Home | History | Annotate | Download | only in src
      1 // Copyright 2011 the V8 project authors. All rights reserved.
      2 // Use of this source code is governed by a BSD-style license that can be
      3 // found in the LICENSE file.
      4 
      5 #include "src/bignum.h"
      6 #include "src/utils.h"
      7 
      8 namespace v8 {
      9 namespace internal {
     10 
     11 Bignum::Bignum()
     12     : bigits_(bigits_buffer_, kBigitCapacity), used_digits_(0), exponent_(0) {
     13   for (int i = 0; i < kBigitCapacity; ++i) {
     14     bigits_[i] = 0;
     15   }
     16 }
     17 
     18 
     19 template<typename S>
     20 static int BitSize(S value) {
     21   return 8 * sizeof(value);
     22 }
     23 
     24 
     25 // Guaranteed to lie in one Bigit.
     26 void Bignum::AssignUInt16(uint16_t value) {
     27   DCHECK_GE(kBigitSize, BitSize(value));
     28   Zero();
     29   if (value == 0) return;
     30 
     31   EnsureCapacity(1);
     32   bigits_[0] = value;
     33   used_digits_ = 1;
     34 }
     35 
     36 
     37 void Bignum::AssignUInt64(uint64_t value) {
     38   const int kUInt64Size = 64;
     39 
     40   Zero();
     41   if (value == 0) return;
     42 
     43   int needed_bigits = kUInt64Size / kBigitSize + 1;
     44   EnsureCapacity(needed_bigits);
     45   for (int i = 0; i < needed_bigits; ++i) {
     46     bigits_[i] = static_cast<Chunk>(value & kBigitMask);
     47     value = value >> kBigitSize;
     48   }
     49   used_digits_ = needed_bigits;
     50   Clamp();
     51 }
     52 
     53 
     54 void Bignum::AssignBignum(const Bignum& other) {
     55   exponent_ = other.exponent_;
     56   for (int i = 0; i < other.used_digits_; ++i) {
     57     bigits_[i] = other.bigits_[i];
     58   }
     59   // Clear the excess digits (if there were any).
     60   for (int i = other.used_digits_; i < used_digits_; ++i) {
     61     bigits_[i] = 0;
     62   }
     63   used_digits_ = other.used_digits_;
     64 }
     65 
     66 
     67 static uint64_t ReadUInt64(Vector<const char> buffer,
     68                            int from,
     69                            int digits_to_read) {
     70   uint64_t result = 0;
     71   int to = from + digits_to_read;
     72 
     73   for (int i = from; i < to; ++i) {
     74     int digit = buffer[i] - '0';
     75     DCHECK(0 <= digit && digit <= 9);
     76     result = result * 10 + digit;
     77   }
     78   return result;
     79 }
     80 
     81 
     82 void Bignum::AssignDecimalString(Vector<const char> value) {
     83   // 2^64 = 18446744073709551616 > 10^19
     84   const int kMaxUint64DecimalDigits = 19;
     85   Zero();
     86   int length = value.length();
     87   int pos = 0;
     88   // Let's just say that each digit needs 4 bits.
     89   while (length >= kMaxUint64DecimalDigits) {
     90     uint64_t digits = ReadUInt64(value, pos, kMaxUint64DecimalDigits);
     91     pos += kMaxUint64DecimalDigits;
     92     length -= kMaxUint64DecimalDigits;
     93     MultiplyByPowerOfTen(kMaxUint64DecimalDigits);
     94     AddUInt64(digits);
     95   }
     96   uint64_t digits = ReadUInt64(value, pos, length);
     97   MultiplyByPowerOfTen(length);
     98   AddUInt64(digits);
     99   Clamp();
    100 }
    101 
    102 
    103 static int HexCharValue(char c) {
    104   if ('0' <= c && c <= '9') return c - '0';
    105   if ('a' <= c && c <= 'f') return 10 + c - 'a';
    106   if ('A' <= c && c <= 'F') return 10 + c - 'A';
    107   UNREACHABLE();
    108 }
    109 
    110 
    111 void Bignum::AssignHexString(Vector<const char> value) {
    112   Zero();
    113   int length = value.length();
    114 
    115   int needed_bigits = length * 4 / kBigitSize + 1;
    116   EnsureCapacity(needed_bigits);
    117   int string_index = length - 1;
    118   for (int i = 0; i < needed_bigits - 1; ++i) {
    119     // These bigits are guaranteed to be "full".
