atarpara/prb-math

Smart contract library for advanced fixed-point math

PRBMath

Smart contract library for advanced fixed-point math that operates with signed 59.18-decimal fixed-point and unsigned 60.18-decimal fixed-point numbers. The name of the number formats stems from the fact that there can be up to 59/60 digits in the integer part and up to 18 decimals in the fractional part. The numbers are bound by the minimum and the maximum values permitted by the Solidity types int256 and uint256.

• Operates with signed and unsigned denary fixed-point numbers, with 18 trailing decimals
• Offers advanced math functions like logarithms, exponentials, powers and square roots
• Gas efficient, but still user-friendly
• Bakes in overflow-safe multiplication and division
• Reverts with custom errors instead of reason strings
• Well-documented via NatSpec comments
• Thoroughly tested with Hardhat and Waffle

I created this because I wanted a fixed-point math library that is at the same time practical, intuitive and efficient. I looked at ABDKMath64x64, which is fast, but I didn't like that it operates with binary numbers and it limits the precision to int128. I then looked at Fixidity, which operates with denary numbers and has wide precision, but is slow and susceptible to phantom overflow.

Install

With yarn:

$ yarn add @prb/math

Or npm:

$ npm install @prb/math

Usage

PRBMath comes in four flavors: basic signed, typed signed, basic unsigned and typed unsigned.

PRBMathSD59x18.sol

// SPDX-License-Identifier: Unlicense
pragma solidity >=0.8.4;

import "@prb/math/contracts/PRBMathSD59x18.sol";

contract SignedConsumer {
  using PRBMathSD59x18 for int256;

  function signedLog2(int256 x) external pure returns (int256 result) {
    result = x.log2();
  }

  /// @notice Calculates x*y÷1e18 while handling possible intermediary overflow.
  /// @dev Try this with x = type(int256).max and y = 5e17.
  function signedMul(int256 x, int256 y) external pure returns (int256 result) {
    result = x.mul(y);
  }

  /// @dev Assuming that 1e18 = 100% and 1e16 = 1%.
  function signedYield(int256 principal, int256 apr) external pure returns (int256 result) {
    result = principal.mul(apr);
  }
}

PRBMathSD59x18Typed.sol

// SPDX-License-Identifier: Unlicense
pragma solidity >=0.8.4;

import "@prb/math/contracts/PRBMathSD59x18Typed.sol";

contract SignedConsumerTyped {
  using PRBMathSD59x18Typed for PRBMath.SD59x18;

  function signedLog2(int256 x) external pure returns (int256 result) {
    PRBMath.SD59x18 memory xsd = PRBMath.SD59x18({ value: x });
    result = xsd.log2().value;
  }

  /// @notice Calculates x*y÷1e18 while handling possible intermediary overflow.
  /// @dev Try this with x = type(int256).max and y = 5e17.
  function signedMul(int256 x, int256 y) external pure returns (int256 result) {
    PRBMath.SD59x18 memory xsd = PRBMath.SD59x18({ value: x });
    PRBMath.SD59x18 memory ysd = PRBMath.SD59x18({ value: y });
    result = xsd.mul(ysd).value;
  }

  /// @dev Assuming that 1e18 = 100% and 1e16 = 1%.
  function signedYield(int256 principal, int256 apr) external pure returns (int256 result) {
    PRBMath.SD59x18 memory principalSd = PRBMath.SD59x18({ value: principal });
    PRBMath.SD59x18 memory aprSd = PRBMath.SD59x18({ value: apr });
    result = principalSd.mul(aprSd).value;
  }
}

PRBMathUD60x18.sol

// SPDX-License-Identifier: Unlicense
pragma solidity >=0.8.4;

import "@prb/math/contracts/PRBMathUD60x18.sol";

contract UnsignedConsumer {
  using PRBMathUD60x18 for uint256;

  /// @dev Note that "x" must be greater than or equal to 1e18, lest the result would be negative, and negative
  /// numbers are not supported by the unsigned 60.18-decimal fixed-point representation.
  function unsignedLog2(uint256 x) external pure returns (uint256 result) {
    result = x.log2();
  }

  /// @notice Calculates x*y÷1e18 while handling possible intermediary overflow.
  /// @dev Try this with x = type(uint256).max and y = 5e17.
  function unsignedMul(uint256 x, uint256 y) external pure returns (uint256 result) {
    result = x.mul(y);
  }

  /// @dev Assuming that 1e18 = 100% and 1e16 = 1%.
  function unsignedYield(uint256 principal, uint256 apr) external pure returns (uint256 result) {
    result = principal.mul(apr);
  }
}

PRBMathUD60x18Typed.sol

// SPDX-License-Identifier: Unlicense
pragma solidity >=0.8.4;

import "@prb/math/contracts/PRBMathUD60x18Typed.sol";

contract UnsignedConsumerTyped {
  using PRBMathUD60x18Typed for PRBMath.UD60x18;

  function unsignedLog2(uint256 x) external pure returns (uint256 result) {
    PRBMath.UD60x18 memory xud = PRBMath.UD60x18({ value: x });
    result = xud.log2().value;
  }

