xref: /freebsd/contrib/bearssl/src/int/i15_moddiv.c (revision 2f513db7)
1 /*
2  * Copyright (c) 2018 Thomas Pornin <pornin@bolet.org>
3  *
4  * Permission is hereby granted, free of charge, to any person obtaining
5  * a copy of this software and associated documentation files (the
6  * "Software"), to deal in the Software without restriction, including
7  * without limitation the rights to use, copy, modify, merge, publish,
8  * distribute, sublicense, and/or sell copies of the Software, and to
9  * permit persons to whom the Software is furnished to do so, subject to
10  * the following conditions:
11  *
12  * The above copyright notice and this permission notice shall be
13  * included in all copies or substantial portions of the Software.
14  *
15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
16  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
17  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
18  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
19  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
20  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
21  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
22  * SOFTWARE.
23  */
24 
25 #include "inner.h"
26 
27 /*
28  * In this file, we handle big integers with a custom format, i.e.
29  * without the usual one-word header. Value is split into 15-bit words,
30  * each stored in a 16-bit slot (top bit is zero) in little-endian
31  * order. The length (in words) is provided explicitly. In some cases,
32  * the value can be negative (using two's complement representation). In
33  * some cases, the top word is allowed to have a 16th bit.
34  */
35 
36 /*
37  * Negate big integer conditionally. The value consists of 'len' words,
38  * with 15 bits in each word (the top bit of each word should be 0,
39  * except possibly for the last word). If 'ctl' is 1, the negation is
40  * computed; otherwise, if 'ctl' is 0, then the value is unchanged.
41  */
42 static void
43 cond_negate(uint16_t *a, size_t len, uint32_t ctl)
44 {
45 	size_t k;
46 	uint32_t cc, xm;
47 
48 	cc = ctl;
49 	xm = 0x7FFF & -ctl;
50 	for (k = 0; k < len; k ++) {
51 		uint32_t aw;
52 
53 		aw = a[k];
54 		aw = (aw ^ xm) + cc;
55 		a[k] = aw & 0x7FFF;
56 		cc = (aw >> 15) & 1;
57 	}
58 }
59 
60 /*
61  * Finish modular reduction. Rules on input parameters:
62  *
63  *   if neg = 1, then -m <= a < 0
64  *   if neg = 0, then 0 <= a < 2*m
65  *
66  * If neg = 0, then the top word of a[] may use 16 bits.
67  *
68  * Also, modulus m must be odd.
69  */
70 static void
71 finish_mod(uint16_t *a, size_t len, const uint16_t *m, uint32_t neg)
72 {
73 	size_t k;
74 	uint32_t cc, xm, ym;
75 
76 	/*
77 	 * First pass: compare a (assumed nonnegative) with m.
78 	 */
79 	cc = 0;
80 	for (k = 0; k < len; k ++) {
81 		uint32_t aw, mw;
82 
83 		aw = a[k];
84 		mw = m[k];
85 		cc = (aw - mw - cc) >> 31;
86 	}
87 
88 	/*
89 	 * At this point:
90 	 *   if neg = 1, then we must add m (regardless of cc)
91 	 *   if neg = 0 and cc = 0, then we must subtract m
92 	 *   if neg = 0 and cc = 1, then we must do nothing
93 	 */
94 	xm = 0x7FFF & -neg;
95 	ym = -(neg | (1 - cc));
96 	cc = neg;
97 	for (k = 0; k < len; k ++) {
98 		uint32_t aw, mw;
99 
100 		aw = a[k];
101 		mw = (m[k] ^ xm) & ym;
102 		aw = aw - mw - cc;
103 		a[k] = aw & 0x7FFF;
104 		cc = aw >> 31;
105 	}
106 }
107 
108 /*
109  * Compute:
110  *   a <- (a*pa+b*pb)/(2^15)
111  *   b <- (a*qa+b*qb)/(2^15)
112  * The division is assumed to be exact (i.e. the low word is dropped).
113  * If the final a is negative, then it is negated. Similarly for b.
114  * Returned value is the combination of two bits:
115  *   bit 0: 1 if a had to be negated, 0 otherwise
116  *   bit 1: 1 if b had to be negated, 0 otherwise
117  *
118  * Factors pa, pb, qa and qb must be at most 2^15 in absolute value.
119  * Source integers a and b must be nonnegative; top word is not allowed
120  * to contain an extra 16th bit.
