1 | /* |
2 | Name: imath.h |
3 | Purpose: Arbitrary precision integer arithmetic routines. |
4 | Author: M. J. Fromberger |
5 | |
6 | Copyright (C) 2002-2007 Michael J. Fromberger, All Rights Reserved. |
7 | |
8 | Permission is hereby granted, free of charge, to any person obtaining a copy |
9 | of this software and associated documentation files (the "Software"), to deal |
10 | in the Software without restriction, including without limitation the rights |
11 | to use, copy, modify, merge, publish, distribute, sublicense, and/or sell |
12 | copies of the Software, and to permit persons to whom the Software is |
13 | furnished to do so, subject to the following conditions: |
14 | |
15 | The above copyright notice and this permission notice shall be included in |
16 | all copies or substantial portions of the Software. |
17 | |
18 | THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR |
19 | IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, |
20 | FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE |
21 | AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER |
22 | LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, |
23 | OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE |
24 | SOFTWARE. |
25 | */ |
26 | |
27 | #ifndef IMATH_H_ |
28 | #define IMATH_H_ |
29 | |
30 | #include <limits.h> |
31 | #include <stdbool.h> |
32 | #include <stdint.h> |
33 | |
34 | #ifdef __cplusplus |
35 | extern "C" { |
36 | #endif |
37 | |
38 | typedef unsigned char mp_sign; |
39 | typedef unsigned int mp_size; |
40 | typedef int mp_result; |
41 | typedef long mp_small; /* must be a signed type */ |
42 | typedef unsigned long mp_usmall; /* must be an unsigned type */ |
43 | |
44 | |
45 | /* Build with words as uint64_t by default. */ |
46 | #ifdef USE_32BIT_WORDS |
47 | typedef uint16_t mp_digit; |
48 | typedef uint32_t mp_word; |
49 | # define MP_DIGIT_MAX (UINT16_MAX * 1UL) |
50 | # define MP_WORD_MAX (UINT32_MAX * 1UL) |
51 | #else |
52 | typedef uint32_t mp_digit; |
53 | typedef uint64_t mp_word; |
54 | # define MP_DIGIT_MAX (UINT32_MAX * UINT64_C(1)) |
55 | # define MP_WORD_MAX (UINT64_MAX) |
56 | #endif |
57 | |
58 | typedef struct { |
59 | mp_digit single; |
60 | mp_digit* digits; |
61 | mp_size alloc; |
62 | mp_size used; |
63 | mp_sign sign; |
64 | } mpz_t, *mp_int; |
65 | |
66 | static inline mp_digit* MP_DIGITS(mp_int Z) { return Z->digits; } |
67 | static inline mp_size MP_ALLOC(mp_int Z) { return Z->alloc; } |
68 | static inline mp_size MP_USED(mp_int Z) { return Z->used; } |
69 | static inline mp_sign MP_SIGN(mp_int Z) { return Z->sign; } |
70 | |
71 | extern const mp_result MP_OK; |
72 | extern const mp_result MP_FALSE; |
73 | extern const mp_result MP_TRUE; |
74 | extern const mp_result MP_MEMORY; |
75 | extern const mp_result MP_RANGE; |
76 | extern const mp_result MP_UNDEF; |
77 | extern const mp_result MP_TRUNC; |
78 | extern const mp_result MP_BADARG; |
79 | extern const mp_result MP_MINERR; |
80 | |
81 | #define MP_DIGIT_BIT (sizeof(mp_digit) * CHAR_BIT) |
82 | #define MP_WORD_BIT (sizeof(mp_word) * CHAR_BIT) |
83 | #define MP_SMALL_MIN LONG_MIN |
84 | #define MP_SMALL_MAX LONG_MAX |
85 | #define MP_USMALL_MAX ULONG_MAX |
86 | |
87 | #define MP_MIN_RADIX 2 |
88 | #define MP_MAX_RADIX 36 |
89 | |
90 | /** Sets the default number of digits allocated to an `mp_int` constructed by |
