Benchmark: unbiased ranged random integer generation functions

Script Preparation code:

const ta16 = new Uint16Array(1);
const ta32 = new Uint32Array(1);
const bound = 13000;

Tests:

Classic 16
if ((bound & -bound) === bound) { window.crypto.getRandomValues(ta16); return bound * ta16[0] >>> 16; } const limit = 65536 - 65536 % bound; do { window.crypto.getRandomValues(ta16); } while (ta16[0] >= limit); return ta16[0] % bound;
Mul shf 16
if ((bound & -bound) === bound) { window.crypto.getRandomValues(ta16); return bound * ta16[0] >>> 16; } const limit = (-bound >>> 0) % bound; let value; do { window.crypto.getRandomValues(ta16); value = ta16[0] * bound; } while ((value & 65535) < limit); return value >>> 16;
Imul shf 16
if ((bound & -bound) === bound) { window.crypto.getRandomValues(ta16); return bound * ta16[0] >>> 16; } const limit = (-bound >>> 0) % bound; let value; do { window.crypto.getRandomValues(ta16); value = Math.imul(ta16[0], bound); } while ((value & 65535) < limit); return value >>> 16;
Classic int
if ((bound & -bound) === bound) { window.crypto.getRandomValues(ta32); return ta32[0] * bound / 4294967296 >>> 0; } const limit = 4294967296 - 4294967296 % bound; do { window.crypto.getRandomValues(ta32); } while (ta32[0] >= limit); return ta32[0] % bound;
Mul shf bigint
if ((bound & -bound) === bound) { window.crypto.getRandomValues(ta32); return ta32[0] * bound / 4294967296 >>> 0; } const limit = (-bound >>> 0) % bound; let bits, mLow; let mHigh; do { window.crypto.getRandomValues(ta32); bits = ta32[0] >>> 1; const m = BigInt(bits) * BigInt(bound); mLow = Number(m & 2147483647n); mHigh = Number(m >> 31n); } while (mLow < limit); return mHigh;
Mul shf 32
if ((bound & -bound) === bound) { window.crypto.getRandomValues(ta32); return ta32[0] * bound / 4294967296 >>> 0; } const limit = (-bound >>> 0) % bound; let value; do { window.crypto.getRandomValues(ta32); value = ta32[0] * bound; } while ((value & 4294967295) % bound < limit); return value / 4294967296 >>> 0;
java nextInt16(bound)
if ((bound & -bound) === bound) { crypto.getRandomValues(ta16); return bound * ta16[0] >>> 16; } let value; do { crypto.getRandomValues(ta16); value = ta16[0] % bound; } while (ta16[0] - value + (bound - 1) < 0); return value;
java nextInt32(bound)
if ((bound & -bound) === bound) { window.crypto.getRandomValues(ta32); return ta32[0] * bound / 4294967296 >>> 0; } let bits, value; do { window.crypto.getRandomValues(ta32); bits = ta32[0] >>> 1; value = bits % bound; } while (bits - value + (bound - 1) < 0); return value;

Historical Performance Data aggregated from 42 community runs

Desktop Results:

Chrome 18 runs

Classic int

2.9M/s

java nextInt16(bound)

2.9M/s

java nextInt32(bound)

2.8M/s

Classic 16

2.8M/s

Imul shf 16

2.8M/s

Mul shf 16

2.8M/s

Mul shf bigint

2.1M/s

Mul shf 32

1.2M/s

Fastest: Classic int (2.4x faster)

View exact numbers

✓ Classic int: 2,909,317 ops/sec
• java nextInt16(bound): 2,854,368 ops/sec
• java nextInt32(bound): 2,796,083 ops/sec
• Classic 16: 2,788,343 ops/sec
• Imul shf 16: 2,787,498 ops/sec
• Mul shf 16: 2,764,021 ops/sec
• Mul shf bigint: 2,106,800 ops/sec
• Mul shf 32: 1,228,541 ops/sec

Firefox 7 runs

java nextInt16(bound)

2.0M/s

Classic 16

1.9M/s

Mul shf 16

1.7M/s

Imul shf 16

1.7M/s

java nextInt32(bound)

1.7M/s

Mul shf bigint

1.3M/s

Mul shf 32

686K/s

Fastest: java nextInt16(bound) (2.9x faster)

