blob: 2c8b62d37965296ee2d8862e941bc7cd052b289d [file] [log] [blame]
/*
* Licensed to the Apache Software Foundation (ASF) under one or more
* contributor license agreements. See the NOTICE file distributed with
* this work for additional information regarding copyright ownership.
* The ASF licenses this file to You under the Apache License, Version 2.0
* (the "License"); you may not use this file except in compliance with
* the License. You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
#include "ignite/ignite_error.h"
#include "ignite/common/bits.h"
#include "ignite/common/big_integer.h"
namespace ignite
{
namespace common
{
BigInteger::BigInteger() :
sign(1),
mag()
{
// No-op.
}
BigInteger::BigInteger(int64_t val) :
sign(1),
mag()
{
AssignInt64(val);
}
BigInteger::BigInteger(const char* val, int32_t len) :
sign(1),
mag()
{
AssignString(val, len);
}
BigInteger::BigInteger(const BigInteger& other) :
sign(other.sign),
mag(other.mag)
{
// No-op.
}
BigInteger::BigInteger(const int8_t* val, int32_t len, int32_t sign, bool bigEndian) :
sign(sign),
mag()
{
assert(val != 0);
assert(len >= 0);
assert(sign == 1 || sign == 0 || sign == -1);
if (bigEndian)
{
int32_t firstNonZero = 0;
while (firstNonZero < len && val[firstNonZero] == 0)
++firstNonZero;
if (firstNonZero == len)
{
AssignInt64(0);
return;
}
int32_t intLength = (len - firstNonZero + 3) / 4;
mag.Resize(intLength);
int32_t b = len - 1;
for (int32_t i = 0; i < intLength - 1; ++i)
{
mag[i] = (val[b] & 0xFFUL)
| ((val[b - 1] & 0xFFUL) << 8)
| ((val[b - 2] & 0xFFUL) << 16)
| ((val[b - 3] & 0xFFUL) << 24);
b -= 4;
}
int32_t bytesRemaining = b - firstNonZero + 1;
assert(bytesRemaining > 0 && bytesRemaining <= 4);
switch (bytesRemaining)
{
case 4:
mag[intLength - 1] |= (val[b - 3] & 0xFF) << 24;
// Fall-through.
case 3:
mag[intLength - 1] |= (val[b - 2] & 0xFF) << 16;
// Fall-through.
case 2:
mag[intLength - 1] |= (val[b - 1] & 0xFF) << 8;
// Fall-through.
case 1:
mag[intLength - 1] |= val[b] & 0xFF;
// Fall-through.
default:
break;
}
}
else
{
int32_t firstNonZero = len - 1;
while (firstNonZero >= 0 && val[firstNonZero] == 0)
--firstNonZero;
if (firstNonZero == -1)
{
AssignInt64(0);
return;
}
int32_t intLength = (firstNonZero + 4) / 4;
mag.Resize(intLength);
int32_t b = 0;
for (int32_t i = 0; i < intLength - 1; ++i)
{
mag[i] = (val[b] & 0xFFUL)
| ((val[b + 1] & 0xFFUL) << 8)
| ((val[b + 2] & 0xFFUL) << 16)
| ((val[b + 3] & 0xFFUL) << 24);
b += 4;
}
int32_t bytesRemaining = firstNonZero - b + 1;
assert(bytesRemaining > 0 && bytesRemaining <= 4);
switch (bytesRemaining)
{
case 4:
mag[intLength - 1] |= (val[b + 3] & 0xFF) << 24;
// Fall-through.
case 3:
mag[intLength - 1] |= (val[b + 2] & 0xFF) << 16;
// Fall-through.
case 2:
mag[intLength - 1] |= (val[b + 1] & 0xFF) << 8;
// Fall-through.
case 1:
mag[intLength - 1] |= val[b] & 0xFF;
// Fall-through.
