Retry fdbff0e3b60139805609f643eb40:
[chromium-blink-merge.git] / crypto / curve25519-donna.c
blobf141ac028b0f7bba419c218aad66275b507eb8c3
1 // Copyright (c) 2013 The Chromium Authors. All rights reserved.
2 // Use of this source code is governed by a BSD-style license that can be
3 // found in the LICENSE file.
5 /*
6 * curve25519-donna: Curve25519 elliptic curve, public key function
8 * http://code.google.com/p/curve25519-donna/
10 * Adam Langley <agl@imperialviolet.org>
12 * Derived from public domain C code by Daniel J. Bernstein <djb@cr.yp.to>
14 * More information about curve25519 can be found here
15 * http://cr.yp.to/ecdh.html
17 * djb's sample implementation of curve25519 is written in a special assembly
18 * language called qhasm and uses the floating point registers.
20 * This is, almost, a clean room reimplementation from the curve25519 paper. It
21 * uses many of the tricks described therein. Only the crecip function is taken
22 * from the sample implementation.
25 #include <string.h>
26 #include <stdint.h>
28 typedef uint8_t u8;
29 typedef int32_t s32;
30 typedef int64_t limb;
32 /* Field element representation:
34 * Field elements are written as an array of signed, 64-bit limbs, least
35 * significant first. The value of the field element is:
36 * x[0] + 2^26·x[1] + x^51·x[2] + 2^102·x[3] + ...
38 * i.e. the limbs are 26, 25, 26, 25, ... bits wide.
41 /* Sum two numbers: output += in */
42 static void fsum(limb *output, const limb *in) {
43 unsigned i;
44 for (i = 0; i < 10; i += 2) {
45 output[0+i] = (output[0+i] + in[0+i]);
46 output[1+i] = (output[1+i] + in[1+i]);
50 /* Find the difference of two numbers: output = in - output
51 * (note the order of the arguments!)
53 static void fdifference(limb *output, const limb *in) {
54 unsigned i;
55 for (i = 0; i < 10; ++i) {
56 output[i] = (in[i] - output[i]);
60 /* Multiply a number my a scalar: output = in * scalar */
61 static void fscalar_product(limb *output, const limb *in, const limb scalar) {
62 unsigned i;
63 for (i = 0; i < 10; ++i) {
64 output[i] = in[i] * scalar;
68 /* Multiply two numbers: output = in2 * in
70 * output must be distinct to both inputs. The inputs are reduced coefficient
71 * form, the output is not.
73 static void fproduct(limb *output, const limb *in2, const limb *in) {
74 output[0] = ((limb) ((s32) in2[0])) * ((s32) in[0]);
75 output[1] = ((limb) ((s32) in2[0])) * ((s32) in[1]) +
76 ((limb) ((s32) in2[1])) * ((s32) in[0]);
77 output[2] = 2 * ((limb) ((s32) in2[1])) * ((s32) in[1]) +
78 ((limb) ((s32) in2[0])) * ((s32) in[2]) +
79 ((limb) ((s32) in2[2])) * ((s32) in[0]);
80 output[3] = ((limb) ((s32) in2[1])) * ((s32) in[2]) +
81 ((limb) ((s32) in2[2])) * ((s32) in[1]) +
82 ((limb) ((s32) in2[0])) * ((s32) in[3]) +
83 ((limb) ((s32) in2[3])) * ((s32) in[0]);
84 output[4] = ((limb) ((s32) in2[2])) * ((s32) in[2]) +
85 2 * (((limb) ((s32) in2[1])) * ((s32) in[3]) +
86 ((limb) ((s32) in2[3])) * ((s32) in[1])) +
