mirror of
				https://github.com/RGBCube/serenity
				synced 2025-10-28 11:02:36 +00:00 
			
		
		
		
	 1682f0b760
			
		
	
	
		1682f0b760
		
	
	
	
	
		
			
			SPDX License Identifiers are a more compact / standardized way of representing file license information. See: https://spdx.dev/resources/use/#identifiers This was done with the `ambr` search and replace tool. ambr --no-parent-ignore --key-from-file --rep-from-file key.txt rep.txt *
		
			
				
	
	
		
			252 lines
		
	
	
	
		
			7 KiB
		
	
	
	
		
			C++
		
	
	
	
	
	
			
		
		
	
	
			252 lines
		
	
	
	
		
			7 KiB
		
	
	
	
		
			C++
		
	
	
	
	
	
| /*
 | |
|  * Copyright (c) 2020, the SerenityOS developers.
 | |
|  *
 | |
|  * SPDX-License-Identifier: BSD-2-Clause
 | |
|  */
 | |
| 
 | |
| #include "SignedBigInteger.h"
 | |
| #include <AK/StringBuilder.h>
 | |
| 
 | |
| namespace Crypto {
 | |
| 
 | |
| SignedBigInteger SignedBigInteger::import_data(const u8* ptr, size_t length)
 | |
| {
 | |
|     bool sign = *ptr;
 | |
|     auto unsigned_data = UnsignedBigInteger::import_data(ptr + 1, length - 1);
 | |
|     return { move(unsigned_data), sign };
 | |
| }
 | |
| 
 | |
| size_t SignedBigInteger::export_data(Bytes data, bool remove_leading_zeros) const
 | |
| {
 | |
|     // FIXME: Support this:
 | |
|     //        m <0XX> -> m <XX> (if remove_leading_zeros)
 | |
|     VERIFY(!remove_leading_zeros);
 | |
| 
 | |
|     data[0] = m_sign;
 | |
|     auto bytes_view = data.slice(1, data.size() - 1);
 | |
|     return m_unsigned_data.export_data(bytes_view, remove_leading_zeros) + 1;
 | |
| }
 | |
| 
 | |
| SignedBigInteger SignedBigInteger::from_base10(StringView str)
 | |
| {
 | |
|     bool sign = false;
 | |
|     if (str.length() > 1) {
 | |
|         auto maybe_sign = str[0];
 | |
|         if (maybe_sign == '-') {
 | |
|             str = str.substring_view(1, str.length() - 1);
 | |
|             sign = true;
 | |
|         }
 | |
|         if (maybe_sign == '+')
 | |
|             str = str.substring_view(1, str.length() - 1);
 | |
|     }
 | |
|     auto unsigned_data = UnsignedBigInteger::from_base10(str);
 | |
|     return { move(unsigned_data), sign };
 | |
| }
 | |
| 
 | |
| String SignedBigInteger::to_base10() const
 | |
| {
 | |
|     StringBuilder builder;
 | |
| 
 | |
|     if (m_sign)
 | |
|         builder.append('-');
 | |
| 
 | |
|     builder.append(m_unsigned_data.to_base10());
 | |
| 
 | |
|     return builder.to_string();
 | |
| }
 | |
| 
 | |
| FLATTEN SignedBigInteger SignedBigInteger::plus(const SignedBigInteger& other) const
 | |
| {
 | |
|     // If both are of the same sign, just add the unsigned data and return.
 | |
|     if (m_sign == other.m_sign)
 | |
|         return { other.m_unsigned_data.plus(m_unsigned_data), m_sign };
 | |
| 
 | |
|     // One value is signed while the other is not.
 | |
|     return m_sign ? other.minus(this->m_unsigned_data) : minus(other.m_unsigned_data);
 | |
| }
 | |
| 
 | |
| FLATTEN SignedBigInteger SignedBigInteger::minus(const SignedBigInteger& other) const
 | |
| {
 | |
|     // If the signs are different, convert the op to an addition.
 | |
|     if (m_sign != other.m_sign) {
 | |
|         // -x - y = - (x + y)