default:
break;
}
}
}
BigInteger::BigInteger(MagArray &mag, int8_t sign) :
sign(sign),
mag()
{
this->mag.Swap(mag);
}
BigInteger& BigInteger::operator=(const BigInteger& other)
{
Assign(other);
return *this;
}
void BigInteger::Assign(const BigInteger& val)
{
if (this != &val)
{
sign = val.sign;
mag = val.mag;
}
}
void BigInteger::AssignInt64(int64_t val)
{
if (val < 0)
{
AssignUint64(val > INT64_MIN ? static_cast<uint64_t>(-val) : static_cast<uint64_t>(val));
sign = -1;
}
else
AssignUint64(static_cast<uint64_t>(val));
}
void BigInteger::AssignString(const char* val, int32_t len)
{
std::stringstream converter;
converter.write(val, len);
converter >> *this;
}
void BigInteger::AssignUint64(uint64_t val)
{
sign = 1;
if (val == 0)
{
mag.Clear();
return;
}
uint32_t highWord = static_cast<uint32_t>(val >> 32);
if (highWord == 0)
mag.Resize(1);
else
{
mag.Resize(2);
mag[1] = highWord;
}
mag[0] = static_cast<uint32_t>(val);
}
int8_t BigInteger::GetSign() const
{
return sign;
}
void BigInteger::Swap(BigInteger& other)
{
using std::swap;
swap(sign, other.sign);
mag.Swap(other.mag);
}
const BigInteger::MagArray& BigInteger::GetMagnitude() const
{
return mag;
}
uint32_t BigInteger::GetBitLength() const
{
if (mag.IsEmpty())
return 0;
int32_t res = bits::BitLengthU32(mag[mag.GetSize() - 1]);
if (mag.GetSize() > 1)
res += (mag.GetSize() - 1) * 32;
return res;
}
int32_t BigInteger::GetPrecision() const
{
// See http://graphics.stanford.edu/~seander/bithacks.html
// for the details on the algorithm.
if (mag.GetSize() == 0)
return 1;
int32_t r = static_cast<uint32_t>(((
static_cast<uint64_t>(GetBitLength()) + 1) * 646456993) >> 31);
BigInteger prec;
BigInteger::GetPowerOfTen(r, prec);
return Compare(prec, true) < 0 ? r : r + 1;
}
void BigInteger::MagnitudeToBytes(common::FixedSizeArray<int8_t>& buffer) const
{
int32_t bytesNum = static_cast<int32_t>((GetBitLength() + 7) / 8);
if (bytesNum == 0)
{
buffer.Reset(1);
return;
}
buffer.Reset(bytesNum);
int32_t i;
for (i = 0; i < mag.GetSize() - 1; ++i)
{
int32_t j = bytesNum - 1 - 4 * i;
buffer[j] = static_cast<int8_t>(mag[i]);
buffer[j - 1] = static_cast<int8_t>(mag[i] >> 8);
buffer[j - 2] = static_cast<int8_t>(mag[i] >> 16);
buffer[j - 3] = static_cast<int8_t>(mag[i] >> 24);
}
int32_t bytesRemaining = bytesNum - 4 * i;
assert(bytesRemaining >= 0 && bytesRemaining <= 4);
i = 0;
switch (bytesRemaining)
{
case 4:
buffer[i++] |= static_cast<int8_t>(mag[mag.GetSize() - 1] >> 24);
// Fall-through.
case 3:
buffer[i++] |= static_cast<int8_t>(mag[mag.GetSize() - 1] >> 16);
// Fall-through.
case 2:
buffer[i++] |= static_cast<int8_t>(mag[mag.GetSize() - 1] >> 8);
// Fall-through.
case 1:
buffer[i++] |= static_cast<int8_t>(mag[mag.GetSize() - 1]);
// Fall-through.