    120     Chunk current_bigit = 0;
    121     for (int j = 0; j < kBigitSize / 4; j++) {
    122       current_bigit += HexCharValue(value[string_index--]) << (j * 4);
    123     }
    124     bigits_[i] = current_bigit;
    125   }
    126   used_digits_ = needed_bigits - 1;
    127 
    128   Chunk most_significant_bigit = 0;  // Could be = 0;
    129   for (int j = 0; j <= string_index; ++j) {
    130     most_significant_bigit <<= 4;
    131     most_significant_bigit += HexCharValue(value[j]);
    132   }
    133   if (most_significant_bigit != 0) {
    134     bigits_[used_digits_] = most_significant_bigit;
    135     used_digits_++;
    136   }
    137   Clamp();
    138 }
    139 
    140 
    141 void Bignum::AddUInt64(uint64_t operand) {
    142   if (operand == 0) return;
    143   Bignum other;
    144   other.AssignUInt64(operand);
    145   AddBignum(other);
    146 }
    147 
    148 
    149 void Bignum::AddBignum(const Bignum& other) {
    150   DCHECK(IsClamped());
    151   DCHECK(other.IsClamped());
    152 
    153   // If this has a greater exponent than other append zero-bigits to this.
    154   // After this call exponent_ <= other.exponent_.
    155   Align(other);
    156 
    157   // There are two possibilities:
    158   //   aaaaaaaaaaa 0000  (where the 0s represent a's exponent)
    159   //     bbbbb 00000000
    160   //   ----------------
    161   //   ccccccccccc 0000
    162   // or
    163   //    aaaaaaaaaa 0000
    164   //  bbbbbbbbb 0000000
    165   //  -----------------
    166   //  cccccccccccc 0000
    167   // In both cases we might need a carry bigit.
    168 
    169   EnsureCapacity(1 + Max(BigitLength(), other.BigitLength()) - exponent_);
    170   Chunk carry = 0;
    171   int bigit_pos = other.exponent_ - exponent_;
    172   DCHECK_GE(bigit_pos, 0);
    173   for (int i = 0; i < other.used_digits_; ++i) {
    174     Chunk sum = bigits_[bigit_pos] + other.bigits_[i] + carry;
    175     bigits_[bigit_pos] = sum & kBigitMask;
    176     carry = sum >> kBigitSize;
    177     bigit_pos++;
    178   }
    179 
    180   while (carry != 0) {
    181     Chunk sum = bigits_[bigit_pos] + carry;
    182     bigits_[bigit_pos] = sum & kBigitMask;
    183     carry = sum >> kBigitSize;
    184     bigit_pos++;
    185   }
    186   used_digits_ = Max(bigit_pos, used_digits_);
    187   DCHECK(IsClamped());
    188 }
    189 
    190 
    191 void Bignum::SubtractBignum(const Bignum& other) {
    192   DCHECK(IsClamped());
    193   DCHECK(other.IsClamped());
    194   // We require this to be bigger than other.
    195   DCHECK(LessEqual(other, *this));
    196 
    197   Align(other);
    198 
    199   int offset = other.exponent_ - exponent_;
    200   Chunk borrow = 0;
    201   int i;
    202   for (i = 0; i < other.used_digits_; ++i) {
    203     DCHECK((borrow == 0) || (borrow == 1));
    204     Chunk difference = bigits_[i + offset] - other.bigits_[i] - borrow;
    205     bigits_[i + offset] = difference & kBigitMask;
    206     borrow = difference >> (kChunkSize - 1);
    207   }
    208   while (borrow != 0) {
    209     Chunk difference = bigits_[i + offset] - borrow;
    210     bigits_[i + offset] = difference & kBigitMask;
    211     borrow = difference >> (kChunkSize - 1);
    212     ++i;
    213   }
    214   Clamp();
    215 }
    216 
    217 
    218 void Bignum::ShiftLeft(int shift_amount) {
    219   if (used_digits_ == 0) return;
    220   exponent_ += shift_amount / kBigitSize;
    221   int local_shift = shift_amount % kBigitSize;
    222   EnsureCapacity(used_digits_ + 1);
    223   BigitsShiftLeft(local_shift);
    224 }
    225 
    226 
    227 void Bignum::MultiplyByUInt32(uint32_t factor) {
    228   if (factor == 1) return;
    229   if (factor == 0) {
    230     Zero();
    231     return;
    232   }
    233   if (used_digits_ == 0) return;
    234 
    235   // The product of a bigit with the factor is of size kBigitSize + 32.
