xref: /freebsd/contrib/bearssl/src/ec/ec_c25519_m64.c (revision e0c4386e7e71d93b0edc0c8fa156263fc4a8b0b6)
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 #if BR_INT128 || BR_UMUL128
28 
29 #if BR_UMUL128
30 #include <intrin.h>
31 #endif
32 
33 static const unsigned char GEN[] = {
34 	0x09, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
35 	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
36 	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
37 	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00
38 };
39 
40 static const unsigned char ORDER[] = {
41 	0x7F, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
42 	0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
43 	0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
44 	0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF
45 };
46 
47 static const unsigned char *
48 api_generator(int curve, size_t *len)
49 {
50 	(void)curve;
51 	*len = 32;
52 	return GEN;
53 }
54 
55 static const unsigned char *
56 api_order(int curve, size_t *len)
57 {
58 	(void)curve;
59 	*len = 32;
60 	return ORDER;
61 }
62 
63 static size_t
64 api_xoff(int curve, size_t *len)
65 {
66 	(void)curve;
67 	*len = 32;
68 	return 0;
69 }
70 
71 /*
72  * A field element is encoded as four 64-bit integers, in basis 2^63.
73  * Operations return partially reduced values, which may range up to
74  * 2^255+37.
75  */
76 
77 #define MASK63   (((uint64_t)1 << 63) - (uint64_t)1)
78 
79 /*
80  * Swap two field elements, conditionally on a flag.
81  */
82 static inline void
83 f255_cswap(uint64_t *a, uint64_t *b, uint32_t ctl)
84 {
85 	uint64_t m, w;
86 
87 	m = -(uint64_t)ctl;
88 	w = m & (a[0] ^ b[0]); a[0] ^= w; b[0] ^= w;
89 	w = m & (a[1] ^ b[1]); a[1] ^= w; b[1] ^= w;
90 	w = m & (a[2] ^ b[2]); a[2] ^= w; b[2] ^= w;
91 	w = m & (a[3] ^ b[3]); a[3] ^= w; b[3] ^= w;
92 }
93 
94 /*
95  * Addition in the field.
96  */
97 static inline void
98 f255_add(uint64_t *d, const uint64_t *a, const uint64_t *b)
99 {
100 #if BR_INT128
101 
102 	uint64_t t0, t1, t2, t3, cc;
103 	unsigned __int128 z;
104 
105 	z = (unsigned __int128)a[0] + (unsigned __int128)b[0];
106 	t0 = (uint64_t)z;
107 	z = (unsigned __int128)a[1] + (unsigned __int128)b[1] + (z >> 64);
108 	t1 = (uint64_t)z;
109 	z = (unsigned __int128)a[2] + (unsigned __int128)b[2] + (z >> 64);
110 	t2 = (uint64_t)z;
111 	z = (unsigned __int128)a[3] + (unsigned __int128)b[3] + (z >> 64);
112 	t3 = (uint64_t)z & MASK63;
113 	cc = (uint64_t)(z >> 63);
114 
115 	/*
116 	 * Since operands are at most 2^255+37, the sum is at most
117 	 * 2^256+74; thus, the carry cc is equal to 0, 1 or 2.
118 	 *
119 	 * We use: 2^255 = 19 mod p.
120 	 * Since we add 0, 19 or 38 to a value that fits on 255 bits,
121 	 * the result is at most 2^255+37.
122 	 */
123 	z = (unsigned __int128)t0 + (unsigned __int128)(19 * cc);
124 	d[0] = (uint64_t)z;
125 	z = (unsigned __int128)t1 + (z >> 64);
126 	d[1] = (uint64_t)z;
127 	z = (unsigned __int128)t2 + (z >> 64);
128 	d[2] = (uint64_t)z;
129 	d[3] = t3 + (uint64_t)(z >> 64);
130 
131 #elif BR_UMUL128
132 
133 	uint64_t t0, t1, t2, t3, cc;
134 	unsigned char k;
135 
136 	k = _addcarry_u64(0, a[0], b[0], &t0);
137 	k = _addcarry_u64(k, a[1], b[1], &t1);
138 	k = _addcarry_u64(k, a[2], b[2], &t2);
139 	k = _addcarry_u64(k, a[3], b[3], &t3);
140 	cc = (k << 1) + (t3 >> 63);
141 	t3 &= MASK63;
142 
143 	/*
144 	 * Since operands are at most 2^255+37, the sum is at most
145 	 * 2^256+74; thus, the carry cc is equal to 0, 1 or 2.
