aave-protocol-v2/contracts/protocol/libraries/math/WadRayMath.sol
2020-11-25 18:33:49 +01:00

136 lines
3.3 KiB
Solidity

// SPDX-License-Identifier: agpl-3.0
pragma solidity 0.6.12;
import {Errors} from '../helpers/Errors.sol';
/**
* @title WadRayMath library
* @author Aave
* @dev Provides mul and div function for wads (decimal numbers with 18 digits precision) and rays (decimals with 27 digits)
**/
library WadRayMath {
uint256 internal constant WAD = 1e18;
uint256 internal constant halfWAD = WAD / 2;
uint256 internal constant RAY = 1e27;
uint256 internal constant halfRAY = RAY / 2;
uint256 internal constant WAD_RAY_RATIO = 1e9;
/**
* @return One ray, 1e27
**/
function ray() internal pure returns (uint256) {
return RAY;
}
/**
* @return One wad, 1e18
**/
function wad() internal pure returns (uint256) {
return WAD;
}
/**
* @return Half ray, 1e27/2
**/
function halfRay() internal pure returns (uint256) {
return halfRAY;
}
/**
* @return Half ray, 1e18/2
**/
function halfWad() internal pure returns (uint256) {
return halfWAD;
}
/**
* @dev Multiplies two wad, rounding half up to the nearest wad
* @param a Wad
* @param b Wad
* @return The result of a*b, in wad
**/
function wadMul(uint256 a, uint256 b) internal pure returns (uint256) {
if (a == 0 || b == 0) {
return 0;
}
require(a <= (type(uint256).max - halfWAD) / b, Errors.MATH_MULTIPLICATION_OVERFLOW);
return (a * b + halfWAD) / WAD;
}
/**
* @dev Divides two wad, rounding half up to the nearest wad
* @param a Wad
* @param b Wad
* @return The result of a/b, in wad
**/
function wadDiv(uint256 a, uint256 b) internal pure returns (uint256) {
require(b != 0, Errors.MATH_DIVISION_BY_ZERO);
uint256 halfB = b / 2;
require(a <= (type(uint256).max - halfB) / WAD, Errors.MATH_MULTIPLICATION_OVERFLOW);
return (a * WAD + halfB) / b;
}
/**
* @dev Multiplies two ray, rounding half up to the nearest ray
* @param a Ray
* @param b Ray
* @return The result of a*b, in ray
**/
function rayMul(uint256 a, uint256 b) internal pure returns (uint256) {
if (a == 0 || b == 0) {
return 0;
}
require(a <= (type(uint256).max - halfRAY) / b, Errors.MATH_MULTIPLICATION_OVERFLOW);
return (a * b + halfRAY) / RAY;
}
/**
* @dev Divides two ray, rounding half up to the nearest ray
* @param a Ray
* @param b Ray
* @return The result of a/b, in ray
**/
function rayDiv(uint256 a, uint256 b) internal pure returns (uint256) {
require(b != 0, Errors.MATH_DIVISION_BY_ZERO);
uint256 halfB = b / 2;
require(a <= (type(uint256).max - halfB) / RAY, Errors.MATH_MULTIPLICATION_OVERFLOW);
return (a * RAY + halfB) / b;
}
/**
* @dev Casts ray down to wad
* @param a Ray
* @return a casted to wad, rounded half up to the nearest wad
**/
function rayToWad(uint256 a) internal pure returns (uint256) {
uint256 halfRatio = WAD_RAY_RATIO / 2;
uint256 result = halfRatio + a;
require(result >= halfRatio, Errors.MATH_ADDITION_OVERFLOW);
return result / WAD_RAY_RATIO;
}
/**
* @dev Converts wad up to ray
* @param a Wad
* @return a converted in ray
**/
function wadToRay(uint256 a) internal pure returns (uint256) {
uint256 result = a * WAD_RAY_RATIO;
require(result / WAD_RAY_RATIO == a, Errors.MATH_MULTIPLICATION_OVERFLOW);
return result;
}
}