121  */
122 static uint32_t
123 co_reduce(uint16_t *a, uint16_t *b, size_t len,
124 	int32_t pa, int32_t pb, int32_t qa, int32_t qb)
125 {
126 	size_t k;
127 	int32_t cca, ccb;
128 	uint32_t nega, negb;
129 
130 	cca = 0;
131 	ccb = 0;
132 	for (k = 0; k < len; k ++) {
133 		uint32_t wa, wb, za, zb;
134 		uint16_t tta, ttb;
135 
136 		/*
137 		 * Since:
138 		 *   |pa| <= 2^15
139 		 *   |pb| <= 2^15
140 		 *   0 <= wa <= 2^15 - 1
141 		 *   0 <= wb <= 2^15 - 1
142 		 *   |cca| <= 2^16 - 1
143 		 * Then:
144 		 *   |za| <= (2^15-1)*(2^16) + (2^16-1) = 2^31 - 1
145 		 *
146 		 * Thus, the new value of cca is such that |cca| <= 2^16 - 1.
147 		 * The same applies to ccb.
148 		 */
149 		wa = a[k];
150 		wb = b[k];
151 		za = wa * (uint32_t)pa + wb * (uint32_t)pb + (uint32_t)cca;
152 		zb = wa * (uint32_t)qa + wb * (uint32_t)qb + (uint32_t)ccb;
153 		if (k > 0) {
154 			a[k - 1] = za & 0x7FFF;
155 			b[k - 1] = zb & 0x7FFF;
156 		}
157 		tta = za >> 15;
158 		ttb = zb >> 15;
159 		cca = *(int16_t *)&tta;
160 		ccb = *(int16_t *)&ttb;
161 	}
162 	a[len - 1] = (uint16_t)cca;
163 	b[len - 1] = (uint16_t)ccb;
164 	nega = (uint32_t)cca >> 31;
165 	negb = (uint32_t)ccb >> 31;
166 	cond_negate(a, len, nega);
167 	cond_negate(b, len, negb);
168 	return nega | (negb << 1);
169 }
170 
171 /*
172  * Compute:
173  *   a <- (a*pa+b*pb)/(2^15) mod m
174  *   b <- (a*qa+b*qb)/(2^15) mod m
175  *
176  * m0i is equal to -1/m[0] mod 2^15.
177  *
178  * Factors pa, pb, qa and qb must be at most 2^15 in absolute value.
179  * Source integers a and b must be nonnegative; top word is not allowed
180  * to contain an extra 16th bit.
181  */
182 static void
183 co_reduce_mod(uint16_t *a, uint16_t *b, size_t len,
184 	int32_t pa, int32_t pb, int32_t qa, int32_t qb,
185 	const uint16_t *m, uint16_t m0i)
186 {
187 	size_t k;
188 	int32_t cca, ccb, fa, fb;
189 
190 	cca = 0;
191 	ccb = 0;
192 	fa = ((a[0] * (uint32_t)pa + b[0] * (uint32_t)pb) * m0i) & 0x7FFF;
193 	fb = ((a[0] * (uint32_t)qa + b[0] * (uint32_t)qb) * m0i) & 0x7FFF;
194 	for (k = 0; k < len; k ++) {
195 		uint32_t wa, wb, za, zb;
196 		uint32_t tta, ttb;
197 
198 		/*
199 		 * In this loop, carries 'cca' and 'ccb' always fit on
200 		 * 17 bits (in absolute value).
201 		 */
202 		wa = a[k];
203 		wb = b[k];
204 		za = wa * (uint32_t)pa + wb * (uint32_t)pb
205 			+ m[k] * (uint32_t)fa + (uint32_t)cca;
206 		zb = wa * (uint32_t)qa + wb * (uint32_t)qb
207 			+ m[k] * (uint32_t)fb + (uint32_t)ccb;
208 		if (k > 0) {
209 			a[k - 1] = za & 0x7FFF;
210 			b[k - 1] = zb & 0x7FFF;
211 		}
212 
213 		/*
214 		 * The XOR-and-sub construction below does an arithmetic
215 		 * right shift in a portable way (technically, right-shifting
216 		 * a negative signed value is implementation-defined in C).
217 		 */
218 #define M   ((uint32_t)1 << 16)
219 		tta = za >> 15;
220 		ttb = zb >> 15;
221 		tta = (tta ^ M) - M;
222 		ttb = (ttb ^ M) - M;
223 		cca = *(int32_t *)&tta;
224 		ccb = *(int32_t *)&ttb;
225 #undef M
226 	}
227 	a[len - 1] = (uint32_t)cca;
228 	b[len - 1] = (uint32_t)ccb;
229 
230 	/*
231 	 * At this point:
232 	 *   -m <= a < 2*m
233 	 *   -m <= b < 2*m
234 	 * (this is a case of Montgomery reduction)
235 	 * The top word of 'a' and 'b' may have a 16-th bit set.
236 	 * We may have to add or subtract the modulus.