146 	 *
147 	 * We use: 2^255 = 19 mod p.
148 	 * Since we add 0, 19 or 38 to a value that fits on 255 bits,
149 	 * the result is at most 2^255+37.
150 	 */
151 	k = _addcarry_u64(0, t0, 19 * cc, &d[0]);
152 	k = _addcarry_u64(k, t1, 0, &d[1]);
153 	k = _addcarry_u64(k, t2, 0, &d[2]);
154 	(void)_addcarry_u64(k, t3, 0, &d[3]);
155 
156 #endif
157 }
158 
159 /*
160  * Subtraction.
161  */
162 static inline void
163 f255_sub(uint64_t *d, const uint64_t *a, const uint64_t *b)
164 {
165 #if BR_INT128
166 
167 	/*
168 	 * We compute t = 2^256 - 38 + a - b, which is necessarily
169 	 * positive but lower than 2^256 + 2^255, since a <= 2^255 + 37
170 	 * and b <= 2^255 + 37. We then subtract 0, p or 2*p, depending
171 	 * on the two upper bits of t (bits 255 and 256).
172 	 */
173 
174 	uint64_t t0, t1, t2, t3, t4, cc;
175 	unsigned __int128 z;
176 
177 	z = (unsigned __int128)a[0] - (unsigned __int128)b[0] - 38;
178 	t0 = (uint64_t)z;
179 	cc = -(uint64_t)(z >> 64);
180 	z = (unsigned __int128)a[1] - (unsigned __int128)b[1]
181 		- (unsigned __int128)cc;
182 	t1 = (uint64_t)z;
183 	cc = -(uint64_t)(z >> 64);
184 	z = (unsigned __int128)a[2] - (unsigned __int128)b[2]
185 		- (unsigned __int128)cc;
186 	t2 = (uint64_t)z;
187 	cc = -(uint64_t)(z >> 64);
188 	z = (unsigned __int128)a[3] - (unsigned __int128)b[3]
189 		- (unsigned __int128)cc;
190 	t3 = (uint64_t)z;
191 	t4 = 1 + (uint64_t)(z >> 64);
192 
193 	/*
194 	 * We have a 257-bit result. The two top bits can be 00, 01 or 10,
195 	 * but not 11 (value t <= 2^256 - 38 + 2^255 + 37 = 2^256 + 2^255 - 1).
196 	 * Therefore, we can truncate to 255 bits, and add 0, 19 or 38.
197 	 * This guarantees that the result is at most 2^255+37.
198 	 */
199 	cc = (38 & -t4) + (19 & -(t3 >> 63));
200 	t3 &= MASK63;
201 	z = (unsigned __int128)t0 + (unsigned __int128)cc;
202 	d[0] = (uint64_t)z;
203 	z = (unsigned __int128)t1 + (z >> 64);
204 	d[1] = (uint64_t)z;
205 	z = (unsigned __int128)t2 + (z >> 64);
206 	d[2] = (uint64_t)z;
207 	d[3] = t3 + (uint64_t)(z >> 64);
208 
209 #elif BR_UMUL128
210 
211 	/*
212 	 * We compute t = 2^256 - 38 + a - b, which is necessarily
213 	 * positive but lower than 2^256 + 2^255, since a <= 2^255 + 37
214 	 * and b <= 2^255 + 37. We then subtract 0, p or 2*p, depending
215 	 * on the two upper bits of t (bits 255 and 256).
216 	 */
217 
218 	uint64_t t0, t1, t2, t3, t4;
219 	unsigned char k;
220 
221 	k = _subborrow_u64(0, a[0], b[0], &t0);
222 	k = _subborrow_u64(k, a[1], b[1], &t1);
223 	k = _subborrow_u64(k, a[2], b[2], &t2);
224 	k = _subborrow_u64(k, a[3], b[3], &t3);
225 	(void)_subborrow_u64(k, 1, 0, &t4);
226 
227 	k = _subborrow_u64(0, t0, 38, &t0);
228 	k = _subborrow_u64(k, t1, 0, &t1);
229 	k = _subborrow_u64(k, t2, 0, &t2);
230 	k = _subborrow_u64(k, t3, 0, &t3);
231 	(void)_subborrow_u64(k, t4, 0, &t4);
232 
233 	/*
234 	 * We have a 257-bit result. The two top bits can be 00, 01 or 10,
235 	 * but not 11 (value t <= 2^256 - 38 + 2^255 + 37 = 2^256 + 2^255 - 1).