87 ((limb) ((s32) in2[0])) * ((s32) in[4]) +
88 ((limb) ((s32) in2[4])) * ((s32) in[0]);
89 output[5] = ((limb) ((s32) in2[2])) * ((s32) in[3]) +
90 ((limb) ((s32) in2[3])) * ((s32) in[2]) +
91 ((limb) ((s32) in2[1])) * ((s32) in[4]) +
92 ((limb) ((s32) in2[4])) * ((s32) in[1]) +
93 ((limb) ((s32) in2[0])) * ((s32) in[5]) +
94 ((limb) ((s32) in2[5])) * ((s32) in[0]);
95 output[6] = 2 * (((limb) ((s32) in2[3])) * ((s32) in[3]) +
96 ((limb) ((s32) in2[1])) * ((s32) in[5]) +
97 ((limb) ((s32) in2[5])) * ((s32) in[1])) +
98 ((limb) ((s32) in2[2])) * ((s32) in[4]) +
99 ((limb) ((s32) in2[4])) * ((s32) in[2]) +
100 ((limb) ((s32) in2[0])) * ((s32) in[6]) +
101 ((limb) ((s32) in2[6])) * ((s32) in[0]);
102 output[7] = ((limb) ((s32) in2[3])) * ((s32) in[4]) +
103 ((limb) ((s32) in2[4])) * ((s32) in[3]) +
104 ((limb) ((s32) in2[2])) * ((s32) in[5]) +
105 ((limb) ((s32) in2[5])) * ((s32) in[2]) +
106 ((limb) ((s32) in2[1])) * ((s32) in[6]) +
107 ((limb) ((s32) in2[6])) * ((s32) in[1]) +
108 ((limb) ((s32) in2[0])) * ((s32) in[7]) +
109 ((limb) ((s32) in2[7])) * ((s32) in[0]);
110 output[8] = ((limb) ((s32) in2[4])) * ((s32) in[4]) +
111 2 * (((limb) ((s32) in2[3])) * ((s32) in[5]) +
112 ((limb) ((s32) in2[5])) * ((s32) in[3]) +
113 ((limb) ((s32) in2[1])) * ((s32) in[7]) +
114 ((limb) ((s32) in2[7])) * ((s32) in[1])) +
115 ((limb) ((s32) in2[2])) * ((s32) in[6]) +
116 ((limb) ((s32) in2[6])) * ((s32) in[2]) +
117 ((limb) ((s32) in2[0])) * ((s32) in[8]) +
118 ((limb) ((s32) in2[8])) * ((s32) in[0]);
119 output[9] = ((limb) ((s32) in2[4])) * ((s32) in[5]) +
120 ((limb) ((s32) in2[5])) * ((s32) in[4]) +
121 ((limb) ((s32) in2[3])) * ((s32) in[6]) +
122 ((limb) ((s32) in2[6])) * ((s32) in[3]) +
123 ((limb) ((s32) in2[2])) * ((s32) in[7]) +
124 ((limb) ((s32) in2[7])) * ((s32) in[2]) +
125 ((limb) ((s32) in2[1])) * ((s32) in[8]) +
126 ((limb) ((s32) in2[8])) * ((s32) in[1]) +
127 ((limb) ((s32) in2[0])) * ((s32) in[9]) +
128 ((limb) ((s32) in2[9])) * ((s32) in[0]);
129 output[10] = 2 * (((limb) ((s32) in2[5])) * ((s32) in[5]) +
130 ((limb) ((s32) in2[3])) * ((s32) in[7]) +
131 ((limb) ((s32) in2[7])) * ((s32) in[3]) +
132 ((limb) ((s32) in2[1])) * ((s32) in[9]) +
133 ((limb) ((s32) in2[9])) * ((s32) in[1])) +
134 ((limb) ((s32) in2[4])) * ((s32) in[6]) +
135 ((limb) ((s32) in2[6])) * ((s32) in[4]) +
136 ((limb) ((s32) in2[2])) * ((s32) in[8]) +
137 ((limb) ((s32) in2[8])) * ((s32) in[2]);
138 output[11] = ((limb) ((s32) in2[5])) * ((s32) in[6]) +
139 ((limb) ((s32) in2[6])) * ((s32) in[5]) +
140 ((limb) ((s32) in2[4])) * ((s32) in[7]) +
141 ((limb) ((s32) in2[7])) * ((s32) in[4]) +
142 ((limb) ((s32) in2[3])) * ((s32) in[8]) +
143 ((limb) ((s32) in2[8])) * ((s32) in[3]) +
144 ((limb) ((s32) in2[2])) * ((s32) in[9]) +
145 ((limb) ((s32) in2[9])) * ((s32) in[2]);
146 output[12] = ((limb) ((s32) in2[6])) * ((s32) in[6]) +
147 2 * (((limb) ((s32) in2[5])) * ((s32) in[7]) +
148 ((limb) ((s32) in2[7])) * ((s32) in[5]) +