default:
break;
}
}
void BigInteger::Pow(int32_t exp)
{
if (exp < 0)
{
AssignInt64(0);
return;
}
uint32_t bitsLen = GetBitLength();
if (!bitsLen)
return;
if (bitsLen == 1)
{
if ((exp % 2 == 0) && sign < 0)
sign = -sign;
return;
}
BigInteger multiplicant(*this);
AssignInt64(1);
int32_t mutExp = exp;
while (mutExp)
{
if (mutExp & 1)
Multiply(multiplicant, *this);
mutExp >>= 1;
if (mutExp)
multiplicant.Multiply(multiplicant, multiplicant);
}
}
void BigInteger::Multiply(const BigInteger& other, BigInteger& res) const
{
MagArray resMag(mag.GetSize() + other.mag.GetSize());
resMag.Resize(mag.GetSize() + other.mag.GetSize());
for (int32_t i = 0; i < other.mag.GetSize(); ++i)
{
uint32_t carry = 0;
for (int32_t j = 0; j < mag.GetSize(); ++j)
{
uint64_t product = static_cast<uint64_t>(mag[j]) * other.mag[i] +
+ resMag[i + j] + carry;
resMag[i + j] = static_cast<uint32_t>(product);
carry = static_cast<uint32_t>(product >> 32);
}
resMag[i + mag.GetSize()] = carry;
}
res.mag.Swap(resMag);
res.sign = sign * other.sign;
res.Normalize();
}
/**
* Shift magnitude left by the specified number of bits.
*
* @param in Input magnitude.
* @param len Magnitude length.
* @param out Output magnitude. Should be not shorter than the input
* magnitude.
* @param n Number of bits to shift to.
*/
void ShiftLeft(const uint32_t* in, int32_t len, uint32_t* out, unsigned n)
{
assert(n < 32);
if (n == 0)
{
std::copy(in, in + len, out);
return;
}
for (int32_t i = len - 1; i > 0; --i)
out[i] = (in[i] << n) | (in[i - 1] >> (32 - n));
out[0] = in[0] << n;
}
/**
* Shift magnitude right by the specified number of bits.
*
* @param in Input magnitude.
* @param len Magnitude length.
* @param out Output magnitude. Should be not shorter than the input
* magnitude.
* @param n Number of bits to shift to.
*/
void ShiftRight(const uint32_t* in, int32_t len, uint32_t* out, unsigned n)
{
assert(n < 32);
if (n == 0)
{
std::copy(in, in + len, out);
return;
}
for (int32_t i = 0; i < len - 1; ++i)
out[i] = (in[i] >> n) | (in[i + 1] << (32 - n));
out[len - 1] = in[len - 1] >> n;
}
/**
* Part of the division algorithm. Computes q - (a * x).
*
* @param q Minuend.
* @param a Multipliplier of the subtrahend.
* @param alen Length of the a.
* @param x Multipliplicand of the subtrahend.
* @return Carry.
*/
uint32_t MultiplyAndSubstruct(uint32_t* q, const uint32_t* a, int32_t alen, uint32_t x)
{
uint64_t carry = 0;
for (int32_t i = 0; i < alen; ++i)
{
uint64_t product = a[i] * static_cast<uint64_t>(x);
int64_t difference = q[i] - carry - (product & 0xFFFFFFFF);
q[i] = static_cast<uint32_t>(difference);
// This will add one if difference is negative.
carry = (product >> 32) - (difference >> 32);
}
return static_cast<uint32_t>(carry);
}
/**
* Add two magnitude arrays and return carry.
*
* @param res First addend. Result is placed here. Length of this addend
* should be equal or greater than len.
* @param addend Second addend.
* @param len Length of the second addend.
* @return Carry.
*/
uint32_t Add(uint32_t* res, const uint32_t* addend, int32_t len)
{
uint64_t carry = 0;
for (int32_t i = 0; i < len; ++i)
{
uint64_t sum = static_cast<uint64_t>(res[i]) + addend[i] + carry;
res[i] = static_cast<uint32_t>(sum);
carry = sum >> 32;
}
return static_cast<uint32_t>(carry);
}
/**
* Add single number to a magnitude array and return carry.
*
* @param res First addend. Result is placed here. Length of this addend
* should be equal or greater than len.
* @param len Length of the First addend.
* @param addend Second addend.
* @return Carry.
