1 //===-- High Precision Decimal ----------------------------------*- C++ -*-===// 2 // 3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 4 // See httpss//llvm.org/LICENSE.txt for license information. 5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 6 // 7 //===----------------------------------------------------------------------===// 8 9 #ifndef LLVM_LIBC_SRC___SUPPORT_HIGH_PRECISION_DECIMAL_H 10 #define LLVM_LIBC_SRC___SUPPORT_HIGH_PRECISION_DECIMAL_H 11 12 #include "src/__support/CPP/limits.h" 13 #include "src/__support/ctype_utils.h" 14 #include "src/__support/str_to_integer.h" 15 #include <stdint.h> 16 17 namespace LIBC_NAMESPACE { 18 namespace internal { 19 20 struct LShiftTableEntry { 21 uint32_t new_digits; 22 char const *power_of_five; 23 }; 24 25 // This is used in both this file and in the main str_to_float.h. 26 // TODO: Figure out where to put this. 27 enum class RoundDirection { Up, Down, Nearest }; 28 29 // This is based on the HPD data structure described as part of the Simple 30 // Decimal Conversion algorithm by Nigel Tao, described at this link: 31 // https://nigeltao.github.io/blog/2020/parse-number-f64-simple.html 32 class HighPrecisionDecimal { 33 34 // This precomputed table speeds up left shifts by having the number of new 35 // digits that will be added by multiplying 5^i by 2^i. If the number is less 36 // than 5^i then it will add one fewer digit. There are only 60 entries since 37 // that's the max shift amount. 38 // This table was generated by the script at 39 // libc/utils/mathtools/GenerateHPDConstants.py 40 static constexpr LShiftTableEntry LEFT_SHIFT_DIGIT_TABLE[] = { 41 {0, ""}, 42 {1, "5"}, 43 {1, "25"}, 44 {1, "125"}, 45 {2, "625"}, 46 {2, "3125"}, 47 {2, "15625"}, 48 {3, "78125"}, 49 {3, "390625"}, 50 {3, "1953125"}, 51 {4, "9765625"}, 52 {4, "48828125"}, 53 {4, "244140625"}, 54 {4, "1220703125"}, 55 {5, "6103515625"}, 56 {5, "30517578125"}, 57 {5, "152587890625"}, 58 {6, "762939453125"}, 59 {6, "3814697265625"}, 60 {6, "19073486328125"}, 61 {7, "95367431640625"}, 62 {7, "476837158203125"}, 63 {7, "2384185791015625"}, 64 {7, "11920928955078125"}, 65 {8, "59604644775390625"}, 66 {8, "298023223876953125"}, 67 {8, "1490116119384765625"}, 68 {9, "7450580596923828125"}, 69 {9, "37252902984619140625"}, 70 {9, "186264514923095703125"}, 71 {10, "931322574615478515625"}, 72 {10, "4656612873077392578125"}, 73 {10, "23283064365386962890625"}, 74 {10, "116415321826934814453125"}, 75 {11, "582076609134674072265625"}, 76 {11, "2910383045673370361328125"}, 77 {11, "14551915228366851806640625"}, 78 {12, "72759576141834259033203125"}, 79 {12, "363797880709171295166015625"}, 80 {12, "1818989403545856475830078125"}, 81 {13, "9094947017729282379150390625"}, 82 {13, "45474735088646411895751953125"}, 83 {13, "227373675443232059478759765625"}, 84 {13, "1136868377216160297393798828125"}, 85 {14, "5684341886080801486968994140625"}, 86 {14, "28421709430404007434844970703125"}, 87 {14, "142108547152020037174224853515625"}, 88 {15, "710542735760100185871124267578125"}, 89 {15, "3552713678800500929355621337890625"}, 90 {15, "17763568394002504646778106689453125"}, 91 {16, "88817841970012523233890533447265625"}, 92 {16, "444089209850062616169452667236328125"}, 93 {16, "2220446049250313080847263336181640625"}, 94 {16, "11102230246251565404236316680908203125"}, 95 {17, "55511151231257827021181583404541015625"}, 96 {17, "277555756156289135105907917022705078125"}, 97 {17, "1387778780781445675529539585113525390625"}, 98 {18, "6938893903907228377647697925567626953125"}, 99 {18, "34694469519536141888238489627838134765625"}, 100 {18, "173472347597680709441192448139190673828125"}, 101 {19, "867361737988403547205962240695953369140625"}, 102 }; 103 104 // The maximum amount we can shift is the number of bits used in the 105 // accumulator, minus the number of bits needed to represent the base (in this 106 // case 4). 