xref: /freebsd/crypto/openssl/crypto/ec/ecp_nistp521.c (revision b6a05070fa77edc7ce6e60b61623fd806e807be6)
1 /* crypto/ec/ecp_nistp521.c */
2 /*
3  * Written by Adam Langley (Google) for the OpenSSL project
4  */
5 /* Copyright 2011 Google Inc.
6  *
7  * Licensed under the Apache License, Version 2.0 (the "License");
8  *
9  * you may not use this file except in compliance with the License.
10  * You may obtain a copy of the License at
11  *
12  *     http://www.apache.org/licenses/LICENSE-2.0
13  *
14  *  Unless required by applicable law or agreed to in writing, software
15  *  distributed under the License is distributed on an "AS IS" BASIS,
16  *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
17  *  See the License for the specific language governing permissions and
18  *  limitations under the License.
19  */
20 
21 /*
22  * A 64-bit implementation of the NIST P-521 elliptic curve point multiplication
23  *
24  * OpenSSL integration was taken from Emilia Kasper's work in ecp_nistp224.c.
25  * Otherwise based on Emilia's P224 work, which was inspired by my curve25519
26  * work which got its smarts from Daniel J. Bernstein's work on the same.
27  */
28 
29 #include <openssl/opensslconf.h>
30 #ifndef OPENSSL_NO_EC_NISTP_64_GCC_128
31 
32 # ifndef OPENSSL_SYS_VMS
33 #  include <stdint.h>
34 # else
35 #  include <inttypes.h>
36 # endif
37 
38 # include <string.h>
39 # include <openssl/err.h>
40 # include "ec_lcl.h"
41 
42 # if defined(__GNUC__) && (__GNUC__ > 3 || (__GNUC__ == 3 && __GNUC_MINOR__ >= 1))
43   /* even with gcc, the typedef won't work for 32-bit platforms */
44 typedef __uint128_t uint128_t;  /* nonstandard; implemented by gcc on 64-bit
45                                  * platforms */
46 # else
47 #  error "Need GCC 3.1 or later to define type uint128_t"
48 # endif
49 
50 typedef uint8_t u8;
51 typedef uint64_t u64;
52 typedef int64_t s64;
53 
54 /*
55  * The underlying field. P521 operates over GF(2^521-1). We can serialise an
56  * element of this field into 66 bytes where the most significant byte
57  * contains only a single bit. We call this an felem_bytearray.
58  */
59 
60 typedef u8 felem_bytearray[66];
61 
62 /*
63  * These are the parameters of P521, taken from FIPS 186-3, section D.1.2.5.
64  * These values are big-endian.
65  */
66 static const felem_bytearray nistp521_curve_params[5] = {
67     {0x01, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, /* p */
68      0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
69      0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
70      0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
71      0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
72      0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
73      0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
74      0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
75      0xff, 0xff},
76     {0x01, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, /* a = -3 */
77      0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
78      0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
79      0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
80      0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
81      0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
82      0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