View exact numbers

✓ java nextInt16(bound): 1,987,199 ops/sec
• Classic 16: 1,864,387 ops/sec
• Mul shf 16: 1,715,306 ops/sec
• Imul shf 16: 1,702,288 ops/sec
• java nextInt32(bound): 1,652,137 ops/sec
• Mul shf bigint: 1,311,774 ops/sec
• Mul shf 32: 685,927 ops/sec

Mobile Results:

Chrome 17 runs

Classic 16

301K/s

Mul shf 16

269K/s

Classic 32

269K/s

Imul shf 16

268K/s

Mul shf bigint

220K/s

rejection sampling of modulo bias

174K/s

Debiased Integer Multiplication

173K/s

Bitmask with Rejection

156K/s

Mul shf 32

116K/s

Fastest: Classic 16 (2.6x faster)

View exact numbers

✓ Classic 16: 300,936 ops/sec
• Mul shf 16: 269,359 ops/sec
• Classic 32: 269,209 ops/sec
• Imul shf 16: 267,977 ops/sec
• Mul shf bigint: 220,488 ops/sec
• rejection sampling of modulo bias: 174,292 ops/sec
• Debiased Integer Multiplication: 173,293 ops/sec
• Bitmask with Rejection: 156,425 ops/sec
• Mul shf 32: 115,938 ops/sec

Results show median performance across all community submissions. Run this benchmark yourself to verify on your device.

Rendered benchmark preparation results:

Suite status: <idle, ready to run>

Previous results

Test case name	Result
Classic 16
Mul shf 16
Imul shf 16
Classic int
Mul shf bigint
Mul shf 32
java nextInt16(bound)
java nextInt32(bound)

Fastest: N/A

Slowest: N/A

Latest run results:

Run details: (Test run date: one year ago)

User agent: Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/137.0.0.0 Safari/537.36

Browser/OS: Chrome 137 on Windows

View result in a separate tab

Classic 16

3.2M/s

Mul shf 16

3.2M/s

Imul shf 16

3.1M/s

Classic int

3.1M/s

java nextInt32(bound)

3.1M/s

java nextInt16(bound)

3.0M/s

Mul shf bigint

2.4M/s

Mul shf 32

1.4M/s

View exact numbers

Test name	Executions per second
✓ Classic 16	3,156,994 Ops/sec
Mul shf 16	3,150,880 Ops/sec
Imul shf 16	3,149,801 Ops/sec
Classic int	3,133,590 Ops/sec
java nextInt32(bound)	3,113,806 Ops/sec
java nextInt16(bound)	2,981,598 Ops/sec
Mul shf bigint	2,397,238 Ops/sec
Mul shf 32	1,395,642 Ops/sec

unbiased ranged random integer generation functions (version: 16)

Comparing performance of: `Classic 16 vs Mul shf 16 vs Imul shf 16 vs Classic int vs Mul shf bigint vs Mul shf 32 vs java nextInt16(bound) vs java nextInt32(bound)`

Created: one year ago by: Registered User

Go to the latest result

Classic 16

Mul shf 16

Imul shf 16

Classic int

Mul shf bigint

Mul shf 32

java nextInt16(bound)

java nextInt32(bound)

Historical Performance Data aggregated from 42 community runs

Desktop Results:

Chrome 18 runs

Firefox 7 runs

Mobile Results:

Chrome 17 runs

Suite status: <idle, ready to run>

Fastest: N/A

Slowest: N/A

unbiased ranged random integer generation functions (version: 16)

Comparing performance of: Classic 16 vs Mul shf 16 vs Imul shf 16 vs Classic int vs Mul shf bigint vs Mul shf 32 vs java nextInt16(bound) vs java nextInt32(bound)

Created: one year ago by: Registered User

Go to the latest result

Classic 16

Mul shf 16

Imul shf 16

Classic int

Mul shf bigint

Mul shf 32

java nextInt16(bound)

java nextInt32(bound)

Historical Performance Data aggregated from 42 community runs

Desktop Results:

Chrome 18 runs

Firefox 7 runs

Mobile Results:

Chrome 17 runs

Suite status: <idle, ready to run>

Fastest: N/A

Slowest: N/A

Related benchmarks

Comparing performance of: `Classic 16 vs Mul shf 16 vs Imul shf 16 vs Classic int vs Mul shf bigint vs Mul shf 32 vs java nextInt16(bound) vs java nextInt32(bound)`