107 static constexpr uint32_t MAX_SHIFT_AMOUNT = sizeof(uint64_t) - 4; 108 109 // 800 is an arbitrary number of digits, but should be 110 // large enough for any practical number. 111 static constexpr uint32_t MAX_NUM_DIGITS = 800; 112 113 uint32_t num_digits = 0; 114 int32_t decimal_point = 0; 115 bool truncated = false; 116 uint8_t digits[MAX_NUM_DIGITS]; 117 118 private: should_round_up(int32_t round_to_digit,RoundDirection round)119 LIBC_INLINE bool should_round_up(int32_t round_to_digit, 120 RoundDirection round) { 121 if (round_to_digit < 0 || 122 static_cast<uint32_t>(round_to_digit) >= this->num_digits) { 123 return false; 124 } 125 126 // The above condition handles all cases where all of the trailing digits 127 // are zero. In that case, if the rounding mode is up, then this number 128 // should be rounded up. Similarly, if the rounding mode is down, then it 129 // should always round down. 130 if (round == RoundDirection::Up) { 131 return true; 132 } else if (round == RoundDirection::Down) { 133 return false; 134 } 135 // Else round to nearest. 136 137 // If we're right in the middle and there are no extra digits 138 if (this->digits[round_to_digit] == 5 && 139 static_cast<uint32_t>(round_to_digit + 1) == this->num_digits) { 140 141 // Round up if we've truncated (since that means the result is slightly 142 // higher than what's represented.) 143 if (this->truncated) { 144 return true; 145 } 146 147 // If this exactly halfway, round to even. 148 if (round_to_digit == 0) 149 // When the input is ".5". 150 return false; 151 return this->digits[round_to_digit - 1] % 2 != 0; 152 } 153 // If there are digits after round_to_digit, they must be non-zero since we 154 // trim trailing zeroes after all operations that change digits. 155 return this->digits[round_to_digit] >= 5; 156 } 157 158 // Takes an amount to left shift and returns the number of new digits needed 159 // to store the result based on LEFT_SHIFT_DIGIT_TABLE. get_num_new_digits(uint32_t lshift_amount)160 LIBC_INLINE uint32_t get_num_new_digits(uint32_t lshift_amount) { 161 const char *power_of_five = 162 LEFT_SHIFT_DIGIT_TABLE[lshift_amount].power_of_five; 163 uint32_t new_digits = LEFT_SHIFT_DIGIT_TABLE[lshift_amount].new_digits; 164 uint32_t digit_index = 0; 165 while (power_of_five[digit_index] != 0) { 166 if (digit_index >= this->num_digits) { 167 return new_digits - 1; 168 } 169 if (this->digits[digit_index] != power_of_five[digit_index] - '0') { 170 return new_digits - 171 ((this->digits[digit_index] < power_of_five[digit_index] - '0') 172 ? 1 173 : 0); 174 } 175 ++digit_index; 176 } 177 return new_digits; 178 } 179 180 // Trim all trailing 0s trim_trailing_zeroes()181 LIBC_INLINE void trim_trailing_zeroes() { 182 while (this->num_digits > 0 && this->digits[this->num_digits - 1] == 0) { 183 --this->num_digits; 184 } 185 if (this->num_digits == 0) { 186 this->decimal_point = 0; 187 } 188 } 189 190 // Perform a digitwise binary non-rounding right shift on this value by 191 // shift_amount. The shift_amount can't be more than MAX_SHIFT_AMOUNT to 192 // prevent overflow. right_shift(uint32_t shift_amount)193 LIBC_INLINE void right_shift(uint32_t shift_amount) { 194 uint32_t read_index = 0; 195 uint32_t write_index = 0; 196 197 uint64_t accumulator = 0; 198 199 const uint64_t shift_mask = (uint64_t(1) << shift_amount) - 1; 200 201 // Warm Up phase: we don't have enough digits to start writing, so just 202 // read them into the accumulator. 203 while (accumulator >> shift_amount == 0) { 204 uint64_t read_digit = 0; 205 // If there are still digits to read, read the next one, else the digit is 206 // assumed to be 0. 