83      0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
84      0xff, 0xfc},
85     {0x00, 0x51, 0x95, 0x3e, 0xb9, 0x61, 0x8e, 0x1c, /* b */
86      0x9a, 0x1f, 0x92, 0x9a, 0x21, 0xa0, 0xb6, 0x85,
87      0x40, 0xee, 0xa2, 0xda, 0x72, 0x5b, 0x99, 0xb3,
88      0x15, 0xf3, 0xb8, 0xb4, 0x89, 0x91, 0x8e, 0xf1,
89      0x09, 0xe1, 0x56, 0x19, 0x39, 0x51, 0xec, 0x7e,
90      0x93, 0x7b, 0x16, 0x52, 0xc0, 0xbd, 0x3b, 0xb1,
91      0xbf, 0x07, 0x35, 0x73, 0xdf, 0x88, 0x3d, 0x2c,
92      0x34, 0xf1, 0xef, 0x45, 0x1f, 0xd4, 0x6b, 0x50,
93      0x3f, 0x00},
94     {0x00, 0xc6, 0x85, 0x8e, 0x06, 0xb7, 0x04, 0x04, /* x */
95      0xe9, 0xcd, 0x9e, 0x3e, 0xcb, 0x66, 0x23, 0x95,
96      0xb4, 0x42, 0x9c, 0x64, 0x81, 0x39, 0x05, 0x3f,
97      0xb5, 0x21, 0xf8, 0x28, 0xaf, 0x60, 0x6b, 0x4d,
98      0x3d, 0xba, 0xa1, 0x4b, 0x5e, 0x77, 0xef, 0xe7,
99      0x59, 0x28, 0xfe, 0x1d, 0xc1, 0x27, 0xa2, 0xff,
100      0xa8, 0xde, 0x33, 0x48, 0xb3, 0xc1, 0x85, 0x6a,
101      0x42, 0x9b, 0xf9, 0x7e, 0x7e, 0x31, 0xc2, 0xe5,
102      0xbd, 0x66},
103     {0x01, 0x18, 0x39, 0x29, 0x6a, 0x78, 0x9a, 0x3b, /* y */
104      0xc0, 0x04, 0x5c, 0x8a, 0x5f, 0xb4, 0x2c, 0x7d,
105      0x1b, 0xd9, 0x98, 0xf5, 0x44, 0x49, 0x57, 0x9b,
106      0x44, 0x68, 0x17, 0xaf, 0xbd, 0x17, 0x27, 0x3e,
107      0x66, 0x2c, 0x97, 0xee, 0x72, 0x99, 0x5e, 0xf4,
108      0x26, 0x40, 0xc5, 0x50, 0xb9, 0x01, 0x3f, 0xad,
109      0x07, 0x61, 0x35, 0x3c, 0x70, 0x86, 0xa2, 0x72,
110      0xc2, 0x40, 0x88, 0xbe, 0x94, 0x76, 0x9f, 0xd1,
111      0x66, 0x50}
112 };
113 
114 /*-
115  * The representation of field elements.
116  * ------------------------------------
117  *
118  * We represent field elements with nine values. These values are either 64 or
119  * 128 bits and the field element represented is:
120  *   v[0]*2^0 + v[1]*2^58 + v[2]*2^116 + ... + v[8]*2^464  (mod p)
121  * Each of the nine values is called a 'limb'. Since the limbs are spaced only
122  * 58 bits apart, but are greater than 58 bits in length, the most significant
123  * bits of each limb overlap with the least significant bits of the next.
124  *
125  * A field element with 64-bit limbs is an 'felem'. One with 128-bit limbs is a
126  * 'largefelem' */
127 
128 # define NLIMBS 9
129 
130 typedef uint64_t limb;
131 typedef limb felem[NLIMBS];
132 typedef uint128_t largefelem[NLIMBS];
133 
134 static const limb bottom57bits = 0x1ffffffffffffff;
135 static const limb bottom58bits = 0x3ffffffffffffff;
136 
137 /*
138  * bin66_to_felem takes a little-endian byte array and converts it into felem
139  * form. This assumes that the CPU is little-endian.
140  */
141 static void bin66_to_felem(felem out, const u8 in[66])
142 {
143     out[0] = (*((limb *) & in[0])) & bottom58bits;
144     out[1] = (*((limb *) & in[7]) >> 2) & bottom58bits;
145     out[2] = (*((limb *) & in[14]) >> 4) & bottom58bits;
146     out[3] = (*((limb *) & in[21]) >> 6) & bottom58bits;
147     out[4] = (*((limb *) & in[29])) & bottom58bits;
148     out[5] = (*((limb *) & in[36]) >> 2) & bottom58bits;
149     out[6] = (*((limb *) & in[43]) >> 4) & bottom58bits;
150     out[7] = (*((limb *) & in[50]) >> 6) & bottom58bits;
151     out[8] = (*((limb *) & in[58])) & bottom57bits;
152 }
153 
154 /*
155  * felem_to_bin66 takes an felem and serialises into a little endian, 66 byte
156  * array. This assumes that the CPU is little-endian.