236 	 * Therefore, we can truncate to 255 bits, and add 0, 19 or 38.
237 	 * This guarantees that the result is at most 2^255+37.
238 	 */
239 	t4 = (38 & -t4) + (19 & -(t3 >> 63));
240 	t3 &= MASK63;
241 	k = _addcarry_u64(0, t0, t4, &d[0]);
242 	k = _addcarry_u64(k, t1, 0, &d[1]);
243 	k = _addcarry_u64(k, t2, 0, &d[2]);
244 	(void)_addcarry_u64(k, t3, 0, &d[3]);
245 
246 #endif
247 }
248 
249 /*
250  * Multiplication.
251  */
252 static inline void
253 f255_mul(uint64_t *d, uint64_t *a, uint64_t *b)
254 {
255 #if BR_INT128
256 
257 	unsigned __int128 z;
258 	uint64_t t0, t1, t2, t3, t4, t5, t6, t7, th;
259 
260 	/*
261 	 * Compute the product a*b over plain integers.
262 	 */
263 	z = (unsigned __int128)a[0] * (unsigned __int128)b[0];
264 	t0 = (uint64_t)z;
265 	z = (unsigned __int128)a[0] * (unsigned __int128)b[1] + (z >> 64);
266 	t1 = (uint64_t)z;
267 	z = (unsigned __int128)a[0] * (unsigned __int128)b[2] + (z >> 64);
268 	t2 = (uint64_t)z;
269 	z = (unsigned __int128)a[0] * (unsigned __int128)b[3] + (z >> 64);
270 	t3 = (uint64_t)z;
271 	t4 = (uint64_t)(z >> 64);
272 
273 	z = (unsigned __int128)a[1] * (unsigned __int128)b[0]
274 		+ (unsigned __int128)t1;
275 	t1 = (uint64_t)z;
276 	z = (unsigned __int128)a[1] * (unsigned __int128)b[1]
277 		+ (unsigned __int128)t2 + (z >> 64);
278 	t2 = (uint64_t)z;
279 	z = (unsigned __int128)a[1] * (unsigned __int128)b[2]
280 		+ (unsigned __int128)t3 + (z >> 64);
281 	t3 = (uint64_t)z;
282 	z = (unsigned __int128)a[1] * (unsigned __int128)b[3]
283 		+ (unsigned __int128)t4 + (z >> 64);
284 	t4 = (uint64_t)z;
285 	t5 = (uint64_t)(z >> 64);
286 
287 	z = (unsigned __int128)a[2] * (unsigned __int128)b[0]
288 		+ (unsigned __int128)t2;
289 	t2 = (uint64_t)z;
290 	z = (unsigned __int128)a[2] * (unsigned __int128)b[1]
291 		+ (unsigned __int128)t3 + (z >> 64);
292 	t3 = (uint64_t)z;
293 	z = (unsigned __int128)a[2] * (unsigned __int128)b[2]
294 		+ (unsigned __int128)t4 + (z >> 64);
295 	t4 = (uint64_t)z;
296 	z = (unsigned __int128)a[2] * (unsigned __int128)b[3]
297 		+ (unsigned __int128)t5 + (z >> 64);
298 	t5 = (uint64_t)z;
299 	t6 = (uint64_t)(z >> 64);
300 
301 	z = (unsigned __int128)a[3] * (unsigned __int128)b[0]
302 		+ (unsigned __int128)t3;
303 	t3 = (uint64_t)z;
304 	z = (unsigned __int128)a[3] * (unsigned __int128)b[1]
305 		+ (unsigned __int128)t4 + (z >> 64);
306 	t4 = (uint64_t)z;
307 	z = (unsigned __int128)a[3] * (unsigned __int128)b[2]
308 		+ (unsigned __int128)t5 + (z >> 64);
309 	t5 = (uint64_t)z;
310 	z = (unsigned __int128)a[3] * (unsigned __int128)b[3]
311 		+ (unsigned __int128)t6 + (z >> 64);
312 	t6 = (uint64_t)z;
313 	t7 = (uint64_t)(z >> 64);
314 
315 	/*
316 	 * Modulo p, we have:
317 	 *
318 	 *   2^255 = 19
319 	 *   2^510 = 19*19 = 361
320 	 *
321 	 * We split the intermediate t into three parts, in basis
322 	 * 2^255. The low one will be in t0..t3; the middle one in t4..t7.
