binary-to-decimal.cpp
11.3 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
//===-- lib/Decimal/binary-to-decimal.cpp ---------------------------------===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
#include "big-radix-floating-point.h"
#include "flang/Decimal/decimal.h"
#include <cassert>
#include <string>
namespace Fortran::decimal {
template <int PREC, int LOG10RADIX>
BigRadixFloatingPointNumber<PREC, LOG10RADIX>::BigRadixFloatingPointNumber(
BinaryFloatingPointNumber<PREC> x, enum FortranRounding rounding)
: rounding_{rounding} {
bool negative{x.IsNegative()};
if (x.IsZero()) {
isNegative_ = negative;
return;
}
if (negative) {
x.Negate();
}
int twoPow{x.UnbiasedExponent()};
twoPow -= x.bits - 1;
if (!x.isImplicitMSB) {
++twoPow;
}
int lshift{x.exponentBits};
if (twoPow <= -lshift) {
twoPow += lshift;
lshift = 0;
} else if (twoPow < 0) {
lshift += twoPow;
twoPow = 0;
}
auto word{x.Fraction()};
word <<= lshift;
SetTo(word);
isNegative_ = negative;
// The significand is now encoded in *this as an integer (D) and
// decimal exponent (E): x = D * 10.**E * 2.**twoPow
// twoPow can be positive or negative.
// The goal now is to get twoPow up or down to zero, leaving us with
// only decimal digits and decimal exponent. This is done by
// fast multiplications and divisions of D by 2 and 5.
// (5*D) * 10.**E * 2.**twoPow -> D * 10.**(E+1) * 2.**(twoPow-1)
for (; twoPow > 0 && IsDivisibleBy<5>(); --twoPow) {
DivideBy<5>();
++exponent_;
}
int overflow{0};
for (; twoPow >= 9; twoPow -= 9) {
// D * 10.**E * 2.**twoPow -> (D*(2**9)) * 10.**E * 2.**(twoPow-9)
overflow |= MultiplyBy<512>();
}
for (; twoPow >= 3; twoPow -= 3) {
// D * 10.**E * 2.**twoPow -> (D*(2**3)) * 10.**E * 2.**(twoPow-3)
overflow |= MultiplyBy<8>();
}
for (; twoPow > 0; --twoPow) {
// D * 10.**E * 2.**twoPow -> (2*D) * 10.**E * 2.**(twoPow-1)
overflow |= MultiplyBy<2>();
}
overflow |= DivideByPowerOfTwoInPlace(-twoPow);
assert(overflow == 0);
Normalize();
}
template <int PREC, int LOG10RADIX>
ConversionToDecimalResult
BigRadixFloatingPointNumber<PREC, LOG10RADIX>::ConvertToDecimal(char *buffer,
std::size_t n, enum DecimalConversionFlags flags, int maxDigits) const {
if (n < static_cast<std::size_t>(3 + digits_ * LOG10RADIX)) {
return {nullptr, 0, 0, Overflow};
}
char *start{buffer};
if (isNegative_) {
*start++ = '-';
} else if (flags & AlwaysSign) {
*start++ = '+';
}
if (IsZero()) {
*start++ = '0';
*start = '\0';
return {buffer, static_cast<std::size_t>(start - buffer), 0, Exact};
}
char *p{start};
static_assert((LOG10RADIX % 2) == 0, "radix not a power of 100");
static const char lut[] = "0001020304050607080910111213141516171819"
"2021222324252627282930313233343536373839"
"4041424344454647484950515253545556575859"
"6061626364656667686970717273747576777879"
"8081828384858687888990919293949596979899";
static constexpr Digit hundredth{radix / 100};
// Treat the MSD specially: don't emit leading zeroes.
