 129, 130, 131, 420)

Individual bits

val bv = BitVector()
bv[142] = true // or bv.set(142)
assert(142 in bv)

bv.clear(142)  // or bv[142] = false
assert(142 !in bv)

or, and, andNot, xor

As with java.util.BitSet, these operations mutate the callee. Do a copy() if the original BitVector needs to stick around.

These functions are not infix functions, as such syntax would suggest a value copy.

val a = bitsOf(0, 1, 2, 3, 120,                130)
val b = bitsOf(0, 1, 2,    120, 121, 122, 123, 130)

assert(a.and(b) == bitsOf(0, 1, 2, 120, 130))

// caveat: bitvector was mutated above
assert(a == bitsOf(0, 1, 2, 120, 130))

vs other BitVectors

// 1, 4 contained in other 
bitsOf(1, 4) in bitsOf(0, 1, 4, 5, 6) 

Looping over bits: fast way

bitsOf(*bits).forEachBit { println("bit $it says hi") }

Looping over bits: Iterable

// can be combined with filter/map etc
bitsOf(*bits).forEach { println("bit $it says hi") }

Getting Started

Maven: JVM/Android

<dependency>
	<groupId>net.onedaybeard.bitvector</groupId>
	<artifactId>bitvector-jvm</artifactId>
	<version>0.1.4</version>
</dependency>

Maven: JavaScript

<dependency>
	<groupId>net.onedaybeard.bitvector</groupId>
	<artifactId>bitvector-js</artifactId>
	<version>0.1.4</version>
</dependency>

Gradle: JVM/Android

  dependencies { compile "net.onedaybeard.bitvector:bitvector-jvm:0.1.4" }

Gradle: JavaScript

  dependencies { compile "net.onedaybeard.bitvector:bitvector-js:0.1.4" }

JVM Benchmarks / enumerating set bits

Mapping true bit positions into integers, inserting each into an IntBag (thin wrapper around int[]).

artemis-odb's BitVector was the basis for this implementation. The benchmark setup favors the artemis implementation, as it provides an optimized toIntBag method: it serves as a reference for best possible performance.

See jmh-logs for the full logs.

Discrepancy to artemis' BitVector is unwelcome. The implementation is for the most part the same, except that this implementation uses int for words, instead of long. 4 or 8 byte words did not have a significant impact on performance.

The for-loop performs poorly due to all the Integer boxing, extra indirection and allocation, compared to forEachBit.

java.util.BitSet suffers from not having a way of enumerating all bits at once, instead relying on repeatedly calling nextSetBit.

Tests

Verifying platform-specific implementations

mvn clean install -P impltest

Running JS tests

Build project, mvn clean install, then point the browser to file://$PROJECT_ROOT/js/target/test-classes/index.html