149 ((limb) ((s32) in2[3])) * ((s32) in[9]) +
150 ((limb) ((s32) in2[9])) * ((s32) in[3])) +
151 ((limb) ((s32) in2[4])) * ((s32) in[8]) +
152 ((limb) ((s32) in2[8])) * ((s32) in[4]);
153 output[13] = ((limb) ((s32) in2[6])) * ((s32) in[7]) +
154 ((limb) ((s32) in2[7])) * ((s32) in[6]) +
155 ((limb) ((s32) in2[5])) * ((s32) in[8]) +
156 ((limb) ((s32) in2[8])) * ((s32) in[5]) +
157 ((limb) ((s32) in2[4])) * ((s32) in[9]) +
158 ((limb) ((s32) in2[9])) * ((s32) in[4]);
159 output[14] = 2 * (((limb) ((s32) in2[7])) * ((s32) in[7]) +
160 ((limb) ((s32) in2[5])) * ((s32) in[9]) +
161 ((limb) ((s32) in2[9])) * ((s32) in[5])) +
162 ((limb) ((s32) in2[6])) * ((s32) in[8]) +
163 ((limb) ((s32) in2[8])) * ((s32) in[6]);
164 output[15] = ((limb) ((s32) in2[7])) * ((s32) in[8]) +
165 ((limb) ((s32) in2[8])) * ((s32) in[7]) +
166 ((limb) ((s32) in2[6])) * ((s32) in[9]) +
167 ((limb) ((s32) in2[9])) * ((s32) in[6]);
168 output[16] = ((limb) ((s32) in2[8])) * ((s32) in[8]) +
169 2 * (((limb) ((s32) in2[7])) * ((s32) in[9]) +
170 ((limb) ((s32) in2[9])) * ((s32) in[7]));
171 output[17] = ((limb) ((s32) in2[8])) * ((s32) in[9]) +
172 ((limb) ((s32) in2[9])) * ((s32) in[8]);
173 output[18] = 2 * ((limb) ((s32) in2[9])) * ((s32) in[9]);
176 /* Reduce a long form to a short form by taking the input mod 2^255 - 19. */
177 static void freduce_degree(limb *output) {
178 /* Each of these shifts and adds ends up multiplying the value by 19. */
179 output[8] += output[18] << 4;
180 output[8] += output[18] << 1;
181 output[8] += output[18];
182 output[7] += output[17] << 4;
183 output[7] += output[17] << 1;
184 output[7] += output[17];
185 output[6] += output[16] << 4;
186 output[6] += output[16] << 1;
187 output[6] += output[16];
188 output[5] += output[15] << 4;
189 output[5] += output[15] << 1;
190 output[5] += output[15];
191 output[4] += output[14] << 4;
192 output[4] += output[14] << 1;
193 output[4] += output[14];
194 output[3] += output[13] << 4;
195 output[3] += output[13] << 1;
196 output[3] += output[13];
197 output[2] += output[12] << 4;
198 output[2] += output[12] << 1;
199 output[2] += output[12];
200 output[1] += output[11] << 4;
201 output[1] += output[11] << 1;
202 output[1] += output[11];
203 output[0] += output[10] << 4;
204 output[0] += output[10] << 1;
205 output[0] += output[10];
208 /* Reduce all coefficients of the short form input so that |x| < 2^26.
210 * On entry: |output[i]| < 2^62
212 static void freduce_coefficients(limb *output) {
213 unsigned i;
214 do {
215 output[10] = 0;
217 for (i = 0; i < 10; i += 2) {
218 limb over = output[i] / 0x4000000l;
219 output[i+1] += over;
220 output[i] -= over * 0x4000000l;
222 over = output[i+1] / 0x2000000;
223 output[i+2] += over;
224 output[i+1] -= over * 0x2000000;
226 output[0] += 19 * output[10];
227 } while (output[10]);
230 /* A helpful wrapper around fproduct: output = in * in2.
232 * output must be distinct to both inputs. The output is reduced degree and
233 * reduced coefficient.