*/
uint32_t Add(uint32_t* res, int32_t len, uint32_t addend)
{
uint64_t carry = addend;
for (int32_t i = 0; (i < len) && carry; ++i)
{
uint64_t sum = static_cast<uint64_t>(res[i]) + carry;
res[i] = static_cast<uint32_t>(sum);
carry = sum >> 32;
}
return static_cast<uint32_t>(carry);
}
void BigInteger::Divide(const BigInteger& divisor, BigInteger& res) const
{
Divide(divisor, res, 0);
}
void BigInteger::Divide(const BigInteger& divisor, BigInteger& res, BigInteger& rem) const
{
Divide(divisor, res, &rem);
}
void BigInteger::Add(const uint32_t* addend, int32_t len)
{
if (mag.GetSize() < len)
{
mag.Reserve(len + 1);
mag.Resize(len);
}
else
mag.Reserve(mag.GetSize() + 1);
uint32_t carry = common::Add(mag.GetData(), addend, len);
if (carry)
{
carry = common::Add(mag.GetData() + len, mag.GetSize() - len, carry);
if (carry)
mag.PushBack(carry);
}
}
void BigInteger::Add(uint64_t x)
{
if (x == 0)
return;
if (IsZero())
{
AssignUint64(x);
return;
}
uint32_t val[2];
val[0] = static_cast<uint32_t>(x);
val[1] = static_cast<uint32_t>(x >> 32);
Add(val, val[1] ? 2 : 1);
}
int32_t BigInteger::Compare(const BigInteger& other, bool ignoreSign) const
{
// What we should return if magnitude is greater.
int32_t mgt = 1;
if (!ignoreSign)
{
if (sign != other.sign)
return sign > other.sign ? 1 : -1;
else
mgt = sign;
}
if (mag.GetSize() != other.mag.GetSize())
return mag.GetSize() > other.mag.GetSize() ? mgt : -mgt;
for (int32_t i = mag.GetSize() - 1; i >= 0; --i)
{
if (mag[i] == other.mag[i])
continue;
else if (mag[i] > other.mag[i])
return mgt;
else
return -mgt;
}
return 0;
}
int64_t BigInteger::ToInt64() const
{
return (static_cast<uint64_t>(GetMagInt(1)) << 32) | GetMagInt(0);
}
void BigInteger::GetPowerOfTen(int32_t pow, BigInteger& res)
{
using namespace common;
assert(pow >= 0);
if (pow < bits::UINT64_MAX_PRECISION)
{
res.AssignUint64(bits::TenPowerU64(pow));
return;
}
res.AssignInt64(10);
res.Pow(pow);
}
uint32_t BigInteger::GetMagInt(int32_t n) const
{
assert(n >= 0);
if (n >= mag.GetSize())
return sign > 0 ? 0 : -1;
return sign * mag[n];
}
void BigInteger::Divide(const BigInteger& divisor, BigInteger& res, BigInteger* rem) const
{
// Can't divide by zero.
if (divisor.mag.IsEmpty())
throw IgniteError(IgniteError::IGNITE_ERR_ILLEGAL_ARGUMENT, "Division by zero.");
int32_t compRes = Compare(divisor, true);
int8_t resSign = sign * divisor.sign;
// The same magnitude. Result is [-]1 and remainder is zero.
if (compRes == 0)
{
res.AssignInt64(resSign);
if (rem)
rem->AssignInt64(0);
return;
}
// Divisor is greater than this. Result is 0 and remainder is this.
if (compRes == -1)
{
// Order is important here! Copy to rem first to handle the case
// when &res == this.
if (rem)
rem->Assign(*this);
res.AssignInt64(0);
return;
}
// If divisor is [-]1 result is [-]this and remainder is zero.
if (divisor.GetBitLength() == 1)
{
// Once again: order is important.
res.Assign(*this);
res.sign = sign * divisor.sign;
if (rem)
rem->AssignInt64(0);
return;
}
// Trivial case.
if (mag.GetSize() <= 2)
{
uint64_t u = mag[0];
uint64_t v = divisor.mag[0];
if (mag.GetSize() == 2)
u |= static_cast<uint64_t>(mag[1]) << 32;
if (divisor.mag.GetSize() == 2)
v |= static_cast<uint64_t>(divisor.mag[1]) << 32;
// Divisor can not be 1, or 0.
assert(v > 1);
// It should also be less than dividend.
assert(v < u);
// (u / v) is always fits into int64_t because abs(v) >= 2.
res.AssignInt64(resSign * static_cast<int64_t>(u / v));
// (u % v) is always fits into int64_t because (u > v) ->
// (u % v) < (u / 2).
if (rem)
rem->AssignInt64(resSign * static_cast<int64_t>(u % v));
return;
}
// Using Knuth division algorithm D for common case.