323 	 * The upper one can only be a single bit (th), since the
324 	 * multiplication operands are at most 2^255+37 each.
325 	 */
326 	th = t7 >> 62;
327 	t7 = ((t7 << 1) | (t6 >> 63)) & MASK63;
328 	t6 = (t6 << 1) | (t5 >> 63);
329 	t5 = (t5 << 1) | (t4 >> 63);
330 	t4 = (t4 << 1) | (t3 >> 63);
331 	t3 &= MASK63;
332 
333 	/*
334 	 * Multiply the middle part (t4..t7) by 19. We truncate it to
335 	 * 255 bits; the extra bits will go along with th.
336 	 */
337 	z = (unsigned __int128)t4 * 19;
338 	t4 = (uint64_t)z;
339 	z = (unsigned __int128)t5 * 19 + (z >> 64);
340 	t5 = (uint64_t)z;
341 	z = (unsigned __int128)t6 * 19 + (z >> 64);
342 	t6 = (uint64_t)z;
343 	z = (unsigned __int128)t7 * 19 + (z >> 64);
344 	t7 = (uint64_t)z & MASK63;
345 
346 	th = (361 & -th) + (19 * (uint64_t)(z >> 63));
347 
348 	/*
349 	 * Add elements together.
350 	 * At this point:
351 	 *   t0..t3 fits on 255 bits.
352 	 *   t4..t7 fits on 255 bits.
353 	 *   th <= 361 + 342 = 703.
354 	 */
355 	z = (unsigned __int128)t0 + (unsigned __int128)t4
356 		+ (unsigned __int128)th;
357 	t0 = (uint64_t)z;
358 	z = (unsigned __int128)t1 + (unsigned __int128)t5 + (z >> 64);
359 	t1 = (uint64_t)z;
360 	z = (unsigned __int128)t2 + (unsigned __int128)t6 + (z >> 64);
361 	t2 = (uint64_t)z;
362 	z = (unsigned __int128)t3 + (unsigned __int128)t7 + (z >> 64);
363 	t3 = (uint64_t)z & MASK63;
364 	th = (uint64_t)(z >> 63);
365 
366 	/*
367 	 * Since the sum is at most 2^256 + 703, the two upper bits, in th,
368 	 * can only have value 0, 1 or 2. We just add th*19, which
369 	 * guarantees a result of at most 2^255+37.
370 	 */
371 	z = (unsigned __int128)t0 + (19 * th);
372 	d[0] = (uint64_t)z;
373 	z = (unsigned __int128)t1 + (z >> 64);
374 	d[1] = (uint64_t)z;
375 	z = (unsigned __int128)t2 + (z >> 64);
376 	d[2] = (uint64_t)z;
377 	d[3] = t3 + (uint64_t)(z >> 64);
378 
379 #elif BR_UMUL128
380 
381 	uint64_t t0, t1, t2, t3, t4, t5, t6, t7, th;
382 	uint64_t h0, h1, h2, h3;
383 	unsigned char k;
384 
385 	/*
386 	 * Compute the product a*b over plain integers.