91 | `mp_int_init_size()` with `prec == 0`. Allocations are rounded up to |
92 | multiples of this value. `MP_DEFAULT_PREC` is the default value. Requires |
93 | `ndigits > 0`. */ |
94 | void mp_int_default_precision(mp_size ndigits); |
95 | |
96 | /** Sets the number of digits below which multiplication will use the standard |
97 | quadratic "schoolbook" multiplication algorithm rather than Karatsuba-Ofman. |
98 | Requires `ndigits >= sizeof(mp_word)`. */ |
99 | void mp_int_multiply_threshold(mp_size ndigits); |
100 | |
101 | /** A sign indicating a (strictly) negative value. */ |
102 | extern const mp_sign MP_NEG; |
103 | |
104 | /** A sign indicating a zero or positive value. */ |
105 | extern const mp_sign MP_ZPOS; |
106 | |
107 | /** Reports whether `z` is odd, having remainder 1 when divided by 2. */ |
108 | static inline bool mp_int_is_odd(mp_int z) { return (z->digits[0] & 1) != 0; } |
109 | |
110 | /** Reports whether `z` is even, having remainder 0 when divided by 2. */ |
111 | static inline bool mp_int_is_even(mp_int z) { return (z->digits[0] & 1) == 0; } |
112 | |
113 | /** Initializes `z` with 1-digit precision and sets it to zero. This function |
114 | cannot fail unless `z == NULL`. */ |
115 | mp_result mp_int_init(mp_int z); |
116 | |
117 | /** Allocates a fresh zero-valued `mpz_t` on the heap, returning NULL in case |
118 | of error. The only possible error is out-of-memory. */ |
119 | mp_int mp_int_alloc(void); |
120 | |
121 | /** Initializes `z` with at least `prec` digits of storage, and sets it to |
122 | zero. If `prec` is zero, the default precision is used. In either case the |
123 | size is rounded up to the nearest multiple of the word size. */ |
124 | mp_result mp_int_init_size(mp_int z, mp_size prec); |
125 | |
126 | /** Initializes `z` to be a copy of an already-initialized value in `old`. The |
127 | new copy does not share storage with the original. */ |
128 | mp_result mp_int_init_copy(mp_int z, mp_int old); |
129 | |
130 | /** Initializes `z` to the specified signed `value` at default precision. */ |
131 | mp_result mp_int_init_value(mp_int z, mp_small value); |
132 | |
133 | /** Initializes `z` to the specified unsigned `value` at default precision. */ |
134 | mp_result mp_int_init_uvalue(mp_int z, mp_usmall uvalue); |
135 | |
136 | /** Sets `z` to the value of the specified signed `value`. */ |
137 | mp_result mp_int_set_value(mp_int z, mp_small value); |
138 | |
139 | /** Sets `z` to the value of the specified unsigned `value`. */ |
140 | mp_result mp_int_set_uvalue(mp_int z, mp_usmall uvalue); |
141 | |
142 | /** Releases the storage used by `z`. */ |
143 | void mp_int_clear(mp_int z); |
144 | |
145 | /** Releases the storage used by `z` and also `z` itself. |
146 | This should only be used for `z` allocated by `mp_int_alloc()`. */ |
147 | void mp_int_free(mp_int z); |
148 | |
149 | /** Replaces the value of `c` with a copy of the value of `a`. No new memory is |
150 | allocated unless `a` has more significant digits than `c` has allocated. */ |
151 | mp_result mp_int_copy(mp_int a, mp_int c); |
152 | |
153 | /** Swaps the values and storage between `a` and `c`. */ |
154 | void mp_int_swap(mp_int a, mp_int c); |
155 | |
156 | /** Sets `z` to zero. The allocated storage of `z` is not changed. */ |
157 | void mp_int_zero(mp_int z); |
158 | |
159 | /** Sets `c` to the absolute value of `a`. */ |
