You cannot select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

218 lines
7.0 KiB
Solidity

1 year ago
// SPDX-License-Identifier: MIT
pragma solidity ^0.8.0;
import "forge-std/Test.sol";
import "../../../contracts/utils/math/Math.sol";
import "../../../contracts/utils/math/SafeMath.sol";
contract MathTest is Test {
// CEILDIV
function testCeilDiv(uint256 a, uint256 b) public {
vm.assume(b > 0);
uint256 result = Math.ceilDiv(a, b);
if (result == 0) {
assertEq(a, 0);
} else {
uint256 maxdiv = UINT256_MAX / b;
bool overflow = maxdiv * b < a;
assertTrue(a > b * (result - 1));
assertTrue(overflow ? result == maxdiv + 1 : a <= b * result);
}
}
// SQRT
function testSqrt(uint256 input, uint8 r) public {
Math.Rounding rounding = _asRounding(r);
uint256 result = Math.sqrt(input, rounding);
// square of result is bigger than input
if (_squareBigger(result, input)) {
assertTrue(rounding == Math.Rounding.Up);
assertTrue(_squareSmaller(result - 1, input));
}
// square of result is smaller than input
else if (_squareSmaller(result, input)) {
assertFalse(rounding == Math.Rounding.Up);
assertTrue(_squareBigger(result + 1, input));
}
// input is perfect square
else {
assertEq(result * result, input);
}
}
function _squareBigger(uint256 value, uint256 ref) private pure returns (bool) {
(bool noOverflow, uint256 square) = SafeMath.tryMul(value, value);
return !noOverflow || square > ref;
}
function _squareSmaller(uint256 value, uint256 ref) private pure returns (bool) {
return value * value < ref;
}
// LOG2
function testLog2(uint256 input, uint8 r) public {
Math.Rounding rounding = _asRounding(r);
uint256 result = Math.log2(input, rounding);
if (input == 0) {
assertEq(result, 0);
} else if (_powerOf2Bigger(result, input)) {
assertTrue(rounding == Math.Rounding.Up);
assertTrue(_powerOf2Smaller(result - 1, input));
} else if (_powerOf2Smaller(result, input)) {
assertFalse(rounding == Math.Rounding.Up);
assertTrue(_powerOf2Bigger(result + 1, input));
} else {
assertEq(2 ** result, input);
}
}
function _powerOf2Bigger(uint256 value, uint256 ref) private pure returns (bool) {
return value >= 256 || 2 ** value > ref; // 2**256 overflows uint256
}
function _powerOf2Smaller(uint256 value, uint256 ref) private pure returns (bool) {
return 2 ** value < ref;
}
// LOG10
function testLog10(uint256 input, uint8 r) public {
Math.Rounding rounding = _asRounding(r);
uint256 result = Math.log10(input, rounding);
if (input == 0) {
assertEq(result, 0);
} else if (_powerOf10Bigger(result, input)) {
assertTrue(rounding == Math.Rounding.Up);
assertTrue(_powerOf10Smaller(result - 1, input));
} else if (_powerOf10Smaller(result, input)) {
assertFalse(rounding == Math.Rounding.Up);
assertTrue(_powerOf10Bigger(result + 1, input));
} else {
assertEq(10 ** result, input);
}
}
function _powerOf10Bigger(uint256 value, uint256 ref) private pure returns (bool) {
return value >= 78 || 10 ** value > ref; // 10**78 overflows uint256
}
function _powerOf10Smaller(uint256 value, uint256 ref) private pure returns (bool) {
return 10 ** value < ref;
}
// LOG256
function testLog256(uint256 input, uint8 r) public {
Math.Rounding rounding = _asRounding(r);
uint256 result = Math.log256(input, rounding);
if (input == 0) {
assertEq(result, 0);
} else if (_powerOf256Bigger(result, input)) {
assertTrue(rounding == Math.Rounding.Up);
assertTrue(_powerOf256Smaller(result - 1, input));
} else if (_powerOf256Smaller(result, input)) {
assertFalse(rounding == Math.Rounding.Up);
assertTrue(_powerOf256Bigger(result + 1, input));
} else {
assertEq(256 ** result, input);
}
}
function _powerOf256Bigger(uint256 value, uint256 ref) private pure returns (bool) {
return value >= 32 || 256 ** value > ref; // 256**32 overflows uint256
}
function _powerOf256Smaller(uint256 value, uint256 ref) private pure returns (bool) {
return 256 ** value < ref;
}
// MULDIV
function testMulDiv(uint256 x, uint256 y, uint256 d) public {
// Full precision for x * y
(uint256 xyHi, uint256 xyLo) = _mulHighLow(x, y);
// Assume result won't overflow (see {testMulDivDomain})
// This also checks that `d` is positive
vm.assume(xyHi < d);
// Perform muldiv
uint256 q = Math.mulDiv(x, y, d);
// Full precision for q * d
(uint256 qdHi, uint256 qdLo) = _mulHighLow(q, d);
// Add remainder of x * y / d (computed as rem = (x * y % d))
(uint256 qdRemLo, uint256 c) = _addCarry(qdLo, _mulmod(x, y, d));
uint256 qdRemHi = qdHi + c;
// Full precision check that x * y = q * d + rem
assertEq(xyHi, qdRemHi);
assertEq(xyLo, qdRemLo);
}
function testMulDivDomain(uint256 x, uint256 y, uint256 d) public {
(uint256 xyHi, ) = _mulHighLow(x, y);
// Violate {testMulDiv} assumption (covers d is 0 and result overflow)
vm.assume(xyHi >= d);
// we are outside the scope of {testMulDiv}, we expect muldiv to revert
try this.muldiv(x, y, d) returns (uint256) {
fail();
} catch {}
}
// External call
function muldiv(uint256 x, uint256 y, uint256 d) external pure returns (uint256) {
return Math.mulDiv(x, y, d);
}
// Helpers
function _asRounding(uint8 r) private returns (Math.Rounding) {
vm.assume(r < uint8(type(Math.Rounding).max));
return Math.Rounding(r);
}
function _mulmod(uint256 x, uint256 y, uint256 z) private pure returns (uint256 r) {
assembly {
r := mulmod(x, y, z)
}
}
function _mulHighLow(uint256 x, uint256 y) private pure returns (uint256 high, uint256 low) {
(uint256 x0, uint256 x1) = (x & type(uint128).max, x >> 128);
(uint256 y0, uint256 y1) = (y & type(uint128).max, y >> 128);
// Karatsuba algorithm
// https://en.wikipedia.org/wiki/Karatsuba_algorithm
uint256 z2 = x1 * y1;
uint256 z1a = x1 * y0;
uint256 z1b = x0 * y1;
uint256 z0 = x0 * y0;
uint256 carry = ((z1a & type(uint128).max) + (z1b & type(uint128).max) + (z0 >> 128)) >> 128;
high = z2 + (z1a >> 128) + (z1b >> 128) + carry;
unchecked {
low = x * y;
}
}
function _addCarry(uint256 x, uint256 y) private pure returns (uint256 res, uint256 carry) {
unchecked {
res = x + y;
}
carry = res < x ? 1 : 0;
}
}