387 	 */
388 	t0 = _umul128(a[0], b[0], &h0);
389 	t1 = _umul128(a[0], b[1], &h1);
390 	k = _addcarry_u64(0, t1, h0, &t1);
391 	t2 = _umul128(a[0], b[2], &h2);
392 	k = _addcarry_u64(k, t2, h1, &t2);
393 	t3 = _umul128(a[0], b[3], &h3);
394 	k = _addcarry_u64(k, t3, h2, &t3);
395 	(void)_addcarry_u64(k, h3, 0, &t4);
396 
397 	k = _addcarry_u64(0, _umul128(a[1], b[0], &h0), t1, &t1);
398 	k = _addcarry_u64(k, _umul128(a[1], b[1], &h1), t2, &t2);
399 	k = _addcarry_u64(k, _umul128(a[1], b[2], &h2), t3, &t3);
400 	k = _addcarry_u64(k, _umul128(a[1], b[3], &h3), t4, &t4);
401 	t5 = k;
402 	k = _addcarry_u64(0, t2, h0, &t2);
403 	k = _addcarry_u64(k, t3, h1, &t3);
404 	k = _addcarry_u64(k, t4, h2, &t4);
405 	(void)_addcarry_u64(k, t5, h3, &t5);
406 
407 	k = _addcarry_u64(0, _umul128(a[2], b[0], &h0), t2, &t2);
408 	k = _addcarry_u64(k, _umul128(a[2], b[1], &h1), t3, &t3);
409 	k = _addcarry_u64(k, _umul128(a[2], b[2], &h2), t4, &t4);
410 	k = _addcarry_u64(k, _umul128(a[2], b[3], &h3), t5, &t5);
411 	t6 = k;
412 	k = _addcarry_u64(0, t3, h0, &t3);
413 	k = _addcarry_u64(k, t4, h1, &t4);
414 	k = _addcarry_u64(k, t5, h2, &t5);
415 	(void)_addcarry_u64(k, t6, h3, &t6);
416 
417 	k = _addcarry_u64(0, _umul128(a[3], b[0], &h0), t3, &t3);
418 	k = _addcarry_u64(k, _umul128(a[3], b[1], &h1), t4, &t4);
419 	k = _addcarry_u64(k, _umul128(a[3], b[2], &h2), t5, &t5);
420 	k = _addcarry_u64(k, _umul128(a[3], b[3], &h3), t6, &t6);
421 	t7 = k;
422 	k = _addcarry_u64(0, t4, h0, &t4);
423 	k = _addcarry_u64(k, t5, h1, &t5);
424 	k = _addcarry_u64(k, t6, h2, &t6);
425 	(void)_addcarry_u64(k, t7, h3, &t7);
426 
427 	/*
428 	 * Modulo p, we have:
429 	 *
430 	 *   2^255 = 19
431 	 *   2^510 = 19*19 = 361
432 	 *
433 	 * We split the intermediate t into three parts, in basis
434 	 * 2^255. The low one will be in t0..t3; the middle one in t4..t7.
435 	 * The upper one can only be a single bit (th), since the
436 	 * multiplication operands are at most 2^255+37 each.
437 	 */
438 	th = t7 >> 62;
439 	t7 = ((t7 << 1) | (t6 >> 63)) & MASK63;
440 	t6 = (t6 << 1) | (t5 >> 63);
441 	t5 = (t5 << 1) | (t4 >> 63);
442 	t4 = (t4 << 1) | (t3 >> 63);
443 	t3 &= MASK63;
444 
445 	/*
446 	 * Multiply the middle part (t4..t7) by 19. We truncate it to
447 	 * 255 bits; the extra bits will go along with th.
448 	 */
449 	t4 = _umul128(t4, 19, &h0);
450 	t5 = _umul128(t5, 19, &h1);
451 	t6 = _umul128(t6, 19, &h2);
452 	t7 = _umul128(t7, 19, &h3);
453 	k = _addcarry_u64(0, t5, h0, &t5);
454 	k = _addcarry_u64(k, t6, h1, &t6);
455 	k = _addcarry_u64(k, t7, h2, &t7);
456 	(void)_addcarry_u64(k, h3, 0, &h3);
457 	th = (361 & -th) + (19 * ((h3 << 1) + (t7 >> 63)));
458 	t7 &= MASK63;
459 
460 	/*
461 	 * Add elements together.
462 	 * At this point:
463 	 *   t0..t3 fits on 255 bits.
464 	 *   t4..t7 fits on 255 bits.
465 	 *   th <= 361 + 342 = 703.
466 	 */
467 	k = _addcarry_u64(0, t0, t4, &t0);
468 	k = _addcarry_u64(k, t1, t5, &t1);
469 	k = _addcarry_u64(k, t2, t6, &t2);
470 	k = _addcarry_u64(k, t3, t7, &t3);
471 	t4 = k;
472 	k = _addcarry_u64(0, t0, th, &t0);
473 	k = _addcarry_u64(k, t1, 0, &t1);
474 	k = _addcarry_u64(k, t2, 0, &t2);
475 	k = _addcarry_u64(k, t3, 0, &t3);
476 	(void)_addcarry_u64(k, t4, 0, &t4);
477 
478 	th = (t4 << 1) + (t3 >> 63);
479 	t3 &= MASK63;
480 
481 	/*
482 	 * Since the sum is at most 2^256 + 703, the two upper bits, in th,
483 	 * can only have value 0, 1 or 2. We just add th*19, which
484 	 * guarantees a result of at most 2^255+37.