235 static void
236 fmul(limb *output, const limb *in, const limb *in2) {
237 limb t[19];
238 fproduct(t, in, in2);
239 freduce_degree(t);
240 freduce_coefficients(t);
241 memcpy(output, t, sizeof(limb) * 10);
244 static void fsquare_inner(limb *output, const limb *in) {
245 output[0] = ((limb) ((s32) in[0])) * ((s32) in[0]);
246 output[1] = 2 * ((limb) ((s32) in[0])) * ((s32) in[1]);
247 output[2] = 2 * (((limb) ((s32) in[1])) * ((s32) in[1]) +
248 ((limb) ((s32) in[0])) * ((s32) in[2]));
249 output[3] = 2 * (((limb) ((s32) in[1])) * ((s32) in[2]) +
250 ((limb) ((s32) in[0])) * ((s32) in[3]));
251 output[4] = ((limb) ((s32) in[2])) * ((s32) in[2]) +
252 4 * ((limb) ((s32) in[1])) * ((s32) in[3]) +
253 2 * ((limb) ((s32) in[0])) * ((s32) in[4]);
254 output[5] = 2 * (((limb) ((s32) in[2])) * ((s32) in[3]) +
255 ((limb) ((s32) in[1])) * ((s32) in[4]) +
256 ((limb) ((s32) in[0])) * ((s32) in[5]));
257 output[6] = 2 * (((limb) ((s32) in[3])) * ((s32) in[3]) +
258 ((limb) ((s32) in[2])) * ((s32) in[4]) +
259 ((limb) ((s32) in[0])) * ((s32) in[6]) +
260 2 * ((limb) ((s32) in[1])) * ((s32) in[5]));
261 output[7] = 2 * (((limb) ((s32) in[3])) * ((s32) in[4]) +
262 ((limb) ((s32) in[2])) * ((s32) in[5]) +
263 ((limb) ((s32) in[1])) * ((s32) in[6]) +
264 ((limb) ((s32) in[0])) * ((s32) in[7]));
265 output[8] = ((limb) ((s32) in[4])) * ((s32) in[4]) +
266 2 * (((limb) ((s32) in[2])) * ((s32) in[6]) +
267 ((limb) ((s32) in[0])) * ((s32) in[8]) +
268 2 * (((limb) ((s32) in[1])) * ((s32) in[7]) +
269 ((limb) ((s32) in[3])) * ((s32) in[5])));
270 output[9] = 2 * (((limb) ((s32) in[4])) * ((s32) in[5]) +
271 ((limb) ((s32) in[3])) * ((s32) in[6]) +
272 ((limb) ((s32) in[2])) * ((s32) in[7]) +
273 ((limb) ((s32) in[1])) * ((s32) in[8]) +
274 ((limb) ((s32) in[0])) * ((s32) in[9]));
275 output[10] = 2 * (((limb) ((s32) in[5])) * ((s32) in[5]) +
276 ((limb) ((s32) in[4])) * ((s32) in[6]) +
277 ((limb) ((s32) in[2])) * ((s32) in[8]) +
278 2 * (((limb) ((s32) in[3])) * ((s32) in[7]) +
279 ((limb) ((s32) in[1])) * ((s32) in[9])));
280 output[11] = 2 * (((limb) ((s32) in[5])) * ((s32) in[6]) +
281 ((limb) ((s32) in[4])) * ((s32) in[7]) +
282 ((limb) ((s32) in[3])) * ((s32) in[8]) +
283 ((limb) ((s32) in[2])) * ((s32) in[9]));
284 output[12] = ((limb) ((s32) in[6])) * ((s32) in[6]) +
285 2 * (((limb) ((s32) in[4])) * ((s32) in[8]) +
286 2 * (((limb) ((s32) in[5])) * ((s32) in[7]) +
287 ((limb) ((s32) in[3])) * ((s32) in[9])));
288 output[13] = 2 * (((limb) ((s32) in[6])) * ((s32) in[7]) +
289 ((limb) ((s32) in[5])) * ((s32) in[8]) +
290 ((limb) ((s32) in[4])) * ((s32) in[9]));
291 output[14] = 2 * (((limb) ((s32) in[7])) * ((s32) in[7]) +