 | |
|         // x - -y = (x + y)
 | |
|         SignedBigInteger result { other.m_unsigned_data.plus(this->m_unsigned_data) };
 | |
|         if (m_sign)
 | |
|             result.negate();
 | |
|         return result;
 | |
|     }
 | |
| 
 | |
|     if (!m_sign) {
 | |
|         // Both operands are positive.
 | |
|         // x - y = - (y - x)
 | |
|         if (m_unsigned_data < other.m_unsigned_data) {
 | |
|             // The result will be negative.
 | |
|             return { other.m_unsigned_data.minus(m_unsigned_data), true };
 | |
|         }
 | |
| 
 | |
|         // The result will be either zero, or positive.
 | |
|         return SignedBigInteger { m_unsigned_data.minus(other.m_unsigned_data) };
 | |
|     }
 | |
| 
 | |
|     // Both operands are negative.
 | |
|     // -x - -y = y - x
 | |
|     if (m_unsigned_data < other.m_unsigned_data) {
 | |
|         // The result will be positive.
 | |
|         return SignedBigInteger { other.m_unsigned_data.minus(m_unsigned_data), true };
 | |
|     }
 | |
|     // The result will be either zero, or negative.
 | |
|     // y - x = - (x - y)
 | |
|     return SignedBigInteger { m_unsigned_data.minus(other.m_unsigned_data) };
 | |
| }
 | |
| 
 | |
| FLATTEN SignedBigInteger SignedBigInteger::plus(const UnsignedBigInteger& other) const
 | |
| {
 | |
|     if (m_sign) {
 | |
|         if (other < m_unsigned_data)
 | |
|             return { m_unsigned_data.minus(other), true };
 | |
| 
 | |
|         return { other.minus(m_unsigned_data), false };
 | |
|     }
 | |
| 
 | |
|     return { m_unsigned_data.plus(other), false };
 | |
| }
 | |
| 
 | |
| FLATTEN SignedBigInteger SignedBigInteger::minus(const UnsignedBigInteger& other) const
 | |
| {
 | |
|     if (m_sign)
 | |
|         return { m_unsigned_data.plus(m_unsigned_data), true };
 | |
| 
 | |
|     if (other < m_unsigned_data)
 | |
|         return { m_unsigned_data.minus(other), false };
 | |
| 
 | |
|     return { other.minus(m_unsigned_data), true };
 | |
| }
 | |
| 
 | |
| FLATTEN SignedBigInteger SignedBigInteger::bitwise_or(const UnsignedBigInteger& other) const
 | |
| {
 | |
|     return { unsigned_value().bitwise_or(other), m_sign };
 | |
| }
 | |
| 
 | |
| FLATTEN SignedBigInteger SignedBigInteger::bitwise_and(const UnsignedBigInteger& other) const
 | |
| {
 | |
|     return { unsigned_value().bitwise_and(other), false };
 | |
| }
 | |
| 
 | |
| FLATTEN SignedBigInteger SignedBigInteger::bitwise_xor(const UnsignedBigInteger& other) const
 | |
| {
 | |
|     return { unsigned_value().bitwise_xor(other), m_sign };
 | |
| }
 | |
| 
 | |
| FLATTEN SignedBigInteger SignedBigInteger::bitwise_not() const
 | |
| {
 | |
|     return { unsigned_value().bitwise_not(), !m_sign };
 | |
| }
 | |
| 
 | |
| FLATTEN SignedBigInteger SignedBigInteger::bitwise_or(const SignedBigInteger& other) const
 | |
| {
 | |
|     auto result = bitwise_or(other.unsigned_value());
 | |
| 
 | |
|     // The sign bit will have to be OR'd manually.
 | |
|     if (other.is_negative())
 | |
|         result.negate();
 | |
| 
 | |
|     return result;
 | |
| }
 | |
| 
 | |
| FLATTEN SignedBigInteger SignedBigInteger::bitwise_and(const SignedBigInteger& other) const