    236   // Assert that this number + 1 (for the carry) fits into double chunk.
    237   DCHECK_GE(kDoubleChunkSize, kBigitSize + 32 + 1);
    238   DoubleChunk carry = 0;
    239   for (int i = 0; i < used_digits_; ++i) {
    240     DoubleChunk product = static_cast<DoubleChunk>(factor) * bigits_[i] + carry;
    241     bigits_[i] = static_cast<Chunk>(product & kBigitMask);
    242     carry = (product >> kBigitSize);
    243   }
    244   while (carry != 0) {
    245     EnsureCapacity(used_digits_ + 1);
    246     bigits_[used_digits_] = static_cast<Chunk>(carry & kBigitMask);
    247     used_digits_++;
    248     carry >>= kBigitSize;
    249   }
    250 }
    251 
    252 
    253 void Bignum::MultiplyByUInt64(uint64_t factor) {
    254   if (factor == 1) return;
    255   if (factor == 0) {
    256     Zero();
    257     return;
    258   }
    259   DCHECK_LT(kBigitSize, 32);
    260   uint64_t carry = 0;
    261   uint64_t low = factor & 0xFFFFFFFF;
    262   uint64_t high = factor >> 32;
    263   for (int i = 0; i < used_digits_; ++i) {
    264     uint64_t product_low = low * bigits_[i];
    265     uint64_t product_high = high * bigits_[i];
    266     uint64_t tmp = (carry & kBigitMask) + product_low;
    267     bigits_[i] = static_cast<Chunk>(tmp & kBigitMask);
    268     carry = (carry >> kBigitSize) + (tmp >> kBigitSize) +
    269         (product_high << (32 - kBigitSize));
    270   }
    271   while (carry != 0) {
    272     EnsureCapacity(used_digits_ + 1);
    273     bigits_[used_digits_] = static_cast<Chunk>(carry & kBigitMask);
    274     used_digits_++;
    275     carry >>= kBigitSize;
    276   }
    277 }
    278 
    279 
    280 void Bignum::MultiplyByPowerOfTen(int exponent) {
    281   const uint64_t kFive27 = V8_2PART_UINT64_C(0x6765C793, fa10079d);
    282   const uint16_t kFive1 = 5;
    283   const uint16_t kFive2 = kFive1 * 5;
    284   const uint16_t kFive3 = kFive2 * 5;
    285   const uint16_t kFive4 = kFive3 * 5;
    286   const uint16_t kFive5 = kFive4 * 5;
    287   const uint16_t kFive6 = kFive5 * 5;
    288   const uint32_t kFive7 = kFive6 * 5;
    289   const uint32_t kFive8 = kFive7 * 5;
    290   const uint32_t kFive9 = kFive8 * 5;
    291   const uint32_t kFive10 = kFive9 * 5;
    292   const uint32_t kFive11 = kFive10 * 5;
    293   const uint32_t kFive12 = kFive11 * 5;
    294   const uint32_t kFive13 = kFive12 * 5;
    295   const uint32_t kFive1_to_12[] =
    296       { kFive1, kFive2, kFive3, kFive4, kFive5, kFive6,
    297         kFive7, kFive8, kFive9, kFive10, kFive11, kFive12 };
    298 
    299   DCHECK_GE(exponent, 0);
    300   if (exponent == 0) return;
    301   if (used_digits_ == 0) return;
    302 
    303   // We shift by exponent at the end just before returning.
    304   int remaining_exponent = exponent;
    305   while (remaining_exponent >= 27) {
    306     MultiplyByUInt64(kFive27);
    307     remaining_exponent -= 27;
    308   }
    309   while (remaining_exponent >= 13) {
    310     MultiplyByUInt32(kFive13);
    311     remaining_exponent -= 13;
    312   }
    313   if (remaining_exponent > 0) {
    314     MultiplyByUInt32(kFive1_to_12[remaining_exponent - 1]);
    315   }
    316   ShiftLeft(exponent);
    317 }
    318 
    319 
    320 void Bignum::Square() {
    321   DCHECK(IsClamped());
    322   int product_length = 2 * used_digits_;
    323   EnsureCapacity(product_length);
    324 
    325   // Comba multiplication: compute each column separately.
    326   // Example: r = a2a1a0 * b2b1b0.