292 ((limb) ((s32) in[6])) * ((s32) in[8]) +
293 2 * ((limb) ((s32) in[5])) * ((s32) in[9]));
294 output[15] = 2 * (((limb) ((s32) in[7])) * ((s32) in[8]) +
295 ((limb) ((s32) in[6])) * ((s32) in[9]));
296 output[16] = ((limb) ((s32) in[8])) * ((s32) in[8]) +
297 4 * ((limb) ((s32) in[7])) * ((s32) in[9]);
298 output[17] = 2 * ((limb) ((s32) in[8])) * ((s32) in[9]);
299 output[18] = 2 * ((limb) ((s32) in[9])) * ((s32) in[9]);
302 static void
303 fsquare(limb *output, const limb *in) {
304 limb t[19];
305 fsquare_inner(t, in);
306 freduce_degree(t);
307 freduce_coefficients(t);
308 memcpy(output, t, sizeof(limb) * 10);
311 /* Take a little-endian, 32-byte number and expand it into polynomial form */
312 static void
313 fexpand(limb *output, const u8 *input) {
314 #define F(n,start,shift,mask) \
315 output[n] = ((((limb) input[start + 0]) | \
316 ((limb) input[start + 1]) << 8 | \
317 ((limb) input[start + 2]) << 16 | \
318 ((limb) input[start + 3]) << 24) >> shift) & mask;
319 F(0, 0, 0, 0x3ffffff);
320 F(1, 3, 2, 0x1ffffff);
321 F(2, 6, 3, 0x3ffffff);
322 F(3, 9, 5, 0x1ffffff);
323 F(4, 12, 6, 0x3ffffff);
324 F(5, 16, 0, 0x1ffffff);
325 F(6, 19, 1, 0x3ffffff);
326 F(7, 22, 3, 0x1ffffff);
327 F(8, 25, 4, 0x3ffffff);
328 F(9, 28, 6, 0x1ffffff);
329 #undef F
332 /* Take a fully reduced polynomial form number and contract it into a
333 * little-endian, 32-byte array
335 static void
336 fcontract(u8 *output, limb *input) {
337 int i;
339 do {
340 for (i = 0; i < 9; ++i) {
341 if ((i & 1) == 1) {
342 while (input[i] < 0) {
343 input[i] += 0x2000000;
344 input[i + 1]--;
346 } else {
347 while (input[i] < 0) {
348 input[i] += 0x4000000;
349 input[i + 1]--;
353 while (input[9] < 0) {
354 input[9] += 0x2000000;
355 input[0] -= 19;
357 } while (input[0] < 0);
359 input[1] <<= 2;
360 input[2] <<= 3;
361 input[3] <<= 5;
362 input[4] <<= 6;
363 input[6] <<= 1;
364 input[7] <<= 3;
365 input[8] <<= 4;
366 input[9] <<= 6;
367 #define F(i, s) \
368 output[s+0] |= input[i] & 0xff; \
369 output[s+1] = (input[i] >> 8) & 0xff; \
370 output[s+2] = (input[i] >> 16) & 0xff; \
371 output[s+3] = (input[i] >> 24) & 0xff;
372 output[0] = 0;
373 output[16] = 0;
374 F(0,0);
375 F(1,3);
376 F(2,6);
377 F(3,9);
378 F(4,12);
379 F(5,16);
380 F(6,19);
381 F(7,22);
382 F(8,25);
383 F(9,28);
384 #undef F
387 /* Input: Q, Q', Q-Q'
388 * Output: 2Q, Q+Q'
390 * x2 z3: long form
391 * x3 z3: long form
392 * x z: short form, destroyed
393 * xprime zprime: short form, destroyed
394 * qmqp: short form, preserved
396 static void fmonty(limb *x2, limb *z2, /* output 2Q */
397 limb *x3, limb *z3, /* output Q + Q' */
398 limb *x, limb *z, /* input Q */
399 limb *xprime, limb *zprime, /* input Q' */