207 if (read_index < this->num_digits) { 208 read_digit = this->digits[read_index]; 209 } 210 accumulator = accumulator * 10 + read_digit; 211 ++read_index; 212 } 213 214 // Shift the decimal point by the number of digits it took to fill the 215 // accumulator. 216 this->decimal_point -= read_index - 1; 217 218 // Middle phase: we have enough digits to write, as well as more digits to 219 // read. Keep reading until we run out of digits. 220 while (read_index < this->num_digits) { 221 uint64_t read_digit = this->digits[read_index]; 222 uint64_t write_digit = accumulator >> shift_amount; 223 accumulator &= shift_mask; 224 this->digits[write_index] = static_cast<uint8_t>(write_digit); 225 accumulator = accumulator * 10 + read_digit; 226 ++read_index; 227 ++write_index; 228 } 229 230 // Cool Down phase: All of the readable digits have been read, so just write 231 // the remainder, while treating any more digits as 0. 232 while (accumulator > 0) { 233 uint64_t write_digit = accumulator >> shift_amount; 234 accumulator &= shift_mask; 235 if (write_index < MAX_NUM_DIGITS) { 236 this->digits[write_index] = static_cast<uint8_t>(write_digit); 237 ++write_index; 238 } else if (write_digit > 0) { 239 this->truncated = true; 240 } 241 accumulator = accumulator * 10; 242 } 243 this->num_digits = write_index; 244 this->trim_trailing_zeroes(); 245 } 246 247 // Perform a digitwise binary non-rounding left shift on this value by 248 // shift_amount. The shift_amount can't be more than MAX_SHIFT_AMOUNT to 249 // prevent overflow. left_shift(uint32_t shift_amount)250 LIBC_INLINE void left_shift(uint32_t shift_amount) { 251 uint32_t new_digits = this->get_num_new_digits(shift_amount); 252 253 int32_t read_index = this->num_digits - 1; 254 uint32_t write_index = this->num_digits + new_digits; 255 256 uint64_t accumulator = 0; 257 258 // No Warm Up phase. Since we're putting digits in at the top and taking 259 // digits from the bottom we don't have to wait for the accumulator to fill. 260 261 // Middle phase: while we have more digits to read, keep reading as well as 262 // writing. 263 while (read_index >= 0) { 264 accumulator += static_cast<uint64_t>(this->digits[read_index]) 265 << shift_amount; 266 uint64_t next_accumulator = accumulator / 10; 267 uint64_t write_digit = accumulator - (10 * next_accumulator); 268 --write_index; 269 if (write_index < MAX_NUM_DIGITS) { 270 this->digits[write_index] = static_cast<uint8_t>(write_digit); 271 } else if (write_digit != 0) { 272 this->truncated = true; 273 } 274 accumulator = next_accumulator; 275 --read_index; 276 } 277 278 // Cool Down phase: there are no more digits to read, so just write the 279 // remaining digits in the accumulator. 280 while (accumulator > 0) { 281 uint64_t next_accumulator = accumulator / 10; 282 uint64_t write_digit = accumulator - (10 * next_accumulator); 283 --write_index; 284 if (write_index < MAX_NUM_DIGITS) { 285 this->digits[write_index] = static_cast<uint8_t>(write_digit); 286 } else if (write_digit != 0) { 287 this->truncated = true; 288 } 289 accumulator = next_accumulator; 290 } 291 292 this->num_digits += new_digits; 293 if (this->num_digits > MAX_NUM_DIGITS) { 294 this->num_digits = MAX_NUM_DIGITS; 295 } 296 this->decimal_point += new_digits; 297 this->trim_trailing_zeroes(); 298 } 299 300 public: 301 // num_string is assumed to be a string of numeric characters. It doesn't 302 // handle leading spaces. 