157  */
158 static void felem_to_bin66(u8 out[66], const felem in)
159 {
160     memset(out, 0, 66);
161     (*((limb *) & out[0])) = in[0];
162     (*((limb *) & out[7])) |= in[1] << 2;
163     (*((limb *) & out[14])) |= in[2] << 4;
164     (*((limb *) & out[21])) |= in[3] << 6;
165     (*((limb *) & out[29])) = in[4];
166     (*((limb *) & out[36])) |= in[5] << 2;
167     (*((limb *) & out[43])) |= in[6] << 4;
168     (*((limb *) & out[50])) |= in[7] << 6;
169     (*((limb *) & out[58])) = in[8];
170 }
171 
172 /* To preserve endianness when using BN_bn2bin and BN_bin2bn */
173 static void flip_endian(u8 *out, const u8 *in, unsigned len)
174 {
175     unsigned i;
176     for (i = 0; i < len; ++i)
177         out[i] = in[len - 1 - i];
178 }
179 
180 /* BN_to_felem converts an OpenSSL BIGNUM into an felem */
181 static int BN_to_felem(felem out, const BIGNUM *bn)
182 {
183     felem_bytearray b_in;
184     felem_bytearray b_out;
185     unsigned num_bytes;
186 
187     /* BN_bn2bin eats leading zeroes */
188     memset(b_out, 0, sizeof b_out);
189     num_bytes = BN_num_bytes(bn);
190     if (num_bytes > sizeof b_out) {
191         ECerr(EC_F_BN_TO_FELEM, EC_R_BIGNUM_OUT_OF_RANGE);
192         return 0;
193     }
194     if (BN_is_negative(bn)) {
195         ECerr(EC_F_BN_TO_FELEM, EC_R_BIGNUM_OUT_OF_RANGE);
196         return 0;
197     }
198     num_bytes = BN_bn2bin(bn, b_in);
199     flip_endian(b_out, b_in, num_bytes);
200     bin66_to_felem(out, b_out);
201     return 1;
202 }
203 
204 /* felem_to_BN converts an felem into an OpenSSL BIGNUM */
205 static BIGNUM *felem_to_BN(BIGNUM *out, const felem in)
206 {
207     felem_bytearray b_in, b_out;
208     felem_to_bin66(b_in, in);
209     flip_endian(b_out, b_in, sizeof b_out);
210     return BN_bin2bn(b_out, sizeof b_out, out);
211 }
212 
213 /*-
214  * Field operations
215  * ----------------
216  */
217 
218 static void felem_one(felem out)
219 {
220     out[0] = 1;
221     out[1] = 0;
222     out[2] = 0;
223     out[3] = 0;
224     out[4] = 0;
225     out[5] = 0;
226     out[6] = 0;
227     out[7] = 0;
228     out[8] = 0;
229 }
230 
231 static void felem_assign(felem out, const felem in)
232 {
233     out[0] = in[0];
234     out[1] = in[1];
235     out[2] = in[2];
236     out[3] = in[3];
237     out[4] = in[4];
238     out[5] = in[5];
239     out[6] = in[6];
240     out[7] = in[7];
241     out[8] = in[8];
242 }
243 
244 /* felem_sum64 sets out = out + in. */
245 static void felem_sum64(felem out, const felem in)
246 {
247     out[0] += in[0];
248     out[1] += in[1];
249     out[2] += in[2];
250     out[3] += in[3];
251     out[4] += in[4];
252     out[5] += in[5];
253     out[6] += in[6];
254     out[7] += in[7];
255     out[8] += in[8];
256 }
257 
258 /* felem_scalar sets out = in * scalar */
259 static void felem_scalar(felem out, const felem in, limb scalar)
260 {
261     out[0] = in[0] * scalar;
262     out[1] = in[1] * scalar;
263     out[2] = in[2] * scalar;
264     out[3] = in[3] * scalar;
265     out[4] = in[4] * scalar;
266     out[5] = in[5] * scalar;
267     out[6] = in[6] * scalar;
268     out[7] = in[7] * scalar;
269     out[8] = in[8] * scalar;
270 }
271 
272 /* felem_scalar64 sets out = out * scalar */
273 static void felem_scalar64(felem out, limb scalar)
274 {
