pytorch

1
// This file is based on the uint128 implementation of protobuf at
2
// https://github.com/protocolbuffers/protobuf/blob/1e88936fce10cf773cb72b44c6a7f48b38c7578b/src/google/protobuf/stubs/int128.cc
3
//
4
// Protocol Buffers - Google's data interchange format
5
// Copyright 2008 Google Inc.  All rights reserved.
6
// https://developers.google.com/protocol-buffers/
7
//
8
// Redistribution and use in source and binary forms, with or without
9
// modification, are permitted provided that the following conditions are
10
// met:
11
//
12
//     * Redistributions of source code must retain the above copyright
13
// notice, this list of conditions and the following disclaimer.
14
//     * Redistributions in binary form must reproduce the above
15
// copyright notice, this list of conditions and the following disclaimer
16
// in the documentation and/or other materials provided with the
17
// distribution.
18
//     * Neither the name of Google Inc. nor the names of its
19
// contributors may be used to endorse or promote products derived from
20
// this software without specific prior written permission.
21
//
22
// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
23
// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
24
// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
25
// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
26
// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
27
// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
28
// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
29
// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
30
// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
31
// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
32
// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
33

34
#include <c10/util/Logging.h>
35
#include <c10/util/int128.h>
36
#include <iomanip>
37
#include <ostream> // NOLINT(readability/streams)
38

39
namespace c10 {
40

41
const uint128_pod kuint128max = {
42
    uint64_t{0xFFFFFFFFFFFFFFFFu},
43
    uint64_t{0xFFFFFFFFFFFFFFFFu}};
44

45
// Returns the 0-based position of the last set bit (i.e., most significant bit)
46
// in the given uint64. The argument may not be 0.
47
//
48
// For example:
49
//   Given: 5 (decimal) == 101 (binary)
50
//   Returns: 2
51
#define STEP(T, n, pos, sh)                   \
52
  do {                                        \
53
    if ((n) >= (static_cast<T>(1) << (sh))) { \
54
      (n) = (n) >> (sh);                      \
55
      (pos) |= (sh);                          \
56
    }                                         \
57
  } while (0)
58
static inline int Fls64(uint64_t n) {
59
  //   GOOGLE_DCHECK_NE(0, n);
60
  uint64_t pos = 0;
61
  STEP(uint64_t, n, pos, 0x20);
62
  uint32_t n32 = n;
63
  STEP(uint32_t, n32, pos, 0x10);
64
  STEP(uint32_t, n32, pos, 0x08);
65
  STEP(uint32_t, n32, pos, 0x04);
66
  return static_cast<int>(
67
      pos + ((uint64_t{0x3333333322221100u} >> (n32 << 2)) & 0x3));
68
}
69
#undef STEP
70

71
// Like Fls64() above, but returns the 0-based position of the last set bit
72
// (i.e., most significant bit) in the given uint128. The argument may not be 0.
73
static inline int Fls128(uint128 n) {
74
  if (uint64_t hi = Uint128High64(n)) {
75
    return Fls64(hi) + 64;
76
  }
77
  return Fls64(Uint128Low64(n));
78
}
79

80
void uint128::DivModImpl(
81
    uint128 dividend,
82
    uint128 divisor,
83
    uint128* quotient_ret,
84
    uint128* remainder_ret) {
85
  if (divisor == 0) {
86
    LOG(FATAL) << "Division or mod by zero: dividend.hi=" << dividend.hi_
87
               << ", lo=" << dividend.lo_;
88
  } else if (dividend < divisor) {
89
    *quotient_ret = 0;
90
    *remainder_ret = dividend;
91
    return;
92
  } else {
93
    int dividend_bit_length = Fls128(dividend);
94
    int divisor_bit_length = Fls128(divisor);
95
    int difference = dividend_bit_length - divisor_bit_length;
96
    uint128 quotient = 0;
97
    while (difference >= 0) {
98
      quotient <<= 1;
99
      uint128 shifted_divisor = divisor << difference;
100
      if (shifted_divisor <= dividend) {
101
        dividend -= shifted_divisor;
102
        quotient += 1;
103
      }
104
      difference -= 1;
105
    }
106
    // record the final quotient and remainder
107
    *quotient_ret = quotient;
108
    *remainder_ret = dividend;
109
  }
110
}
111

112
uint128& uint128::operator/=(const uint128& divisor) {
113
  uint128 quotient = 0;
114
  uint128 remainder = 0;
115
  DivModImpl(*this, divisor, &quotient, &remainder);
116
  *this = quotient;
117
  return *this;
118
}
119
uint128& uint128::operator%=(const uint128& divisor) {
120
  uint128 quotient = 0;
121
  uint128 remainder = 0;
122
  DivModImpl(*this, divisor, &quotient, &remainder);
123
  *this = remainder;
124
  return *this;
125
}
126

127
std::ostream& operator<<(std::ostream& o, const uint128& b) {
128
  std::ios_base::fmtflags flags = o.flags();
129

130
  // Select a divisor which is the largest power of the base < 2^64.
131
  uint128 div;
132
  int div_base_log = 0;
133
  switch (flags & std::ios::basefield) {
134
    case std::ios::hex:
135
      div = (uint64_t)0x1000000000000000u; // 16^15
136
      div_base_log = 15;
137
      break;
138
    case std::ios::oct:
139
      div = (uint64_t)01000000000000000000000u; // 8^21
140
      div_base_log = 21;
141
      break;
142
    default: // std::ios::dec
143
      div = (uint64_t)10000000000000000000u; // 10^19
144
      div_base_log = 19;
145
      break;
146
  }
147

148
  // Now piece together the uint128 representation from three chunks of
149
  // the original value, each less than "div" and therefore representable
150
  // as a uint64.
151
  std::ostringstream os;
152
  std::ios_base::fmtflags copy_mask =
153
      std::ios::basefield | std::ios::showbase | std::ios::uppercase;
154
  os.setf(flags & copy_mask, copy_mask);
155
  uint128 high = b;
156
  uint128 low;
157
  uint128::DivModImpl(high, div, &high, &low);
158
  uint128 mid;
159
  uint128::DivModImpl(high, div, &high, &mid);
160
  if (high.lo_ != 0) {
161
    os << high.lo_;
162
    os << std::noshowbase << std::setfill('0') << std::setw(div_base_log);
163
    os << mid.lo_;
164
    os << std::setw(div_base_log);
165
  } else if (mid.lo_ != 0) {
166
    os << mid.lo_;
167
    os << std::noshowbase << std::setfill('0') << std::setw(div_base_log);
168
  }
169
  os << low.lo_;
170
  std::string rep = os.str();
171

172
  // Add the requisite padding.
173
  std::streamsize width = o.width(0);
174
  if (width > static_cast<std::streamsize>(rep.size())) {
175
    if ((flags & std::ios::adjustfield) == std::ios::left) {
176
      rep.append(width - rep.size(), o.fill());
177
    } else {
178
      rep.insert(
179
          static_cast<std::string::size_type>(0), width - rep.size(), o.fill());
180
    }
181
  }
182

183
  // Stream the final representation in a single "<<" call.
184
  return o << rep;
185
}
186

187
} // namespace c10
188