    327   //    r =  1    * a0b0 +
    328   //        10    * (a1b0 + a0b1) +
    329   //        100   * (a2b0 + a1b1 + a0b2) +
    330   //        1000  * (a2b1 + a1b2) +
    331   //        10000 * a2b2
    332   //
    333   // In the worst case we have to accumulate nb-digits products of digit*digit.
    334   //
    335   // Assert that the additional number of bits in a DoubleChunk are enough to
    336   // sum up used_digits of Bigit*Bigit.
    337   if ((1 << (2 * (kChunkSize - kBigitSize))) <= used_digits_) {
    338     UNIMPLEMENTED();
    339   }
    340   DoubleChunk accumulator = 0;
    341   // First shift the digits so we don't overwrite them.
    342   int copy_offset = used_digits_;
    343   for (int i = 0; i < used_digits_; ++i) {
    344     bigits_[copy_offset + i] = bigits_[i];
    345   }
    346   // We have two loops to avoid some 'if's in the loop.
    347   for (int i = 0; i < used_digits_; ++i) {
    348     // Process temporary digit i with power i.
    349     // The sum of the two indices must be equal to i.
    350     int bigit_index1 = i;
    351     int bigit_index2 = 0;
    352     // Sum all of the sub-products.
    353     while (bigit_index1 >= 0) {
    354       Chunk chunk1 = bigits_[copy_offset + bigit_index1];
    355       Chunk chunk2 = bigits_[copy_offset + bigit_index2];
    356       accumulator += static_cast<DoubleChunk>(chunk1) * chunk2;
    357       bigit_index1--;
    358       bigit_index2++;
    359     }
    360     bigits_[i] = static_cast<Chunk>(accumulator) & kBigitMask;
    361     accumulator >>= kBigitSize;
    362   }
    363   for (int i = used_digits_; i < product_length; ++i) {
    364     int bigit_index1 = used_digits_ - 1;
    365     int bigit_index2 = i - bigit_index1;
    366     // Invariant: sum of both indices is again equal to i.
    367     // Inner loop runs 0 times on last iteration, emptying accumulator.
    368     while (bigit_index2 < used_digits_) {
    369       Chunk chunk1 = bigits_[copy_offset + bigit_index1];
    370       Chunk chunk2 = bigits_[copy_offset + bigit_index2];
    371       accumulator += static_cast<DoubleChunk>(chunk1) * chunk2;
    372       bigit_index1--;
    373       bigit_index2++;
    374     }
    375     // The overwritten bigits_[i] will never be read in further loop iterations,
    376     // because bigit_index1 and bigit_index2 are always greater
    377     // than i - used_digits_.
    378     bigits_[i] = static_cast<Chunk>(accumulator) & kBigitMask;
    379     accumulator >>= kBigitSize;
    380   }
    381   // Since the result was guaranteed to lie inside the number the
    382   // accumulator must be 0 now.
    383   DCHECK_EQ(accumulator, 0);
    384 
    385   // Don't forget to update the used_digits and the exponent.
    386   used_digits_ = product_length;
    387   exponent_ *= 2;
    388   Clamp();
    389 }
    390 
    391 
    392 void Bignum::AssignPowerUInt16(uint16_t base, int power_exponent) {
    393   DCHECK_NE(base, 0);
    394   DCHECK_GE(power_exponent, 0);
    395   if (power_exponent == 0) {
    396     AssignUInt16(1);
    397     return;
    398   }
    399   Zero();
    400   int shifts = 0;
    401   // We expect base to be in range 2-32, and most often to be 10.
    402   // It does not make much sense to implement different algorithms for counting
    403   // the bits.
    404   while ((base & 1) == 0) {
    405     base >>= 1;
    406     shifts++;
    407   }
    408   int bit_size = 0;
    409   int tmp_base = base;
    410   while (tmp_base != 0) {
    411     tmp_base >>= 1;
    412     bit_size++;
    413   }
    414   int final_size = bit_size * power_exponent;
    415   // 1 extra bigit for the shifting, and one for rounded final_size.
    416   EnsureCapacity(final_size / kBigitSize + 2);
    417 
    418   // Left to Right exponentiation.
    419   int mask = 1;
    420   while (power_exponent >= mask) mask <<= 1;
    421 
    422   // The mask is now pointing to the bit above the most significant 1-bit of
    423   // power_exponent.