303 LIBC_INLINE 304 HighPrecisionDecimal( 305 const char *__restrict num_string, 306 const size_t num_len = cpp::numeric_limits<size_t>::max()) { 307 bool saw_dot = false; 308 size_t num_cur = 0; 309 // This counts the digits in the number, even if there isn't space to store 310 // them all. 311 uint32_t total_digits = 0; 312 while (num_cur < num_len && 313 (isdigit(num_string[num_cur]) || num_string[num_cur] == '.')) { 314 if (num_string[num_cur] == '.') { 315 if (saw_dot) { 316 break; 317 } 318 this->decimal_point = total_digits; 319 saw_dot = true; 320 } else { 321 if (num_string[num_cur] == '0' && this->num_digits == 0) { 322 --this->decimal_point; 323 ++num_cur; 324 continue; 325 } 326 ++total_digits; 327 if (this->num_digits < MAX_NUM_DIGITS) { 328 this->digits[this->num_digits] = 329 static_cast<uint8_t>(num_string[num_cur] - '0'); 330 ++this->num_digits; 331 } else if (num_string[num_cur] != '0') { 332 this->truncated = true; 333 } 334 } 335 ++num_cur; 336 } 337 338 if (!saw_dot) 339 this->decimal_point = total_digits; 340 341 if (num_cur < num_len && ((num_string[num_cur] | 32) == 'e')) { 342 ++num_cur; 343 if (isdigit(num_string[num_cur]) || num_string[num_cur] == '+' || 344 num_string[num_cur] == '-') { 345 auto result = 346 strtointeger<int32_t>(num_string + num_cur, 10, num_len - num_cur); 347 if (result.has_error()) { 348 // TODO: handle error 349 } 350 int32_t add_to_exponent = result.value; 351 352 // Here we do this operation as int64 to avoid overflow. 353 int64_t temp_exponent = static_cast<int64_t>(this->decimal_point) + 354 static_cast<int64_t>(add_to_exponent); 355 356 // Theoretically these numbers should be MAX_BIASED_EXPONENT for long 357 // double, but that should be ~16,000 which is much less than 1 << 30. 358 if (temp_exponent > (1 << 30)) { 359 temp_exponent = (1 << 30); 360 } else if (temp_exponent < -(1 << 30)) { 361 temp_exponent = -(1 << 30); 362 } 363 this->decimal_point = static_cast<int32_t>(temp_exponent); 364 } 365 } 366 367 this->trim_trailing_zeroes(); 368 } 369 370 // Binary shift left (shift_amount > 0) or right (shift_amount < 0) shift(int shift_amount)371 LIBC_INLINE void shift(int shift_amount) { 372 if (shift_amount == 0) { 373 return; 374 } 375 // Left 376 else if (shift_amount > 0) { 377 while (static_cast<uint32_t>(shift_amount) > MAX_SHIFT_AMOUNT) { 378 this->left_shift(MAX_SHIFT_AMOUNT); 379 shift_amount -= MAX_SHIFT_AMOUNT; 380 } 381 this->left_shift(shift_amount); 382 } 383 // Right 384 else { 385 while (static_cast<uint32_t>(shift_amount) < -MAX_SHIFT_AMOUNT) { 386 this->right_shift(MAX_SHIFT_AMOUNT); 387 shift_amount += MAX_SHIFT_AMOUNT; 388 } 389 this->right_shift(-shift_amount); 390 } 391 } 392 393 // Round the number represented to the closest value of unsigned int type T. 394 // This is done ignoring overflow. 395 template <class T> 396 LIBC_INLINE T 397 round_to_integer_type(RoundDirection round = RoundDirection::Nearest) { 398 T result = 0; 399 uint32_t cur_digit = 0; 400 401 while (static_cast<int32_t>(cur_digit) < this->decimal_point && 402 cur_digit < this->num_digits) { 403 result = result * 10 + (this->digits[cur_digit]); 404 ++cur_digit; 405 } 406 407 // If there are implicit 0s at the end of the number, include those. 408 while (static_cast<int32_t>(cur_digit) < this->decimal_point) { 409 result *= 10; 410 ++cur_digit; 411 } 412 return result + this->should_round_up(this->decimal_point, round); 413 } 414 415 // Extra functions for testing. 416 get_digits()417 LIBC_INLINE uint8_t *get_digits() { return this->digits; } get_num_digits()418 LIBC_INLINE uint32_t get_num_digits() { return this->num_digits; } get_decimal_point()419 LIBC_INLINE int32_t get_decimal_point() { return this->decimal_point; } set_truncated(bool trunc)420 LIBC_INLINE void set_truncated(bool trunc) { this->truncated = trunc; } 421 }; 422 423 } // namespace internal 424 } // namespace LIBC_NAMESPACE 425 426 #endif // LLVM_LIBC_SRC___SUPPORT_HIGH_PRECISION_DECIMAL_H 427