485 	 */
486 	k = _addcarry_u64(0, t0, 19 * th, &d[0]);
487 	k = _addcarry_u64(k, t1, 0, &d[1]);
488 	k = _addcarry_u64(k, t2, 0, &d[2]);
489 	(void)_addcarry_u64(k, t3, 0, &d[3]);
490 
491 #endif
492 }
493 
494 /*
495  * Multiplication by A24 = 121665.
496  */
497 static inline void
498 f255_mul_a24(uint64_t *d, const uint64_t *a)
499 {
500 #if BR_INT128
501 
502 	uint64_t t0, t1, t2, t3;
503 	unsigned __int128 z;
504 
505 	z = (unsigned __int128)a[0] * 121665;
506 	t0 = (uint64_t)z;
507 	z = (unsigned __int128)a[1] * 121665 + (z >> 64);
508 	t1 = (uint64_t)z;
509 	z = (unsigned __int128)a[2] * 121665 + (z >> 64);
510 	t2 = (uint64_t)z;
511 	z = (unsigned __int128)a[3] * 121665 + (z >> 64);
512 	t3 = (uint64_t)z & MASK63;
513 
514 	z = (unsigned __int128)t0 + (19 * (uint64_t)(z >> 63));
515 	t0 = (uint64_t)z;
516 	z = (unsigned __int128)t1 + (z >> 64);
517 	t1 = (uint64_t)z;
518 	z = (unsigned __int128)t2 + (z >> 64);
519 	t2 = (uint64_t)z;
520 	t3 = t3 + (uint64_t)(z >> 64);
521 
522 	z = (unsigned __int128)t0 + (19 & -(t3 >> 63));
523 	d[0] = (uint64_t)z;
524 	z = (unsigned __int128)t1 + (z >> 64);
525 	d[1] = (uint64_t)z;
526 	z = (unsigned __int128)t2 + (z >> 64);
527 	d[2] = (uint64_t)z;
528 	d[3] = (t3 & MASK63) + (uint64_t)(z >> 64);
529 
530 #elif BR_UMUL128
531 
532 	uint64_t t0, t1, t2, t3, t4, h0, h1, h2, h3;
533 	unsigned char k;
534 
535 	t0 = _umul128(a[0], 121665, &h0);
536 	t1 = _umul128(a[1], 121665, &h1);
537 	k = _addcarry_u64(0, t1, h0, &t1);
538 	t2 = _umul128(a[2], 121665, &h2);
539 	k = _addcarry_u64(k, t2, h1, &t2);
540 	t3 = _umul128(a[3], 121665, &h3);
541 	k = _addcarry_u64(k, t3, h2, &t3);
542 	(void)_addcarry_u64(k, h3, 0, &t4);
543 
544 	t4 = (t4 << 1) + (t3 >> 63);
545 	t3 &= MASK63;
546 	k = _addcarry_u64(0, t0, 19 * t4, &t0);
547 	k = _addcarry_u64(k, t1, 0, &t1);
548 	k = _addcarry_u64(k, t2, 0, &t2);
549 	(void)_addcarry_u64(k, t3, 0, &t3);
550 
551 	t4 = 19 & -(t3 >> 63);
552 	t3 &= MASK63;
553 	k = _addcarry_u64(0, t0, t4, &d[0]);
554 	k = _addcarry_u64(k, t1, 0, &d[1]);
555 	k = _addcarry_u64(k, t2, 0, &d[2]);
556 	(void)_addcarry_u64(k, t3, 0, &d[3]);
557 
558 #endif
559 }
560 
561 /*
562  * Finalize reduction.
563  */
564 static inline void
565 f255_final_reduce(uint64_t *a)
566 {
567 #if BR_INT128
568 
569 	uint64_t t0, t1, t2, t3, m;
570 	unsigned __int128 z;
571 
572 	/*
573 	 * We add 19. If the result (in t) is below 2^255, then a[]
574 	 * is already less than 2^255-19, thus already reduced.
575 	 * Otherwise, we subtract 2^255 from t[], in which case we
576 	 * have t = a - (2^255-19), and that's our result.