237 	 */
238 	finish_mod(a, len, m, (uint32_t)cca >> 31);
239 	finish_mod(b, len, m, (uint32_t)ccb >> 31);
240 }
241 
242 /* see inner.h */
243 uint32_t
244 br_i15_moddiv(uint16_t *x, const uint16_t *y, const uint16_t *m, uint16_t m0i,
245 	uint16_t *t)
246 {
247 	/*
248 	 * Algorithm is an extended binary GCD. We maintain four values
249 	 * a, b, u and v, with the following invariants:
250 	 *
251 	 *   a * x = y * u mod m
252 	 *   b * x = y * v mod m
253 	 *
254 	 * Starting values are:
255 	 *
256 	 *   a = y
257 	 *   b = m
258 	 *   u = x
259 	 *   v = 0
260 	 *
261 	 * The formal definition of the algorithm is a sequence of steps:
262 	 *
263 	 *   - If a is even, then a <- a/2 and u <- u/2 mod m.
264 	 *   - Otherwise, if b is even, then b <- b/2 and v <- v/2 mod m.
265 	 *   - Otherwise, if a > b, then a <- (a-b)/2 and u <- (u-v)/2 mod m.
266 	 *   - Otherwise, b <- (b-a)/2 and v <- (v-u)/2 mod m.
267 	 *
268 	 * Algorithm stops when a = b. At that point, they both are equal
269 	 * to GCD(y,m); the modular division succeeds if that value is 1.
270 	 * The result of the modular division is then u (or v: both are
271 	 * equal at that point).
272 	 *
273 	 * Each step makes either a or b shrink by at least one bit; hence,
274 	 * if m has bit length k bits, then 2k-2 steps are sufficient.
275 	 *
276 	 *
277 	 * Though complexity is quadratic in the size of m, the bit-by-bit
278 	 * processing is not very efficient. We can speed up processing by
279 	 * remarking that the decisions are taken based only on observation
280 	 * of the top and low bits of a and b.
281 	 *
282 	 * In the loop below, at each iteration, we use the two top words
283 	 * of a and b, and the low words of a and b, to compute reduction
284 	 * parameters pa, pb, qa and qb such that the new values for a
285 	 * and b are:
286 	 *
287 	 *   a' = (a*pa + b*pb) / (2^15)
288 	 *   b' = (a*qa + b*qb) / (2^15)
289 	 *
290 	 * the division being exact.
291 	 *
292 	 * Since the choices are based on the top words, they may be slightly
293 	 * off, requiring an optional correction: if a' < 0, then we replace
294 	 * pa with -pa, and pb with -pb. The total length of a and b is
295 	 * thus reduced by at least 14 bits at each iteration.
296 	 *
297 	 * The stopping conditions are still the same, though: when a
298 	 * and b become equal, they must be both odd (since m is odd,
299 	 * the GCD cannot be even), therefore the next operation is a
300 	 * subtraction, and one of the values becomes 0. At that point,
301 	 * nothing else happens, i.e. one value is stuck at 0, and the
302 	 * other one is the GCD.
303 	 */
304 	size_t len, k;
305 	uint16_t *a, *b, *u, *v;
306 	uint32_t num, r;
307 
308 	len = (m[0] + 15) >> 4;
309 	a = t;
310 	b = a + len;
311 	u = x + 1;
312 	v = b + len;
313 	memcpy(a, y + 1, len * sizeof *y);
314 	memcpy(b, m + 1, len * sizeof *m);
315 	memset(v, 0, len * sizeof *v);
316 
317 	/*
318 	 * Loop below ensures that a and b are reduced by some bits each,
319 	 * for a total of at least 14 bits.
320 	 */
321 	for (num = ((m[0] - (m[0] >> 4)) << 1) + 14; num >= 14; num -= 14) {
322 		size_t j;
323 		uint32_t c0, c1;
324 		uint32_t a0, a1, b0, b1;
325 		uint32_t a_hi, b_hi, a_lo, b_lo;
326 		int32_t pa, pb, qa, qb;
327 		int i;
328 
329 		/*
330 		 * Extract top words of a and b. If j is the highest
331 		 * index >= 1 such that a[j] != 0 or b[j] != 0, then we want
332 		 * (a[j] << 15) + a[j - 1], and (b[j] << 15) + b[j - 1].
333 		 * If a and b are down to one word each, then we use a[0]
334 		 * and b[0].
335 		 */
336 		c0 = (uint32_t)-1;
337 		c1 = (uint32_t)-1;
338 		a0 = 0;
339 		a1 = 0;
340 		b0 = 0;
341 		b1 = 0;
342 		j = len;
343 		while (j -- > 0) {
344 			uint32_t aw, bw;
345 
346 			aw = a[j];
347 			bw = b[j];
348 			a0 ^= (a0 ^ aw) & c0;
349 			a1 ^= (a1 ^ aw) & c1;
350 			b0 ^= (b0 ^ bw) & c0;
351 			b1 ^= (b1 ^ bw) & c1;
352 			c1 = c0;
353 			c0 &= (((aw | bw) + 0xFFFF) >> 16) - (uint32_t)1;
354 		}
355 
356 		/*
357 		 * If c1 = 0, then we grabbed two words for a and b.