160 | mp_result mp_int_abs(mp_int a, mp_int c); |
161 | |
162 | /** Sets `c` to the additive inverse (negation) of `a`. */ |
163 | mp_result mp_int_neg(mp_int a, mp_int c); |
164 | |
165 | /** Sets `c` to the sum of `a` and `b`. */ |
166 | mp_result mp_int_add(mp_int a, mp_int b, mp_int c); |
167 | |
168 | /** Sets `c` to the sum of `a` and `value`. */ |
169 | mp_result mp_int_add_value(mp_int a, mp_small value, mp_int c); |
170 | |
171 | /** Sets `c` to the difference of `a` less `b`. */ |
172 | mp_result mp_int_sub(mp_int a, mp_int b, mp_int c); |
173 | |
174 | /** Sets `c` to the difference of `a` less `value`. */ |
175 | mp_result mp_int_sub_value(mp_int a, mp_small value, mp_int c); |
176 | |
177 | /** Sets `c` to the product of `a` and `b`. */ |
178 | mp_result mp_int_mul(mp_int a, mp_int b, mp_int c); |
179 | |
180 | /** Sets `c` to the product of `a` and `value`. */ |
181 | mp_result mp_int_mul_value(mp_int a, mp_small value, mp_int c); |
182 | |
183 | /** Sets `c` to the product of `a` and `2^p2`. Requires `p2 >= 0`. */ |
184 | mp_result mp_int_mul_pow2(mp_int a, mp_small p2, mp_int c); |
185 | |
186 | /** Sets `c` to the square of `a`. */ |
187 | mp_result mp_int_sqr(mp_int a, mp_int c); |
188 | |
189 | /** Sets `q` and `r` to the quotent and remainder of `a / b`. Division by |
190 | powers of 2 is detected and handled efficiently. The remainder is pinned |
191 | to `0 <= r < b`. |
192 | |
193 | Either of `q` or `r` may be NULL, but not both, and `q` and `r` may not |
194 | point to the same value. */ |
195 | mp_result mp_int_div(mp_int a, mp_int b, mp_int q, mp_int r); |
196 | |
197 | /** Sets `q` and `*r` to the quotent and remainder of `a / value`. Division by |
198 | powers of 2 is detected and handled efficiently. The remainder is pinned to |
199 | `0 <= *r < b`. Either of `q` or `r` may be NULL. */ |
200 | mp_result mp_int_div_value(mp_int a, mp_small value, mp_int q, mp_small *r); |
201 | |
202 | /** Sets `q` and `r` to the quotient and remainder of `a / 2^p2`. This is a |
203 | special case for division by powers of two that is more efficient than |
204 | using ordinary division. Note that `mp_int_div()` will automatically handle |
205 | this case, this function is for cases where you have only the exponent. */ |
206 | mp_result mp_int_div_pow2(mp_int a, mp_small p2, mp_int q, mp_int r); |
207 | |
208 | /** Sets `c` to the remainder of `a / m`. |
209 | The remainder is pinned to `0 <= c < m`. */ |
210 | mp_result mp_int_mod(mp_int a, mp_int m, mp_int c); |
211 | |
212 | /** Sets `c` to the value of `a` raised to the `b` power. |
213 | It returns `MP_RANGE` if `b < 0`. */ |
214 | mp_result mp_int_expt(mp_int a, mp_small b, mp_int c); |
215 | |
216 | /** Sets `c` to the value of `a` raised to the `b` power. |
217 | It returns `MP_RANGE` if `b < 0`. */ |
218 | mp_result mp_int_expt_value(mp_small a, mp_small b, mp_int c); |
219 | |
220 | /** Sets `c` to the value of `a` raised to the `b` power. |
221 | It returns `MP_RANGE`) if `b < 0`. */ |
222 | mp_result mp_int_expt_full(mp_int a, mp_int b, mp_int c); |
223 | |
224 | /** Sets `*r` to the remainder of `a / value`. |
225 | The remainder is pinned to `0 <= r < value`. */ |
226 | static inline |
227 | mp_result mp_int_mod_value(mp_int a, mp_small value, mp_small* r) { |
228 | return mp_int_div_value(a, value, q: 0, r); |