275     out[0] *= scalar;
276     out[1] *= scalar;
277     out[2] *= scalar;
278     out[3] *= scalar;
279     out[4] *= scalar;
280     out[5] *= scalar;
281     out[6] *= scalar;
282     out[7] *= scalar;
283     out[8] *= scalar;
284 }
285 
286 /* felem_scalar128 sets out = out * scalar */
287 static void felem_scalar128(largefelem out, limb scalar)
288 {
289     out[0] *= scalar;
290     out[1] *= scalar;
291     out[2] *= scalar;
292     out[3] *= scalar;
293     out[4] *= scalar;
294     out[5] *= scalar;
295     out[6] *= scalar;
296     out[7] *= scalar;
297     out[8] *= scalar;
298 }
299 
300 /*-
301  * felem_neg sets |out| to |-in|
302  * On entry:
303  *   in[i] < 2^59 + 2^14
304  * On exit:
305  *   out[i] < 2^62
306  */
307 static void felem_neg(felem out, const felem in)
308 {
309     /* In order to prevent underflow, we subtract from 0 mod p. */
310     static const limb two62m3 = (((limb) 1) << 62) - (((limb) 1) << 5);
311     static const limb two62m2 = (((limb) 1) << 62) - (((limb) 1) << 4);
312 
313     out[0] = two62m3 - in[0];
314     out[1] = two62m2 - in[1];
315     out[2] = two62m2 - in[2];
316     out[3] = two62m2 - in[3];
317     out[4] = two62m2 - in[4];
318     out[5] = two62m2 - in[5];
319     out[6] = two62m2 - in[6];
320     out[7] = two62m2 - in[7];
321     out[8] = two62m2 - in[8];
322 }
323 
324 /*-
325  * felem_diff64 subtracts |in| from |out|
326  * On entry:
327  *   in[i] < 2^59 + 2^14
328  * On exit:
329  *   out[i] < out[i] + 2^62
330  */
331 static void felem_diff64(felem out, const felem in)
332 {
333     /*
334      * In order to prevent underflow, we add 0 mod p before subtracting.
335      */
336     static const limb two62m3 = (((limb) 1) << 62) - (((limb) 1) << 5);
337     static const limb two62m2 = (((limb) 1) << 62) - (((limb) 1) << 4);
338 
339     out[0] += two62m3 - in[0];
340     out[1] += two62m2 - in[1];
341     out[2] += two62m2 - in[2];
342     out[3] += two62m2 - in[3];
343     out[4] += two62m2 - in[4];
344     out[5] += two62m2 - in[5];
345     out[6] += two62m2 - in[6];
346     out[7] += two62m2 - in[7];
347     out[8] += two62m2 - in[8];
348 }
349 
350 /*-
351  * felem_diff_128_64 subtracts |in| from |out|
352  * On entry:
353  *   in[i] < 2^62 + 2^17
354  * On exit:
355  *   out[i] < out[i] + 2^63
356  */
357 static void felem_diff_128_64(largefelem out, const felem in)
358 {
359     /*
360      * In order to prevent underflow, we add 0 mod p before subtracting.
361      */
362     static const limb two63m6 = (((limb) 1) << 62) - (((limb) 1) << 5);
363     static const limb two63m5 = (((limb) 1) << 62) - (((limb) 1) << 4);
364 
365     out[0] += two63m6 - in[0];
366     out[1] += two63m5 - in[1];
367     out[2] += two63m5 - in[2];
368     out[3] += two63m5 - in[3];
369     out[4] += two63m5 - in[4];
370     out[5] += two63m5 - in[5];
371     out[6] += two63m5 - in[6];
372     out[7] += two63m5 - in[7];
373     out[8] += two63m5 - in[8];
374 }
375 
376 /*-
377  * felem_diff_128_64 subtracts |in| from |out|
378  * On entry:
379  *   in[i] < 2^126
380  * On exit:
381  *   out[i] < out[i] + 2^127 - 2^69
382  */
383 static void felem_diff128(largefelem out, const largefelem in)
384 {
385     /*
386      * In order to prevent underflow, we add 0 mod p before subtracting.