400 const limb *qmqp /* input Q - Q' */) {
401 limb origx[10], origxprime[10], zzz[19], xx[19], zz[19], xxprime[19],
402 zzprime[19], zzzprime[19], xxxprime[19];
404 memcpy(origx, x, 10 * sizeof(limb));
405 fsum(x, z);
406 fdifference(z, origx); // does x - z
408 memcpy(origxprime, xprime, sizeof(limb) * 10);
409 fsum(xprime, zprime);
410 fdifference(zprime, origxprime);
411 fproduct(xxprime, xprime, z);
412 fproduct(zzprime, x, zprime);
413 freduce_degree(xxprime);
414 freduce_coefficients(xxprime);
415 freduce_degree(zzprime);
416 freduce_coefficients(zzprime);
417 memcpy(origxprime, xxprime, sizeof(limb) * 10);
418 fsum(xxprime, zzprime);
419 fdifference(zzprime, origxprime);
420 fsquare(xxxprime, xxprime);
421 fsquare(zzzprime, zzprime);
422 fproduct(zzprime, zzzprime, qmqp);
423 freduce_degree(zzprime);
424 freduce_coefficients(zzprime);
425 memcpy(x3, xxxprime, sizeof(limb) * 10);
426 memcpy(z3, zzprime, sizeof(limb) * 10);
428 fsquare(xx, x);
429 fsquare(zz, z);
430 fproduct(x2, xx, zz);
431 freduce_degree(x2);
432 freduce_coefficients(x2);
433 fdifference(zz, xx); // does zz = xx - zz
434 memset(zzz + 10, 0, sizeof(limb) * 9);
435 fscalar_product(zzz, zz, 121665);
436 freduce_degree(zzz);
437 freduce_coefficients(zzz);
438 fsum(zzz, xx);
439 fproduct(z2, zz, zzz);
440 freduce_degree(z2);
441 freduce_coefficients(z2);
444 /* Calculates nQ where Q is the x-coordinate of a point on the curve
446 * resultx/resultz: the x coordinate of the resulting curve point (short form)
447 * n: a little endian, 32-byte number
448 * q: a point of the curve (short form)
450 static void
451 cmult(limb *resultx, limb *resultz, const u8 *n, const limb *q) {
452 limb a[19] = {0}, b[19] = {1}, c[19] = {1}, d[19] = {0};
453 limb *nqpqx = a, *nqpqz = b, *nqx = c, *nqz = d, *t;
454 limb e[19] = {0}, f[19] = {1}, g[19] = {0}, h[19] = {1};
455 limb *nqpqx2 = e, *nqpqz2 = f, *nqx2 = g, *nqz2 = h;
457 unsigned i, j;
459 memcpy(nqpqx, q, sizeof(limb) * 10);
461 for (i = 0; i < 32; ++i) {
462 u8 byte = n[31 - i];
463 for (j = 0; j < 8; ++j) {
464 if (byte & 0x80) {
465 fmonty(nqpqx2, nqpqz2,
466 nqx2, nqz2,
467 nqpqx, nqpqz,
468 nqx, nqz,
470 } else {
471 fmonty(nqx2, nqz2,
472 nqpqx2, nqpqz2,
473 nqx, nqz,
474 nqpqx, nqpqz,
478 t = nqx;
479 nqx = nqx2;
480 nqx2 = t;
481 t = nqz;
482 nqz = nqz2;
483 nqz2 = t;
484 t = nqpqx;
485 nqpqx = nqpqx2;
486 nqpqx2 = t;
487 t = nqpqz;
488 nqpqz = nqpqz2;
489 nqpqz2 = t;
491 byte <<= 1;
495 memcpy(resultx, nqx, sizeof(limb) * 10);
496 memcpy(resultz, nqz, sizeof(limb) * 10);
499 // -----------------------------------------------------------------------------
500 // Shamelessly copied from djb's code