577 	 */
578 	z = (unsigned __int128)a[0] + 19;
579 	t0 = (uint64_t)z;
580 	z = (unsigned __int128)a[1] + (z >> 64);
581 	t1 = (uint64_t)z;
582 	z = (unsigned __int128)a[2] + (z >> 64);
583 	t2 = (uint64_t)z;
584 	t3 = a[3] + (uint64_t)(z >> 64);
585 
586 	m = -(t3 >> 63);
587 	t3 &= MASK63;
588 	a[0] ^= m & (a[0] ^ t0);
589 	a[1] ^= m & (a[1] ^ t1);
590 	a[2] ^= m & (a[2] ^ t2);
591 	a[3] ^= m & (a[3] ^ t3);
592 
593 #elif BR_UMUL128
594 
595 	uint64_t t0, t1, t2, t3, m;
596 	unsigned char k;
597 
598 	/*
599 	 * We add 19. If the result (in t) is below 2^255, then a[]
600 	 * is already less than 2^255-19, thus already reduced.
601 	 * Otherwise, we subtract 2^255 from t[], in which case we
602 	 * have t = a - (2^255-19), and that's our result.
603 	 */
604 	k = _addcarry_u64(0, a[0], 19, &t0);
605 	k = _addcarry_u64(k, a[1], 0, &t1);
606 	k = _addcarry_u64(k, a[2], 0, &t2);
607 	(void)_addcarry_u64(k, a[3], 0, &t3);
608 
609 	m = -(t3 >> 63);
610 	t3 &= MASK63;
611 	a[0] ^= m & (a[0] ^ t0);
612 	a[1] ^= m & (a[1] ^ t1);
613 	a[2] ^= m & (a[2] ^ t2);
614 	a[3] ^= m & (a[3] ^ t3);
615 
616 #endif
617 }
618 
619 static uint32_t
620 api_mul(unsigned char *G, size_t Glen,
621 	const unsigned char *kb, size_t kblen, int curve)
622 {
623 	unsigned char k[32];
624 	uint64_t x1[4], x2[4], z2[4], x3[4], z3[4];
625 	uint32_t swap;
626 	int i;
627 
628 	(void)curve;
629 
630 	/*
631 	 * Points are encoded over exactly 32 bytes. Multipliers must fit
632 	 * in 32 bytes as well.
633 	 */
634 	if (Glen != 32 || kblen > 32) {
635 		return 0;
636 	}
637 
638 	/*
639 	 * RFC 7748 mandates that the high bit of the last point byte must
640 	 * be ignored/cleared.
641 	 */
642 	x1[0] = br_dec64le(&G[ 0]);
643 	x1[1] = br_dec64le(&G[ 8]);
644 	x1[2] = br_dec64le(&G[16]);
645 	x1[3] = br_dec64le(&G[24]) & MASK63;
646 
647 	/*
648 	 * We can use memset() to clear values, because exact-width types
649 	 * like uint64_t are guaranteed to have no padding bits or
650 	 * trap representations.
651 	 */
652 	memset(x2, 0, sizeof x2);
653 	x2[0] = 1;
654 	memset(z2, 0, sizeof z2);
655 	memcpy(x3, x1, sizeof x1);
656 	memcpy(z3, x2, sizeof x2);
657 
658 	/*
659 	 * The multiplier is provided in big-endian notation, and
660 	 * possibly shorter than 32 bytes.