// Short aliases.
const MagArray& u = mag;
const MagArray& v = divisor.mag;
MagArray& q = res.mag;
int32_t ulen = u.GetSize();
int32_t vlen = v.GetSize();
// First we need to normilize divisor.
MagArray nv;
nv.Resize(v.GetSize());
int32_t shift = bits::NumberOfLeadingZerosU32(v.Back());
ShiftLeft(v.GetData(), vlen, nv.GetData(), shift);
// Divisor is normilized. Now we need to normilize divident.
MagArray nu;
// First find out what is the size of it.
if (bits::NumberOfLeadingZerosU32(u.Back()) >= shift)
{
// Everything is the same as with divisor. Just add leading zero.
nu.Resize(ulen + 1);
ShiftLeft(u.GetData(), ulen, nu.GetData(), shift);
assert((static_cast<uint64_t>(u.Back()) >> (32 - shift)) == 0);
}
else
{
// We need one more byte here. Also adding leading zero.
nu.Resize(ulen + 2);
ShiftLeft(u.GetData(), ulen, nu.GetData(), shift);
nu[ulen] = u[ulen - 1] >> (32 - shift);
assert(nu[ulen] != 0);
}
assert(nu.Back() == 0);
// Resizing resulting array.
q.Resize(ulen - vlen + 1);
// Main loop
for (int32_t i = ulen - vlen; i >= 0; --i)
{
uint64_t base = bits::MakeU64(nu[i + vlen], nu[i + vlen - 1]);
uint64_t qhat = base / nv[vlen - 1]; // Estimated quotient.
uint64_t rhat = base % nv[vlen - 1]; // A remainder.
// Adjusting result if needed.
while (qhat > UINT32_MAX ||
((qhat * nv[vlen - 2]) > ((UINT32_MAX + static_cast<uint64_t>(1)) * rhat + nu[i + vlen - 2])))
{
--qhat;
rhat += nv[vlen - 1];
if (rhat > UINT32_MAX)
break;
}
uint32_t qhat32 = static_cast<uint32_t>(qhat);
// Multiply and subtract.
uint32_t carry = MultiplyAndSubstruct(nu.GetData() + i, nv.GetData(), vlen, qhat32);
int64_t difference = nu[i + vlen] - carry;
nu[i + vlen] = static_cast<uint32_t>(difference);
if (difference < 0)
{
--qhat32;
carry = common::Add(nu.GetData() + i, nv.GetData(), vlen);
assert(carry == 0);
}
q[i] = qhat32;
}
res.sign = resSign;
res.Normalize();
// If remainder is needed unnormolize it.
if (rem)
{
rem->sign = resSign;
rem->mag.Resize(vlen);
ShiftRight(nu.GetData(), rem->mag.GetSize(), rem->mag.GetData(), shift);
rem->Normalize();
}
}
void BigInteger::Normalize()
{
int32_t lastNonZero = mag.GetSize() - 1;
while (lastNonZero >= 0 && mag[lastNonZero] == 0)
--lastNonZero;
mag.Resize(lastNonZero + 1);
}
bool operator==(const BigInteger& val1, const BigInteger& val2)
{
return val1.Compare(val2) == 0;
}
bool operator!=(const BigInteger& val1, const BigInteger& val2)
{
return val1.Compare(val2) != 0;
}
bool operator<(const BigInteger& val1, const BigInteger& val2)
{
return val1.Compare(val2) < 0;
}
bool operator<=(const BigInteger& val1, const BigInteger& val2)
{
return val1.Compare(val2) <= 0;
}
bool operator>(const BigInteger& val1, const BigInteger& val2)
{
return val1.Compare(val2) > 0;
}
bool operator>=(const BigInteger& val1, const BigInteger& val2)
{
return val1.Compare(val2) >= 0;
}
}
}