229 | } |
230 | |
231 | /** Returns the comparator of `a` and `b`. */ |
232 | int mp_int_compare(mp_int a, mp_int b); |
233 | |
234 | /** Returns the comparator of the magnitudes of `a` and `b`, disregarding their |
235 | signs. Neither `a` nor `b` is modified by the comparison. */ |
236 | int mp_int_compare_unsigned(mp_int a, mp_int b); |
237 | |
238 | /** Returns the comparator of `z` and zero. */ |
239 | int mp_int_compare_zero(mp_int z); |
240 | |
241 | /** Returns the comparator of `z` and the signed value `v`. */ |
242 | int mp_int_compare_value(mp_int z, mp_small v); |
243 | |
244 | /** Returns the comparator of `z` and the unsigned value `uv`. */ |
245 | int mp_int_compare_uvalue(mp_int z, mp_usmall uv); |
246 | |
247 | /** Reports whether `a` is divisible by `v`. */ |
248 | bool mp_int_divisible_value(mp_int a, mp_small v); |
249 | |
250 | /** Returns `k >= 0` such that `z` is `2^k`, if such a `k` exists. If no such |
251 | `k` exists, the function returns -1. */ |
252 | int mp_int_is_pow2(mp_int z); |
253 | |
254 | /** Sets `c` to the value of `a` raised to the `b` power, reduced modulo `m`. |
255 | It returns `MP_RANGE` if `b < 0` or `MP_UNDEF` if `m == 0`. */ |
256 | mp_result mp_int_exptmod(mp_int a, mp_int b, mp_int m, mp_int c); |
257 | |
258 | /** Sets `c` to the value of `a` raised to the `value` power, modulo `m`. |
259 | It returns `MP_RANGE` if `value < 0` or `MP_UNDEF` if `m == 0`. */ |
260 | mp_result mp_int_exptmod_evalue(mp_int a, mp_small value, mp_int m, mp_int c); |
261 | |
262 | /** Sets `c` to the value of `value` raised to the `b` power, modulo `m`. |
263 | It returns `MP_RANGE` if `b < 0` or `MP_UNDEF` if `m == 0`. */ |
264 | mp_result mp_int_exptmod_bvalue(mp_small value, mp_int b, mp_int m, mp_int c); |
265 | |
266 | /** Sets `c` to the value of `a` raised to the `b` power, reduced modulo `m`, |
267 | given a precomputed reduction constant `mu` defined for Barrett's modular |
268 | reduction algorithm. |
269 | |
270 | It returns `MP_RANGE` if `b < 0` or `MP_UNDEF` if `m == 0`. */ |
271 | mp_result mp_int_exptmod_known(mp_int a, mp_int b, mp_int m, mp_int mu, mp_int c); |
272 | |
273 | /** Sets `c` to the reduction constant for Barrett reduction by modulus `m`. |
274 | Requires that `c` and `m` point to distinct locations. */ |
275 | mp_result mp_int_redux_const(mp_int m, mp_int c); |
276 | |
277 | /** Sets `c` to the multiplicative inverse of `a` modulo `m`, if it exists. |
278 | The least non-negative representative of the congruence class is computed. |
279 | |
280 | It returns `MP_UNDEF` if the inverse does not exist, or `MP_RANGE` if `a == |
281 | 0` or `m <= 0`. */ |
282 | mp_result mp_int_invmod(mp_int a, mp_int m, mp_int c); |
283 | |
284 | /** Sets `c` to the greatest common divisor of `a` and `b`. |
285 | |
286 | It returns `MP_UNDEF` if the GCD is undefined, such as for example if `a` |
287 | and `b` are both zero. */ |
288 | mp_result mp_int_gcd(mp_int a, mp_int b, mp_int c); |
289 | |
290 | /** Sets `c` to the greatest common divisor of `a` and `b`, and sets `x` and |
291 | `y` to values satisfying Bezout's identity `gcd(a, b) = ax + by`. |
292 | |
293 | It returns `MP_UNDEF` if the GCD is undefined, such as for example if `a` |
294 | and `b` are both zero. */ |
295 | mp_result mp_int_egcd(mp_int a, mp_int b, mp_int c, mp_int x, mp_int y); |
296 | |