661 	 */
662 	memset(k, 0, (sizeof k) - kblen);
663 	memcpy(k + (sizeof k) - kblen, kb, kblen);
664 	k[31] &= 0xF8;
665 	k[0] &= 0x7F;
666 	k[0] |= 0x40;
667 
668 	swap = 0;
669 
670 	for (i = 254; i >= 0; i --) {
671 		uint64_t a[4], aa[4], b[4], bb[4], e[4];
672 		uint64_t c[4], d[4], da[4], cb[4];
673 		uint32_t kt;
674 
675 		kt = (k[31 - (i >> 3)] >> (i & 7)) & 1;
676 		swap ^= kt;
677 		f255_cswap(x2, x3, swap);
678 		f255_cswap(z2, z3, swap);
679 		swap = kt;
680 
681 		/* A = x_2 + z_2 */
682 		f255_add(a, x2, z2);
683 
684 		/* AA = A^2 */
685 		f255_mul(aa, a, a);
686 
687 		/* B = x_2 - z_2 */
688 		f255_sub(b, x2, z2);
689 
690 		/* BB = B^2 */
691 		f255_mul(bb, b, b);
692 
693 		/* E = AA - BB */
694 		f255_sub(e, aa, bb);
695 
696 		/* C = x_3 + z_3 */
697 		f255_add(c, x3, z3);
698 
699 		/* D = x_3 - z_3 */
700 		f255_sub(d, x3, z3);
701 
702 		/* DA = D * A */
703 		f255_mul(da, d, a);
704 
705 		/* CB = C * B */
706 		f255_mul(cb, c, b);
707 
708 		/* x_3 = (DA + CB)^2 */
709 		f255_add(x3, da, cb);
710 		f255_mul(x3, x3, x3);
711 
712 		/* z_3 = x_1 * (DA - CB)^2 */
713 		f255_sub(z3, da, cb);
714 		f255_mul(z3, z3, z3);
715 		f255_mul(z3, x1, z3);
716 
717 		/* x_2 = AA * BB */
718 		f255_mul(x2, aa, bb);
719 
720 		/* z_2 = E * (AA + a24 * E) */
721 		f255_mul_a24(z2, e);
722 		f255_add(z2, aa, z2);
723 		f255_mul(z2, e, z2);
724 	}
725 
726 	f255_cswap(x2, x3, swap);
727 	f255_cswap(z2, z3, swap);
728 
729 	/*
730 	 * Compute 1/z2 = z2^(p-2). Since p = 2^255-19, we can mutualize
731 	 * most non-squarings. We use x1 and x3, now useless, as temporaries.
732 	 */
733 	memcpy(x1, z2, sizeof z2);
734 	for (i = 0; i < 15; i ++) {
735 		f255_mul(x1, x1, x1);
736 		f255_mul(x1, x1, z2);
737 	}
738 	memcpy(x3, x1, sizeof x1);
739 	for (i = 0; i < 14; i ++) {
740 		int j;
741 
742 		for (j = 0; j < 16; j ++) {
743 			f255_mul(x3, x3, x3);
744 		}
745 		f255_mul(x3, x3, x1);
746 	}
747 	for (i = 14; i >= 0; i --) {
748 		f255_mul(x3, x3, x3);
749 		if ((0xFFEB >> i) & 1) {
750 			f255_mul(x3, z2, x3);
751 		}
752 	}
753 
754 	/*
755 	 * Compute x2/z2. We have 1/z2 in x3.
756 	 */
757 	f255_mul(x2, x2, x3);
758 	f255_final_reduce(x2);
759 
760 	/*
761 	 * Encode the final x2 value in little-endian.
762 	 */
763 	br_enc64le(G,      x2[0]);
764 	br_enc64le(G +  8, x2[1]);
765 	br_enc64le(G + 16, x2[2]);
766 	br_enc64le(G + 24, x2[3]);
767 	return 1;
768 }
769 
770 static size_t
771 api_mulgen(unsigned char *R,
772 	const unsigned char *x, size_t xlen, int curve)
773 {
774 	const unsigned char *G;
775 	size_t Glen;
776 
777 	G = api_generator(curve, &Glen);
778 	memcpy(R, G, Glen);
779 	api_mul(R, Glen, x, xlen, curve);
780 	return Glen;
781 }
782 
783 static uint32_t
784 api_muladd(unsigned char *A, const unsigned char *B, size_t len,
785 	const unsigned char *x, size_t xlen,
786 	const unsigned char *y, size_t ylen, int curve)
787 {
788 	/*
789 	 * We don't implement this method, since it is used for ECDSA
790 	 * only, and there is no ECDSA over Curve25519 (which instead
791 	 * uses EdDSA).
792 	 */
793 	(void)A;
794 	(void)B;
795 	(void)len;
796 	(void)x;
797 	(void)xlen;
798 	(void)y;
799 	(void)ylen;
800 	(void)curve;
801 	return 0;
802 }
803 
804 /* see bearssl_ec.h */
805 const br_ec_impl br_ec_c25519_m64 = {
806 	(uint32_t)0x20000000,
807 	&api_generator,
808 	&api_order,
809 	&api_xoff,
810 	&api_mul,
811 	&api_mulgen,
812 	&api_muladd
813 };
814 
815 /* see bearssl_ec.h */
816 const br_ec_impl *
817 br_ec_c25519_m64_get(void)
818 {
819 	return &br_ec_c25519_m64;
820 }
821 
822 #else
823 
824 /* see bearssl_ec.h */
825 const br_ec_impl *
826 br_ec_c25519_m64_get(void)
827 {
828 	return 0;
829 }
830 
831 #endif
832