  /// @notice Calculates x*y÷1e18 while handling possible intermediary overflow.
  /// @dev Try this with x = type(uint256).max and y = 5e17.
  function unsignedMul(uint256 x, uint256 y) external pure returns (uint256 result) {
    PRBMath.UD60x18 memory xud = PRBMath.UD60x18({ value: x });
    PRBMath.UD60x18 memory yud = PRBMath.UD60x18({ value: y });
    result = xud.mul(yud).value;
  }

  /// @dev Assuming that 1e18 = 100% and 1e16 = 1%.
  function unsignedYield(uint256 principal, uint256 apr) external pure returns (uint256 result) {
    PRBMath.UD60x18 memory principalUd = PRBMath.UD60x18({ value: principal });
    PRBMath.UD60x18 memory aprUd = PRBMath.UD60x18({ value: apr });
    result = principalUd.mul(aprUd).value;
  }
}

JavaScript SDK

PRBMath is accompanied by a JavaScript SDK. Whatever functions there are in the Solidity library, you should find them replicated in the SDK. The only exception are fromUint and toUint, for which you can use evm-bn.

Here's an example for how to calculate the binary logarithm:

import type { BigNumber } from "@ethersproject/bignumber";
import { log2 } from "@prb/math";
import { fromBn, toBn } from "evm-bn";

(async function () {
  const x: BigNumber = toBn("16");
  const result: BigNumber = log2(x);
  console.log({ result: fromBn(result) });
})();

Pro tip: see how the SDK is used in the tests for PRBMath.

Gas Efficiency

The typeless PRBMath library is faster than ABDKMath for abs, exp, exp2, gm, inv, ln, log2. Conversely, it is slower than ABDKMath for avg, div, mul, powu and sqrt. There are two technical reasons why PRBMath lags behind ABDKMath's mul and div functions:

1. PRBMath operates with 256-bit word sizes, so it has to account for possible intermediary overflow. ABDKMath operates with 128-bit word sizes.
2. PRBMath rounds up instead of truncating in certain cases (see listing 6 and text above it in this article), which makes it slightly more precise than ABDKMath but comes at a gas cost.

PRBMath

Based on the v2.0.1 of the library.

SD59x18	Min	Max	Avg		UD60x18	Min	Max	Avg
abs	68	72	70		n/a	n/a	n/a	n/a
avg	57	57	57		avg	57	57	57
ceil	82	117	101		ceil	78	78	78
div	431	483	451		div	205	205	205
exp	38	2797	2263		exp	1874	2742	2244
exp2	63	2678	2104		exp2	1784	2652	2156
floor	82	117	101		floor	43	43	43
frac	23	23	23		frac	23	23	23
fromInt	83	83	83		fromUint	49	49	49
gm	26	892	690		gm	26	893	691
inv	40	40	40		inv	40	40	40
ln	463	7306	4724		ln	419	6902	3814
log10	104	9074	4337		log10	503	8695	4571
log2	377	7241	4243		log2	330	6825	3426
mul	455	463	459		mul	219	275	247
pow	64	11338	8518		pow	64	10637	6635
powu	293	24745	5681		powu	83	24535	5471
sqrt	140	839	716		sqrt	114	846	710
toInt	23	23	23		toUint	23	23	23

PRBMathTyped

Based on the v2.0.1 of the library.

SD59x18	Min	Max	Avg		UD60x18	Min	Max	Avg
abs	128	132	130		n/a	n/a	n/a	n/a
add	221	221	221		add	97	97	97
avg	120	120	120		avg	120	120	120
ceil	95	166	141		ceil	132	132	132
div	524	582	546		div	259	259	259
exp	82	3076	2617		exp	2086	2954	2456
exp2	110	2768	2233		exp2	1840	2708	2212
floor	95	171	143		floor	97	97	97
fromInt	118	118	118		fromUint	84	84	84
frac	82	82	82		frac	77	77	77
gm	67	947	741		gm	67	948	743
inv	82	82	82		inv	82	82	82
ln	645	7503	5041		ln	626	7124	4029
log10	182	9287	4337		log10	2414	8912	7280
log2	422	7285	4701		log2	407	6910	4108
mul	538	546	542		mul	273	336	305
pow	115	11824	8471		pow	115	11129	7346
powu	479	25213	5931		powu	132	24426	4207
sqrt	195	918	798		sqrt	153	903	769
sub	218	218	218		sub	98	98	98
toInt	29	29	29		toUint	29	29	29

ABDKMath64x64

Based on v3.0 of the library. See abdk-gas-estimations.

Method	Min	Max	Avg
abs	88	92	90
avg	41	41	41
div	168	168	168
exp	77	3780	2687
exp2	77	3600	2746
gavg	166	875	719
inv	157	157	157
ln	7074	7164	7126
log2	6972	7062	7024
mul	111	111	111
pow	303	4740	1792
sqrt	129	809	699

Security

While I set a high bar for code quality and test coverage, you shouldn't assume that this project is completely safe to use. The contracts have not been audited by a security researcher.

Caveat Emptor

This is experimental software and is provided on an "as is" and "as available" basis. I do not give any warranties and will not be liable for any loss, direct or indirect through continued use of this codebase.

Contact

If you discover any security issues, please report them via Keybase.

Acknowledgements

• Mikhail Vladimirov for the insights he shared in his Math in Solidity series.
• Remco Bloemen for his work on overflow-safe multiplication and division and for responding to the questions I asked him while developing the library.

License

Unlicense