 | |
| {
 | |
|     auto result = bitwise_and(other.unsigned_value());
 | |
| 
 | |
|     // The sign bit will have to be AND'd manually.
 | |
|     result.m_sign = is_negative() || other.is_negative();
 | |
| 
 | |
|     return result;
 | |
| }
 | |
| 
 | |
| FLATTEN SignedBigInteger SignedBigInteger::bitwise_xor(const SignedBigInteger& other) const
 | |
| {
 | |
|     auto result = bitwise_xor(other.unsigned_value());
 | |
| 
 | |
|     // The sign bit will have to be XOR'd manually.
 | |
|     result.m_sign = is_negative() ^ other.is_negative();
 | |
| 
 | |
|     return result;
 | |
| }
 | |
| 
 | |
| bool SignedBigInteger::operator==(const UnsignedBigInteger& other) const
 | |
| {
 | |
|     if (m_sign)
 | |
|         return false;
 | |
|     return m_unsigned_data == other;
 | |
| }
 | |
| 
 | |
| bool SignedBigInteger::operator!=(const UnsignedBigInteger& other) const
 | |
| {
 | |
|     if (m_sign)
 | |
|         return true;
 | |
|     return m_unsigned_data != other;
 | |
| }
 | |
| 
 | |
| bool SignedBigInteger::operator<(const UnsignedBigInteger& other) const
 | |
| {
 | |
|     if (m_sign)
 | |
|         return true;
 | |
|     return m_unsigned_data < other;
 | |
| }
 | |
| 
 | |
| FLATTEN SignedBigInteger SignedBigInteger::shift_left(size_t num_bits) const
 | |
| {
 | |
|     return SignedBigInteger { m_unsigned_data.shift_left(num_bits), m_sign };
 | |
| }
 | |
| 
 | |
| FLATTEN SignedBigInteger SignedBigInteger::multiplied_by(const SignedBigInteger& other) const
 | |
| {
 | |
|     bool result_sign = m_sign ^ other.m_sign;
 | |
|     return { m_unsigned_data.multiplied_by(other.m_unsigned_data), result_sign };
 | |
| }
 | |
| 
 | |
| FLATTEN SignedDivisionResult SignedBigInteger::divided_by(const SignedBigInteger& divisor) const
 | |
| {
 | |
|     // Aa / Bb -> (A^B)q, Ar
 | |
|     bool result_sign = m_sign ^ divisor.m_sign;
 | |
|     auto unsigned_division_result = m_unsigned_data.divided_by(divisor.m_unsigned_data);
 | |
|     return {
 | |
|         { move(unsigned_division_result.quotient), result_sign },
 | |
|         { move(unsigned_division_result.remainder), m_sign }
 | |
|     };
 | |
| }
 | |
| 
 | |
| void SignedBigInteger::set_bit_inplace(size_t bit_index)
 | |
| {
 | |
|     m_unsigned_data.set_bit_inplace(bit_index);
 | |
| }
 | |
| 
 | |
| bool SignedBigInteger::operator==(const SignedBigInteger& other) const
 | |
| {
 | |
|     if (is_invalid() != other.is_invalid())
 | |
|         return false;
 | |
| 
 | |
|     if (m_unsigned_data == 0 && other.m_unsigned_data == 0)
 | |
|         return true;
 | |
| 
 | |
|     return m_sign == other.m_sign && m_unsigned_data == other.m_unsigned_data;
 | |
| }
 | |
| 
 | |
| bool SignedBigInteger::operator!=(const SignedBigInteger& other) const
 | |
| {
 | |
|     return !(*this == other);
 | |
| }
 | |
| 
 | |
| bool SignedBigInteger::operator<(const SignedBigInteger& other) const
 | |
| {
 | |
|     if (m_sign ^ other.m_sign)
 | |
|         return m_sign;
 | |
| 
 | |
|     if (m_sign)
 | |
|         return other.m_unsigned_data < m_unsigned_data;
 | |
| 
 | |
|     return m_unsigned_data < other.m_unsigned_data;
 | |
| }
 | |
| 
 | |
| }
 |