Digit dig{digit_[digits_ - 1]};
for (int k{0}; k < LOG10RADIX; k += 2) {
Digit d{common::DivideUnsignedBy<Digit, hundredth>(dig)};
dig = 100 * (dig - d * hundredth);
const char *q{lut + 2 * d};
if (q[0] != '0' || p > start) {
*p++ = q[0];
*p++ = q[1];
} else if (q[1] != '0') {
*p++ = q[1];
}
}
for (int j{digits_ - 1}; j-- > 0;) {
Digit dig{digit_[j]};
for (int k{0}; k < log10Radix; k += 2) {
Digit d{common::DivideUnsignedBy<Digit, hundredth>(dig)};
dig = 100 * (dig - d * hundredth);
const char *q{lut + 2 * d};
*p++ = q[0];
*p++ = q[1];
}
}
// Adjust exponent so the effective decimal point is to
// the left of the first digit.
int expo = exponent_ + p - start;
// Trim trailing zeroes.
while (p[-1] == '0') {
--p;
}
char *end{start + maxDigits};
if (maxDigits == 0) {
p = end;
}
if (p <= end) {
*p = '\0';
return {buffer, static_cast<std::size_t>(p - buffer), expo, Exact};
} else {
// Apply a digit limit, possibly with rounding.
bool incr{false};
switch (rounding_) {
case RoundNearest:
case RoundDefault:
incr = *end > '5' ||
(*end == '5' && (p > end + 1 || ((end[-1] - '0') & 1) != 0));
break;
case RoundUp:
incr = !isNegative_;
break;
case RoundDown:
incr = isNegative_;
break;
case RoundToZero:
break;
case RoundCompatible:
incr = *end >= '5';
break;
}
p = end;
if (incr) {
while (p > start && p[-1] == '9') {
--p;
}
if (p == start) {
*p++ = '1';
++expo;
} else {
++p[-1];
}
}
*p = '\0';
return {buffer, static_cast<std::size_t>(p - buffer), expo, Inexact};
}
}
template <int PREC, int LOG10RADIX>
bool BigRadixFloatingPointNumber<PREC, LOG10RADIX>::Mean(
const BigRadixFloatingPointNumber &that) {
while (digits_ < that.digits_) {
digit_[digits_++] = 0;
}
int carry{0};
for (int j{0}; j < that.digits_; ++j) {
Digit v{digit_[j] + that.digit_[j] + carry};
if (v >= radix) {
digit_[j] = v - radix;
carry = 1;
} else {
digit_[j] = v;
carry = 0;
}
}
if (carry != 0) {
AddCarry(that.digits_, carry);
}
return DivideBy<2>() != 0;
}
template <int PREC, int LOG10RADIX>
void BigRadixFloatingPointNumber<PREC, LOG10RADIX>::Minimize(
BigRadixFloatingPointNumber &&less, BigRadixFloatingPointNumber &&more) {
int leastExponent{exponent_};
if (less.exponent_ < leastExponent) {
leastExponent = less.exponent_;
}
if (more.exponent_ < leastExponent) {
leastExponent = more.exponent_;
}
while (exponent_ > leastExponent) {
--exponent_;
MultiplyBy<10>();
}
while (less.exponent_ > leastExponent) {
--less.exponent_;
less.MultiplyBy<10>();
}
while (more.exponent_ > leastExponent) {
--more.exponent_;
more.MultiplyBy<10>();
}
if (less.Mean(*this)) {
less.AddCarry(); // round up
}
if (!more.Mean(*this)) {
more.Decrement(); // round down
}
while (less.digits_ < more.digits_) {
less.digit_[less.digits_++] = 0;
}
while (more.digits_ < less.digits_) {
more.digit_[more.digits_++] = 0;
}
int digits{more.digits_};
int same{0};
while (same < digits &&
less.digit_[digits - 1 - same] == more.digit_[digits - 1 - same]) {
++same;
}
if (same == digits) {
return;
}
digits_ = same + 1;
int offset{digits - digits_};
exponent_ += offset * log10Radix;
for (int j{0}; j < digits_; ++j) {
digit_[j] = more.digit_[j + offset];
}
Digit least{less.digit_[offset]};
Digit my{digit_[0]};
while (true) {
Digit q{common::DivideUnsignedBy<Digit, 10>(my)};
Digit r{my - 10 * q};
Digit lq{common::DivideUnsignedBy<Digit, 10>(least)};
Digit lr{least - 10 * lq};
if (r != 0 && lq == q) {
Digit sub{(r - lr) >> 1};
digit_[0] -= sub;
break;
} else {
least = lq;
my = q;
DivideBy<10>();
++exponent_;
}
}
Normalize();
}
template <int PREC>
ConversionToDecimalResult ConvertToDecimal(char *buffer, std::size_t size,
enum DecimalConversionFlags flags, int digits,
enum FortranRounding rounding, BinaryFloatingPointNumber<PREC> x) {
if (x.IsNaN()) {
return {"NaN", 3, 0, Invalid};
} else if (x.IsInfinite()) {
if (x.IsNegative()) {
return {"-Inf", 4, 0, Exact};
} else if (flags & AlwaysSign) {
return {"+Inf", 4, 0, Exact};
} else {
return {"Inf", 3, 0, Exact};
}
} else {
using Big = BigRadixFloatingPointNumber<PREC>;
Big number{x, rounding};
if ((flags & Minimize) && !x.IsZero()) {
// To emit the fewest decimal digits necessary to represent the value
// in such a way that decimal-to-binary conversion to the same format
// with a fixed assumption about rounding will return the same binary
// value, we also perform binary-to-decimal conversion on the two
// binary values immediately adjacent to this one, use them to identify
// the bounds of the range of decimal values that will map back to the
// original binary value, and find a (not necessary unique) shortest
// decimal sequence in that range.