    424   // Get rid of first 1-bit;
    425   mask >>= 2;
    426   uint64_t this_value = base;
    427 
    428   bool delayed_multipliciation = false;
    429   const uint64_t max_32bits = 0xFFFFFFFF;
    430   while (mask != 0 && this_value <= max_32bits) {
    431     this_value = this_value * this_value;
    432     // Verify that there is enough space in this_value to perform the
    433     // multiplication.  The first bit_size bits must be 0.
    434     if ((power_exponent & mask) != 0) {
    435       uint64_t base_bits_mask =
    436           ~((static_cast<uint64_t>(1) << (64 - bit_size)) - 1);
    437       bool high_bits_zero = (this_value & base_bits_mask) == 0;
    438       if (high_bits_zero) {
    439         this_value *= base;
    440       } else {
    441         delayed_multipliciation = true;
    442       }
    443     }
    444     mask >>= 1;
    445   }
    446   AssignUInt64(this_value);
    447   if (delayed_multipliciation) {
    448     MultiplyByUInt32(base);
    449   }
    450 
    451   // Now do the same thing as a bignum.
    452   while (mask != 0) {
    453     Square();
    454     if ((power_exponent & mask) != 0) {
    455       MultiplyByUInt32(base);
    456     }
    457     mask >>= 1;
    458   }
    459 
    460   // And finally add the saved shifts.
    461   ShiftLeft(shifts * power_exponent);
    462 }
    463 
    464 
    465 // Precondition: this/other < 16bit.
    466 uint16_t Bignum::DivideModuloIntBignum(const Bignum& other) {
    467   DCHECK(IsClamped());
    468   DCHECK(other.IsClamped());
    469   DCHECK_GT(other.used_digits_, 0);
    470 
    471   // Easy case: if we have less digits than the divisor than the result is 0.
    472   // Note: this handles the case where this == 0, too.
    473   if (BigitLength() < other.BigitLength()) {
    474     return 0;
    475   }
    476 
    477   Align(other);
    478 
    479   uint16_t result = 0;
    480 
    481   // Start by removing multiples of 'other' until both numbers have the same
    482   // number of digits.
    483   while (BigitLength() > other.BigitLength()) {
    484     // This naive approach is extremely inefficient if the this divided other
    485     // might be big. This function is implemented for doubleToString where
    486     // the result should be small (less than 10).
    487     DCHECK(other.bigits_[other.used_digits_ - 1] >= ((1 << kBigitSize) / 16));
    488     // Remove the multiples of the first digit.
    489     // Example this = 23 and other equals 9. -> Remove 2 multiples.
    490     result += bigits_[used_digits_ - 1];
    491     SubtractTimes(other, bigits_[used_digits_ - 1]);
    492   }
    493 
    494   DCHECK(BigitLength() == other.BigitLength());
    495 
    496   // Both bignums are at the same length now.
    497   // Since other has more than 0 digits we know that the access to
    498   // bigits_[used_digits_ - 1] is safe.
    499   Chunk this_bigit = bigits_[used_digits_ - 1];
    500   Chunk other_bigit = other.bigits_[other.used_digits_ - 1];
    501 
    502   if (other.used_digits_ == 1) {
    503     // Shortcut for easy (and common) case.
    504     int quotient = this_bigit / other_bigit;
    505     bigits_[used_digits_ - 1] = this_bigit - other_bigit * quotient;
    506     result += quotient;
    507     Clamp();
    508     return result;
    509   }
    510 
    511   int division_estimate = this_bigit / (other_bigit + 1);
    512   result += division_estimate;
    513   SubtractTimes(other, division_estimate);
    514 
    515   if (other_bigit * (division_estimate + 1) > this_bigit) {
    516     // No need to even try to subtract. Even if other's remaining digits were 0
    517     // another subtraction would be too much.
    518     return result;
    519   }
    520 
    521   while (LessEqual(other, *this)) {
    522     SubtractBignum(other);
    523     result++;
    524   }
    525   return result;
    526 }
    527 
    528 
    529 template<typename S>
    530 static int SizeInHexChars(S number) {
    531   DCHECK_GT(number, 0);
    532   int result = 0;
    533   while (number != 0) {
    534     number >>= 4;
    535     result++;
    536   }
    537   return result;
    538 }
    539 
    540 
    541 bool Bignum::ToHexString(char* buffer, int buffer_size) const {
    542   DCHECK(IsClamped());
    543   // Each bigit must be printable as separate hex-character.