387      */
388     static const uint128_t two127m70 =
389         (((uint128_t) 1) << 127) - (((uint128_t) 1) << 70);
390     static const uint128_t two127m69 =
391         (((uint128_t) 1) << 127) - (((uint128_t) 1) << 69);
392 
393     out[0] += (two127m70 - in[0]);
394     out[1] += (two127m69 - in[1]);
395     out[2] += (two127m69 - in[2]);
396     out[3] += (two127m69 - in[3]);
397     out[4] += (two127m69 - in[4]);
398     out[5] += (two127m69 - in[5]);
399     out[6] += (two127m69 - in[6]);
400     out[7] += (two127m69 - in[7]);
401     out[8] += (two127m69 - in[8]);
402 }
403 
404 /*-
405  * felem_square sets |out| = |in|^2
406  * On entry:
407  *   in[i] < 2^62
408  * On exit:
409  *   out[i] < 17 * max(in[i]) * max(in[i])
410  */
411 static void felem_square(largefelem out, const felem in)
412 {
413     felem inx2, inx4;
414     felem_scalar(inx2, in, 2);
415     felem_scalar(inx4, in, 4);
416 
417     /*-
418      * We have many cases were we want to do
419      *   in[x] * in[y] +
420      *   in[y] * in[x]
421      * This is obviously just
422      *   2 * in[x] * in[y]
423      * However, rather than do the doubling on the 128 bit result, we
424      * double one of the inputs to the multiplication by reading from
425      * |inx2|
426      */
427 
428     out[0] = ((uint128_t) in[0]) * in[0];
429     out[1] = ((uint128_t) in[0]) * inx2[1];
430     out[2] = ((uint128_t) in[0]) * inx2[2] + ((uint128_t) in[1]) * in[1];
431     out[3] = ((uint128_t) in[0]) * inx2[3] + ((uint128_t) in[1]) * inx2[2];
432     out[4] = ((uint128_t) in[0]) * inx2[4] +
433         ((uint128_t) in[1]) * inx2[3] + ((uint128_t) in[2]) * in[2];
434     out[5] = ((uint128_t) in[0]) * inx2[5] +
435         ((uint128_t) in[1]) * inx2[4] + ((uint128_t) in[2]) * inx2[3];
436     out[6] = ((uint128_t) in[0]) * inx2[6] +
437         ((uint128_t) in[1]) * inx2[5] +
438         ((uint128_t) in[2]) * inx2[4] + ((uint128_t) in[3]) * in[3];
439     out[7] = ((uint128_t) in[0]) * inx2[7] +
440         ((uint128_t) in[1]) * inx2[6] +
441         ((uint128_t) in[2]) * inx2[5] + ((uint128_t) in[3]) * inx2[4];
442     out[8] = ((uint128_t) in[0]) * inx2[8] +
443         ((uint128_t) in[1]) * inx2[7] +
444         ((uint128_t) in[2]) * inx2[6] +
445         ((uint128_t) in[3]) * inx2[5] + ((uint128_t) in[4]) * in[4];
446 
447     /*
448      * The remaining limbs fall above 2^521, with the first falling at 2^522.
449      * They correspond to locations one bit up from the limbs produced above
450      * so we would have to multiply by two to align them. Again, rather than
451      * operate on the 128-bit result, we double one of the inputs to the
452      * multiplication. If we want to double for both this reason, and the
453      * reason above, then we end up multiplying by four.