297 | /** Sets `c` to the least common multiple of `a` and `b`. |
298 | |
299 | It returns `MP_UNDEF` if the LCM is undefined, such as for example if `a` |
300 | and `b` are both zero. */ |
301 | mp_result mp_int_lcm(mp_int a, mp_int b, mp_int c); |
302 | |
303 | /** Sets `c` to the greatest integer not less than the `b`th root of `a`, |
304 | using Newton's root-finding algorithm. |
305 | It returns `MP_UNDEF` if `a < 0` and `b` is even. */ |
306 | mp_result mp_int_root(mp_int a, mp_small b, mp_int c); |
307 | |
308 | /** Sets `c` to the greatest integer not less than the square root of `a`. |
309 | This is a special case of `mp_int_root()`. */ |
310 | static inline |
311 | mp_result mp_int_sqrt(mp_int a, mp_int c) { return mp_int_root(a, b: 2, c); } |
312 | |
313 | /** Returns `MP_OK` if `z` is representable as `mp_small`, else `MP_RANGE`. |
314 | If `out` is not NULL, `*out` is set to the value of `z` when `MP_OK`. */ |
315 | mp_result mp_int_to_int(mp_int z, mp_small *out); |
316 | |
317 | /** Returns `MP_OK` if `z` is representable as `mp_usmall`, or `MP_RANGE`. |
318 | If `out` is not NULL, `*out` is set to the value of `z` when `MP_OK`. */ |
319 | mp_result mp_int_to_uint(mp_int z, mp_usmall *out); |
320 | |
321 | /** Converts `z` to a zero-terminated string of characters in the specified |
322 | `radix`, writing at most `limit` characters to `str` including the |
323 | terminating NUL value. A leading `-` is used to indicate a negative value. |
324 | |
325 | Returns `MP_TRUNC` if `limit` was to small to write all of `z`. |
326 | Requires `MP_MIN_RADIX <= radix <= MP_MAX_RADIX`. */ |
327 | mp_result mp_int_to_string(mp_int z, mp_size radix, char *str, int limit); |
328 | |
329 | /** Reports the minimum number of characters required to represent `z` as a |
330 | zero-terminated string in the given `radix`. |
331 | Requires `MP_MIN_RADIX <= radix <= MP_MAX_RADIX`. */ |
332 | mp_result mp_int_string_len(mp_int z, mp_size radix); |
333 | |
334 | /** Reads a string of ASCII digits in the specified `radix` from the zero |
335 | terminated `str` provided into `z`. For values of `radix > 10`, the letters |
336 | `A`..`Z` or `a`..`z` are accepted. Letters are interpreted without respect |
337 | to case. |
338 | |
339 | Leading whitespace is ignored, and a leading `+` or `-` is interpreted as a |
340 | sign flag. Processing stops when a NUL or any other character out of range |
341 | for a digit in the given radix is encountered. |
342 | |
343 | If the whole string was consumed, `MP_OK` is returned; otherwise |
344 | `MP_TRUNC`. is returned. |
345 | |
346 | Requires `MP_MIN_RADIX <= radix <= MP_MAX_RADIX`. */ |
347 | mp_result mp_int_read_string(mp_int z, mp_size radix, const char *str); |
348 | |
349 | /** Reads a string of ASCII digits in the specified `radix` from the zero |
350 | terminated `str` provided into `z`. For values of `radix > 10`, the letters |
351 | `A`..`Z` or `a`..`z` are accepted. Letters are interpreted without respect |
352 | to case. |
353 | |
354 | Leading whitespace is ignored, and a leading `+` or `-` is interpreted as a |
355 | sign flag. Processing stops when a NUL or any other character out of range |
356 | for a digit in the given radix is encountered. |
357 | |
358 | If the whole string was consumed, `MP_OK` is returned; otherwise |
359 | `MP_TRUNC`. is returned. If `end` is not NULL, `*end` is set to point to |