501 // -----------------------------------------------------------------------------
502 static void
503 crecip(limb *out, const limb *z) {
504 limb z2[10];
505 limb z9[10];
506 limb z11[10];
507 limb z2_5_0[10];
508 limb z2_10_0[10];
509 limb z2_20_0[10];
510 limb z2_50_0[10];
511 limb z2_100_0[10];
512 limb t0[10];
513 limb t1[10];
514 int i;
516 /* 2 */ fsquare(z2,z);
517 /* 4 */ fsquare(t1,z2);
518 /* 8 */ fsquare(t0,t1);
519 /* 9 */ fmul(z9,t0,z);
520 /* 11 */ fmul(z11,z9,z2);
521 /* 22 */ fsquare(t0,z11);
522 /* 2^5 - 2^0 = 31 */ fmul(z2_5_0,t0,z9);
524 /* 2^6 - 2^1 */ fsquare(t0,z2_5_0);
525 /* 2^7 - 2^2 */ fsquare(t1,t0);
526 /* 2^8 - 2^3 */ fsquare(t0,t1);
527 /* 2^9 - 2^4 */ fsquare(t1,t0);
528 /* 2^10 - 2^5 */ fsquare(t0,t1);
529 /* 2^10 - 2^0 */ fmul(z2_10_0,t0,z2_5_0);
531 /* 2^11 - 2^1 */ fsquare(t0,z2_10_0);
532 /* 2^12 - 2^2 */ fsquare(t1,t0);
533 /* 2^20 - 2^10 */
534 for (i = 2;i < 10;i += 2) { fsquare(t0,t1); fsquare(t1,t0); }
535 /* 2^20 - 2^0 */ fmul(z2_20_0,t1,z2_10_0);
537 /* 2^21 - 2^1 */ fsquare(t0,z2_20_0);
538 /* 2^22 - 2^2 */ fsquare(t1,t0);
539 /* 2^40 - 2^20 */
540 for (i = 2;i < 20;i += 2) { fsquare(t0,t1); fsquare(t1,t0); }
541 /* 2^40 - 2^0 */ fmul(t0,t1,z2_20_0);
543 /* 2^41 - 2^1 */ fsquare(t1,t0);
544 /* 2^42 - 2^2 */ fsquare(t0,t1);
545 /* 2^50 - 2^10 */
546 for (i = 2;i < 10;i += 2) { fsquare(t1,t0); fsquare(t0,t1); }
547 /* 2^50 - 2^0 */ fmul(z2_50_0,t0,z2_10_0);
549 /* 2^51 - 2^1 */ fsquare(t0,z2_50_0);
550 /* 2^52 - 2^2 */ fsquare(t1,t0);
551 /* 2^100 - 2^50 */
552 for (i = 2;i < 50;i += 2) { fsquare(t0,t1); fsquare(t1,t0); }
553 /* 2^100 - 2^0 */ fmul(z2_100_0,t1,z2_50_0);
555 /* 2^101 - 2^1 */ fsquare(t1,z2_100_0);
556 /* 2^102 - 2^2 */ fsquare(t0,t1);
557 /* 2^200 - 2^100 */
558 for (i = 2;i < 100;i += 2) { fsquare(t1,t0); fsquare(t0,t1); }
559 /* 2^200 - 2^0 */ fmul(t1,t0,z2_100_0);
561 /* 2^201 - 2^1 */ fsquare(t0,t1);
562 /* 2^202 - 2^2 */ fsquare(t1,t0);
563 /* 2^250 - 2^50 */
564 for (i = 2;i < 50;i += 2) { fsquare(t0,t1); fsquare(t1,t0); }
565 /* 2^250 - 2^0 */ fmul(t0,t1,z2_50_0);
567 /* 2^251 - 2^1 */ fsquare(t1,t0);
568 /* 2^252 - 2^2 */ fsquare(t0,t1);
569 /* 2^253 - 2^3 */ fsquare(t1,t0);
570 /* 2^254 - 2^4 */ fsquare(t0,t1);
571 /* 2^255 - 2^5 */ fsquare(t1,t0);
572 /* 2^255 - 21 */ fmul(out,t1,z11);
576 curve25519_donna(u8 *mypublic, const u8 *secret, const u8 *basepoint) {
577 limb bp[10], x[10], z[10], zmone[10];
578 uint8_t e[32];
579 int i;
581 for (i = 0; i < 32; ++i) e[i] = secret[i];
582 e[0] &= 248;
583 e[31] &= 127;
584 e[31] |= 64;
586 fexpand(bp, basepoint);
587 cmult(x, z, e, bp);
588 crecip(zmone, z);
589 fmul(z, x, zmone);
590 fcontract(mypublic, z);
591 return 0;