    544   DCHECK_EQ(kBigitSize % 4, 0);
    545   const int kHexCharsPerBigit = kBigitSize / 4;
    546 
    547   if (used_digits_ == 0) {
    548     if (buffer_size < 2) return false;
    549     buffer[0] = '0';
    550     buffer[1] = '\0';
    551     return true;
    552   }
    553   // We add 1 for the terminating '\0' character.
    554   int needed_chars = (BigitLength() - 1) * kHexCharsPerBigit +
    555       SizeInHexChars(bigits_[used_digits_ - 1]) + 1;
    556   if (needed_chars > buffer_size) return false;
    557   int string_index = needed_chars - 1;
    558   buffer[string_index--] = '\0';
    559   for (int i = 0; i < exponent_; ++i) {
    560     for (int j = 0; j < kHexCharsPerBigit; ++j) {
    561       buffer[string_index--] = '0';
    562     }
    563   }
    564   for (int i = 0; i < used_digits_ - 1; ++i) {
    565     Chunk current_bigit = bigits_[i];
    566     for (int j = 0; j < kHexCharsPerBigit; ++j) {
    567       buffer[string_index--] = HexCharOfValue(current_bigit & 0xF);
    568       current_bigit >>= 4;
    569     }
    570   }
    571   // And finally the last bigit.
    572   Chunk most_significant_bigit = bigits_[used_digits_ - 1];
    573   while (most_significant_bigit != 0) {
    574     buffer[string_index--] = HexCharOfValue(most_significant_bigit & 0xF);
    575     most_significant_bigit >>= 4;
    576   }
    577   return true;
    578 }
    579 
    580 
    581 Bignum::Chunk Bignum::BigitAt(int index) const {
    582   if (index >= BigitLength()) return 0;
    583   if (index < exponent_) return 0;
    584   return bigits_[index - exponent_];
    585 }
    586 
    587 
    588 int Bignum::Compare(const Bignum& a, const Bignum& b) {
    589   DCHECK(a.IsClamped());
    590   DCHECK(b.IsClamped());
    591   int bigit_length_a = a.BigitLength();
    592   int bigit_length_b = b.BigitLength();
    593   if (bigit_length_a < bigit_length_b) return -1;
    594   if (bigit_length_a > bigit_length_b) return +1;
    595   for (int i = bigit_length_a - 1; i >= Min(a.exponent_, b.exponent_); --i) {
    596     Chunk bigit_a = a.BigitAt(i);
    597     Chunk bigit_b = b.BigitAt(i);
    598     if (bigit_a < bigit_b) return -1;
    599     if (bigit_a > bigit_b) return +1;
    600     // Otherwise they are equal up to this digit. Try the next digit.
    601   }
    602   return 0;
    603 }
    604 
    605 
    606 int Bignum::PlusCompare(const Bignum& a, const Bignum& b, const Bignum& c) {
    607   DCHECK(a.IsClamped());
    608   DCHECK(b.IsClamped());
    609   DCHECK(c.IsClamped());
    610   if (a.BigitLength() < b.BigitLength()) {
    611     return PlusCompare(b, a, c);
    612   }
    613   if (a.BigitLength() + 1 < c.BigitLength()) return -1;
    614   if (a.BigitLength() > c.BigitLength()) return +1;
    615   // The exponent encodes 0-bigits. So if there are more 0-digits in 'a' than
    616   // 'b' has digits, then the bigit-length of 'a'+'b' must be equal to the one
    617   // of 'a'.
    618   if (a.exponent_ >= b.BigitLength() && a.BigitLength() < c.BigitLength()) {
    619     return -1;
    620   }
    621 
    622   Chunk borrow = 0;
    623   // Starting at min_exponent all digits are == 0. So no need to compare them.
    624   int min_exponent = Min(Min(a.exponent_, b.exponent_), c.exponent_);
    625   for (int i = c.BigitLength() - 1; i >= min_exponent; --i) {
    626     Chunk chunk_a = a.BigitAt(i);
    627     Chunk chunk_b = b.BigitAt(i);
    628     Chunk chunk_c = c.BigitAt(i);
    629     Chunk sum = chunk_a + chunk_b;
    630     if (sum > chunk_c + borrow) {
    631       return +1;
    632     } else {
    633       borrow = chunk_c + borrow - sum;
    634       if (borrow > 1) return -1;
    635       borrow <<= kBigitSize;
    636     }
    637   }
    638   if (borrow == 0) return 0;
    639   return -1;
    640 }
    641 
    642 
    643 void Bignum::Clamp() {
    644   while (used_digits_ > 0 && bigits_[used_digits_ - 1] == 0) {
    645     used_digits_--;
    646   }
    647   if (used_digits_ == 0) {
    648     // Zero.