360 | the first unconsumed byte of the input string (the NUL byte if the whole |
361 | string was consumed). This emulates the behavior of the standard C |
362 | `strtol()` function. |
363 | |
364 | Requires `MP_MIN_RADIX <= radix <= MP_MAX_RADIX`. */ |
365 | mp_result mp_int_read_cstring(mp_int z, mp_size radix, const char *str, char **end); |
366 | |
367 | /** Returns the number of significant bits in `z`. */ |
368 | mp_result mp_int_count_bits(mp_int z); |
369 | |
370 | /** Converts `z` to 2's complement binary, writing at most `limit` bytes into |
371 | the given `buf`. Returns `MP_TRUNC` if the buffer limit was too small to |
372 | contain the whole value. If this occurs, the contents of buf will be |
373 | effectively garbage, as the function uses the buffer as scratch space. |
374 | |
375 | The binary representation of `z` is in base-256 with digits ordered from |
376 | most significant to least significant (network byte ordering). The |
377 | high-order bit of the first byte is set for negative values, clear for |
378 | non-negative values. |
379 | |
380 | As a result, non-negative values will be padded with a leading zero byte if |
381 | the high-order byte of the base-256 magnitude is set. This extra byte is |
382 | accounted for by the `mp_int_binary_len()` function. */ |
383 | mp_result mp_int_to_binary(mp_int z, unsigned char *buf, int limit); |
384 | |
385 | /** Reads a 2's complement binary value from `buf` into `z`, where `len` is the |
386 | length of the buffer. The contents of `buf` may be overwritten during |
387 | processing, although they will be restored when the function returns. */ |
388 | mp_result mp_int_read_binary(mp_int z, unsigned char *buf, int len); |
389 | |
390 | /** Returns the number of bytes to represent `z` in 2's complement binary. */ |
391 | mp_result mp_int_binary_len(mp_int z); |
392 | |
393 | /** Converts the magnitude of `z` to unsigned binary, writing at most `limit` |
394 | bytes into the given `buf`. The sign of `z` is ignored, but `z` is not |
395 | modified. Returns `MP_TRUNC` if the buffer limit was too small to contain |
396 | the whole value. If this occurs, the contents of `buf` will be effectively |
397 | garbage, as the function uses the buffer as scratch space during |
398 | conversion. |
399 | |
400 | The binary representation of `z` is in base-256 with digits ordered from |
401 | most significant to least significant (network byte ordering). */ |
402 | mp_result mp_int_to_unsigned(mp_int z, unsigned char *buf, int limit); |
403 | |
404 | /** Reads an unsigned binary value from `buf` into `z`, where `len` is the |
405 | length of the buffer. The contents of `buf` are not modified during |
406 | processing. */ |
407 | mp_result mp_int_read_unsigned(mp_int z, unsigned char *buf, int len); |
408 | |
409 | /** Returns the number of bytes required to represent `z` as an unsigned binary |
410 | value in base 256. */ |
411 | mp_result mp_int_unsigned_len(mp_int z); |
412 | |
413 | /** Returns a pointer to a brief, human-readable, zero-terminated string |
414 | describing `res`. The returned string is statically allocated and must not |
415 | be freed by the caller. */ |
416 | const char *mp_error_string(mp_result res); |
417 | |
418 | #ifdef __cplusplus |
419 | } |
420 | #endif |
421 | #endif /* end IMATH_H_ */ |
422 | |