454      */
455 
456     /* 9 */
457     out[0] += ((uint128_t) in[1]) * inx4[8] +
458         ((uint128_t) in[2]) * inx4[7] +
459         ((uint128_t) in[3]) * inx4[6] + ((uint128_t) in[4]) * inx4[5];
460 
461     /* 10 */
462     out[1] += ((uint128_t) in[2]) * inx4[8] +
463         ((uint128_t) in[3]) * inx4[7] +
464         ((uint128_t) in[4]) * inx4[6] + ((uint128_t) in[5]) * inx2[5];
465 
466     /* 11 */
467     out[2] += ((uint128_t) in[3]) * inx4[8] +
468         ((uint128_t) in[4]) * inx4[7] + ((uint128_t) in[5]) * inx4[6];
469 
470     /* 12 */
471     out[3] += ((uint128_t) in[4]) * inx4[8] +
472         ((uint128_t) in[5]) * inx4[7] + ((uint128_t) in[6]) * inx2[6];
473 
474     /* 13 */
475     out[4] += ((uint128_t) in[5]) * inx4[8] + ((uint128_t) in[6]) * inx4[7];
476 
477     /* 14 */
478     out[5] += ((uint128_t) in[6]) * inx4[8] + ((uint128_t) in[7]) * inx2[7];
479 
480     /* 15 */
481     out[6] += ((uint128_t) in[7]) * inx4[8];
482 
483     /* 16 */
484     out[7] += ((uint128_t) in[8]) * inx2[8];
485 }
486 
487 /*-
488  * felem_mul sets |out| = |in1| * |in2|
489  * On entry:
490  *   in1[i] < 2^64
491  *   in2[i] < 2^63
492  * On exit:
493  *   out[i] < 17 * max(in1[i]) * max(in2[i])
494  */
495 static void felem_mul(largefelem out, const felem in1, const felem in2)
496 {
497     felem in2x2;
498     felem_scalar(in2x2, in2, 2);
499 
500     out[0] = ((uint128_t) in1[0]) * in2[0];
501 
502     out[1] = ((uint128_t) in1[0]) * in2[1] + ((uint128_t) in1[1]) * in2[0];
503 
504     out[2] = ((uint128_t) in1[0]) * in2[2] +
505         ((uint128_t) in1[1]) * in2[1] + ((uint128_t) in1[2]) * in2[0];
506 
507     out[3] = ((uint128_t) in1[0]) * in2[3] +
508         ((uint128_t) in1[1]) * in2[2] +
509         ((uint128_t) in1[2]) * in2[1] + ((uint128_t) in1[3]) * in2[0];
510 
511     out[4] = ((uint128_t) in1[0]) * in2[4] +
512         ((uint128_t) in1[1]) * in2[3] +
513         ((uint128_t) in1[2]) * in2[2] +
514         ((uint128_t) in1[3]) * in2[1] + ((uint128_t) in1[4]) * in2[0];
515 
516     out[5] = ((uint128_t) in1[0]) * in2[5] +
517         ((uint128_t) in1[1]) * in2[4] +
518         ((uint128_t) in1[2]) * in2[3] +
519         ((uint128_t) in1[3]) * in2[2] +
520         ((uint128_t) in1[4]) * in2[1] + ((uint128_t) in1[5]) * in2[0];
521 
522     out[6] = ((uint128_t) in1[0]) * in2[6] +
523         ((uint128_t) in1[1]) * in2[5] +
524         ((uint128_t) in1[2]) * in2[4] +
525         ((uint128_t) in1[3]) * in2[3] +
526         ((uint128_t) in1[4]) * in2[2] +
527         ((uint128_t) in1[5]) * in2[1] + ((uint128_t) in1[6]) * in2[0];
528 
529     out[7] = ((uint128_t) in1[0]) * in2[7] +
530         ((uint128_t) in1[1]) * in2[6] +
531         ((uint128_t) in1[2]) * in2[5] +
532         ((uint128_t) in1[3]) * in2[4] +
533         ((uint128_t) in1[4]) * in2[3] +
534         ((uint128_t) in1[5]) * in2[2] +
535         ((uint128_t) in1[6]) * in2[1] + ((uint128_t) in1[7]) * in2[0];
536 
537     out[8] = ((uint128_t) in1[0]) * in2[8] +
538         ((uint128_t) in1[1]) * in2[7] +
539         ((uint128_t) in1[2]) * in2[6] +
540         ((uint128_t) in1[3]) * in2[5] +
541         ((uint128_t) in1[4]) * in2[4] +
542         ((uint128_t) in1[5]) * in2[3] +
543         ((uint128_t) in1[6]) * in2[2] +
544         ((uint128_t) in1[7]) * in2[1] + ((uint128_t) in1[8]) * in2[0];
545 
546     /* See comment in felem_square about the use of in2x2 here */