    649     exponent_ = 0;
    650   }
    651 }
    652 
    653 
    654 bool Bignum::IsClamped() const {
    655   return used_digits_ == 0 || bigits_[used_digits_ - 1] != 0;
    656 }
    657 
    658 
    659 void Bignum::Zero() {
    660   for (int i = 0; i < used_digits_; ++i) {
    661     bigits_[i] = 0;
    662   }
    663   used_digits_ = 0;
    664   exponent_ = 0;
    665 }
    666 
    667 
    668 void Bignum::Align(const Bignum& other) {
    669   if (exponent_ > other.exponent_) {
    670     // If "X" represents a "hidden" digit (by the exponent) then we are in the
    671     // following case (a == this, b == other):
    672     // a:  aaaaaaXXXX   or a:   aaaaaXXX
    673     // b:     bbbbbbX      b: bbbbbbbbXX
    674     // We replace some of the hidden digits (X) of a with 0 digits.
    675     // a:  aaaaaa000X   or a:   aaaaa0XX
    676     int zero_digits = exponent_ - other.exponent_;
    677     EnsureCapacity(used_digits_ + zero_digits);
    678     for (int i = used_digits_ - 1; i >= 0; --i) {
    679       bigits_[i + zero_digits] = bigits_[i];
    680     }
    681     for (int i = 0; i < zero_digits; ++i) {
    682       bigits_[i] = 0;
    683     }
    684     used_digits_ += zero_digits;
    685     exponent_ -= zero_digits;
    686     DCHECK_GE(used_digits_, 0);
    687     DCHECK_GE(exponent_, 0);
    688   }
    689 }
    690 
    691 
    692 void Bignum::BigitsShiftLeft(int shift_amount) {
    693   DCHECK_LT(shift_amount, kBigitSize);
    694   DCHECK_GE(shift_amount, 0);
    695   Chunk carry = 0;
    696   for (int i = 0; i < used_digits_; ++i) {
    697     Chunk new_carry = bigits_[i] >> (kBigitSize - shift_amount);
    698     bigits_[i] = ((bigits_[i] << shift_amount) + carry) & kBigitMask;
    699     carry = new_carry;
    700   }
    701   if (carry != 0) {
    702     bigits_[used_digits_] = carry;
    703     used_digits_++;
    704   }
    705 }
    706 
    707 
    708 void Bignum::SubtractTimes(const Bignum& other, int factor) {
    709 #ifdef DEBUG
    710   Bignum a, b;
    711   a.AssignBignum(*this);
    712   b.AssignBignum(other);
    713   b.MultiplyByUInt32(factor);
    714   a.SubtractBignum(b);
    715 #endif
    716   DCHECK(exponent_ <= other.exponent_);
    717   if (factor < 3) {
    718     for (int i = 0; i < factor; ++i) {
    719       SubtractBignum(other);
    720     }
    721     return;
    722   }
    723   Chunk borrow = 0;
    724   int exponent_diff = other.exponent_ - exponent_;
    725   for (int i = 0; i < other.used_digits_; ++i) {
    726     DoubleChunk product = static_cast<DoubleChunk>(factor) * other.bigits_[i];
    727     DoubleChunk remove = borrow + product;
    728     Chunk difference =
    729         bigits_[i + exponent_diff] - static_cast<Chunk>(remove & kBigitMask);
    730     bigits_[i + exponent_diff] = difference & kBigitMask;
    731     borrow = static_cast<Chunk>((difference >> (kChunkSize - 1)) +
    732                                 (remove >> kBigitSize));
    733   }
    734   for (int i = other.used_digits_ + exponent_diff; i < used_digits_; ++i) {
    735     if (borrow == 0) return;
    736     Chunk difference = bigits_[i] - borrow;
    737     bigits_[i] = difference & kBigitMask;
    738     borrow = difference >> (kChunkSize - 1);
    739   }
    740   Clamp();
    741   DCHECK(Bignum::Equal(a, *this));
    742 }
    743 
    744 
    745 }  // namespace internal
    746 }  // namespace v8
    747