using Binary = typename Big::Real;
Binary less{x};
less.Previous();
Binary more{x};
if (!x.IsMaximalFiniteMagnitude()) {
more.Next();
}
number.Minimize(Big{less, rounding}, Big{more, rounding});
} else {
}
return number.ConvertToDecimal(buffer, size, flags, digits);
}
}
template ConversionToDecimalResult ConvertToDecimal<8>(char *, std::size_t,
enum DecimalConversionFlags, int, enum FortranRounding,
BinaryFloatingPointNumber<8>);
template ConversionToDecimalResult ConvertToDecimal<11>(char *, std::size_t,
enum DecimalConversionFlags, int, enum FortranRounding,
BinaryFloatingPointNumber<11>);
template ConversionToDecimalResult ConvertToDecimal<24>(char *, std::size_t,
enum DecimalConversionFlags, int, enum FortranRounding,
BinaryFloatingPointNumber<24>);
template ConversionToDecimalResult ConvertToDecimal<53>(char *, std::size_t,
enum DecimalConversionFlags, int, enum FortranRounding,
BinaryFloatingPointNumber<53>);
template ConversionToDecimalResult ConvertToDecimal<64>(char *, std::size_t,
enum DecimalConversionFlags, int, enum FortranRounding,
BinaryFloatingPointNumber<64>);
template ConversionToDecimalResult ConvertToDecimal<113>(char *, std::size_t,
enum DecimalConversionFlags, int, enum FortranRounding,
BinaryFloatingPointNumber<113>);
extern "C" {
ConversionToDecimalResult ConvertFloatToDecimal(char *buffer, std::size_t size,
enum DecimalConversionFlags flags, int digits,
enum FortranRounding rounding, float x) {
return Fortran::decimal::ConvertToDecimal(buffer, size, flags, digits,
rounding, Fortran::decimal::BinaryFloatingPointNumber<24>(x));
}
ConversionToDecimalResult ConvertDoubleToDecimal(char *buffer, std::size_t size,
enum DecimalConversionFlags flags, int digits,
enum FortranRounding rounding, double x) {
return Fortran::decimal::ConvertToDecimal(buffer, size, flags, digits,
rounding, Fortran::decimal::BinaryFloatingPointNumber<53>(x));
}
#if __x86_64__ && !defined(_MSC_VER)
ConversionToDecimalResult ConvertLongDoubleToDecimal(char *buffer,
std::size_t size, enum DecimalConversionFlags flags, int digits,
enum FortranRounding rounding, long double x) {
return Fortran::decimal::ConvertToDecimal(buffer, size, flags, digits,
rounding, Fortran::decimal::BinaryFloatingPointNumber<64>(x));
}
#endif
}
template <int PREC, int LOG10RADIX>
template <typename STREAM>
STREAM &BigRadixFloatingPointNumber<PREC, LOG10RADIX>::Dump(STREAM &o) const {
if (isNegative_) {
o << '-';
}
o << "10**(" << exponent_ << ") * ...\n";
for (int j{digits_}; --j >= 0;) {
std::string str{std::to_string(digit_[j])};
o << std::string(20 - str.size(), ' ') << str << " [" << j << ']';
if (j + 1 == digitLimit_) {
o << " (limit)";
}
o << '\n';
}
return o;
}
} // namespace Fortran::decimal