358 		 * If c1 != 0 but c0 = 0, then we grabbed one word. It
359 		 * is not possible that c1 != 0 and c0 != 0, because that
360 		 * would mean that both integers are zero.
361 		 */
362 		a1 |= a0 & c1;
363 		a0 &= ~c1;
364 		b1 |= b0 & c1;
365 		b0 &= ~c1;
366 		a_hi = (a0 << 15) + a1;
367 		b_hi = (b0 << 15) + b1;
368 		a_lo = a[0];
369 		b_lo = b[0];
370 
371 		/*
372 		 * Compute reduction factors:
373 		 *
374 		 *   a' = a*pa + b*pb
375 		 *   b' = a*qa + b*qb
376 		 *
377 		 * such that a' and b' are both multiple of 2^15, but are
378 		 * only marginally larger than a and b.
379 		 */
380 		pa = 1;
381 		pb = 0;
382 		qa = 0;
383 		qb = 1;
384 		for (i = 0; i < 15; i ++) {
385 			/*
386 			 * At each iteration:
387 			 *
388 			 *   a <- (a-b)/2 if: a is odd, b is odd, a_hi > b_hi
389 			 *   b <- (b-a)/2 if: a is odd, b is odd, a_hi <= b_hi
390 			 *   a <- a/2 if: a is even
391 			 *   b <- b/2 if: a is odd, b is even
392 			 *
393 			 * We multiply a_lo and b_lo by 2 at each
394 			 * iteration, thus a division by 2 really is a
395 			 * non-multiplication by 2.
396 			 */
397 			uint32_t r, oa, ob, cAB, cBA, cA;
398 
399 			/*
400 			 * cAB = 1 if b must be subtracted from a
401 			 * cBA = 1 if a must be subtracted from b
402 			 * cA = 1 if a is divided by 2, 0 otherwise
403 			 *
404 			 * Rules:
405 			 *
406 			 *   cAB and cBA cannot be both 1.
407 			 *   if a is not divided by 2, b is.
408 			 */
409 			r = GT(a_hi, b_hi);
410 			oa = (a_lo >> i) & 1;
411 			ob = (b_lo >> i) & 1;
412 			cAB = oa & ob & r;
413 			cBA = oa & ob & NOT(r);
414 			cA = cAB | NOT(oa);
415 
416 			/*
417 			 * Conditional subtractions.
418 			 */
419 			a_lo -= b_lo & -cAB;
420 			a_hi -= b_hi & -cAB;
421 			pa -= qa & -(int32_t)cAB;
422 			pb -= qb & -(int32_t)cAB;
423 			b_lo -= a_lo & -cBA;
424 			b_hi -= a_hi & -cBA;
425 			qa -= pa & -(int32_t)cBA;
426 			qb -= pb & -(int32_t)cBA;
427 
428 			/*
429 			 * Shifting.
430 			 */
431 			a_lo += a_lo & (cA - 1);
432 			pa += pa & ((int32_t)cA - 1);
433 			pb += pb & ((int32_t)cA - 1);
434 			a_hi ^= (a_hi ^ (a_hi >> 1)) & -cA;
435 			b_lo += b_lo & -cA;
436 			qa += qa & -(int32_t)cA;
437 			qb += qb & -(int32_t)cA;
438 			b_hi ^= (b_hi ^ (b_hi >> 1)) & (cA - 1);
439 		}
440 
441 		/*
442 		 * Replace a and b with new values a' and b'.
443 		 */
444 		r = co_reduce(a, b, len, pa, pb, qa, qb);
445 		pa -= pa * ((r & 1) << 1);
446 		pb -= pb * ((r & 1) << 1);
447 		qa -= qa * (r & 2);
448 		qb -= qb * (r & 2);
449 		co_reduce_mod(u, v, len, pa, pb, qa, qb, m + 1, m0i);
450 	}
451 
452 	/*
453 	 * Now one of the arrays should be 0, and the other contains
454 	 * the GCD. If a is 0, then u is 0 as well, and v contains
455 	 * the division result.
456 	 * Result is correct if and only if GCD is 1.
457 	 */
458 	r = (a[0] | b[0]) ^ 1;
459 	u[0] |= v[0];
460 	for (k = 1; k < len; k ++) {
461 		r |= a[k] | b[k];
462 		u[k] |= v[k];
463 	}
464 	return EQ0(r);
465 }
466