547 
548     out[0] += ((uint128_t) in1[1]) * in2x2[8] +
549         ((uint128_t) in1[2]) * in2x2[7] +
550         ((uint128_t) in1[3]) * in2x2[6] +
551         ((uint128_t) in1[4]) * in2x2[5] +
552         ((uint128_t) in1[5]) * in2x2[4] +
553         ((uint128_t) in1[6]) * in2x2[3] +
554         ((uint128_t) in1[7]) * in2x2[2] + ((uint128_t) in1[8]) * in2x2[1];
555 
556     out[1] += ((uint128_t) in1[2]) * in2x2[8] +
557         ((uint128_t) in1[3]) * in2x2[7] +
558         ((uint128_t) in1[4]) * in2x2[6] +
559         ((uint128_t) in1[5]) * in2x2[5] +
560         ((uint128_t) in1[6]) * in2x2[4] +
561         ((uint128_t) in1[7]) * in2x2[3] + ((uint128_t) in1[8]) * in2x2[2];
562 
563     out[2] += ((uint128_t) in1[3]) * in2x2[8] +
564         ((uint128_t) in1[4]) * in2x2[7] +
565         ((uint128_t) in1[5]) * in2x2[6] +
566         ((uint128_t) in1[6]) * in2x2[5] +
567         ((uint128_t) in1[7]) * in2x2[4] + ((uint128_t) in1[8]) * in2x2[3];
568 
569     out[3] += ((uint128_t) in1[4]) * in2x2[8] +
570         ((uint128_t) in1[5]) * in2x2[7] +
571         ((uint128_t) in1[6]) * in2x2[6] +
572         ((uint128_t) in1[7]) * in2x2[5] + ((uint128_t) in1[8]) * in2x2[4];
573 
574     out[4] += ((uint128_t) in1[5]) * in2x2[8] +
575         ((uint128_t) in1[6]) * in2x2[7] +
576         ((uint128_t) in1[7]) * in2x2[6] + ((uint128_t) in1[8]) * in2x2[5];
577 
578     out[5] += ((uint128_t) in1[6]) * in2x2[8] +
579         ((uint128_t) in1[7]) * in2x2[7] + ((uint128_t) in1[8]) * in2x2[6];
580 
581     out[6] += ((uint128_t) in1[7]) * in2x2[8] +
582         ((uint128_t) in1[8]) * in2x2[7];
583 
584     out[7] += ((uint128_t) in1[8]) * in2x2[8];
585 }
586 
587 static const limb bottom52bits = 0xfffffffffffff;
588 
589 /*-
590  * felem_reduce converts a largefelem to an felem.
591  * On entry:
592  *   in[i] < 2^128
593  * On exit:
594  *   out[i] < 2^59 + 2^14
595  */
596 static void felem_reduce(felem out, const largefelem in)
597 {
598     u64 overflow1, overflow2;
599 
600     out[0] = ((limb) in[0]) & bottom58bits;
601     out[1] = ((limb) in[1]) & bottom58bits;
602     out[2] = ((limb) in[2]) & bottom58bits;
603     out[3] = ((limb) in[3]) & bottom58bits;
604     out[4] = ((limb) in[4]) & bottom58bits;
605     out[5] = ((limb) in[5]) & bottom58bits;
606     out[6] = ((limb) in[6]) & bottom58bits;
607     out[7] = ((limb) in[7]) & bottom58bits;
608     out[8] = ((limb) in[8]) & bottom58bits;
609 
610     /* out[i] < 2^58 */
611 
612     out[1] += ((limb) in[0]) >> 58;
613     out[1] += (((limb) (in[0] >> 64)) & bottom52bits) << 6;
614     /*-
615      * out[1] < 2^58 + 2^6 + 2^58
616      *        = 2^59 + 2^6
617      */
618     out[2] += ((limb) (in[0] >> 64)) >> 52;
619 
620     out[2] += ((limb) in[1]) >> 58;
621     out[2] += (((limb) (in[1] >> 64)) & bottom52bits) << 6;
622     out[3] += ((limb) (in[1] >> 64)) >> 52;
623 
624     out[3] += ((limb) in[2]) >> 58;
625     out[3] += (((limb) (in[2] >> 64)) & bottom52bits) << 6;
626     out[4] += ((limb) (in[2] >> 64)) >> 52;
627 
628     out[4] += ((limb) in[3]) >> 58;
629     out[4] += (((limb) (in[3] >> 64)) & bottom52bits) << 6;
630     out[5] += ((limb) (in[3] >> 64)) >> 52;
631 
632     out[5] += ((limb) in[4]) >> 58;
633     out[5] += (((limb) (in[4] >> 64)) & bottom52bits) << 6;
634     out[6] += ((limb) (in[4] >> 64)) >> 52;
635 
636     out[6] += ((limb) in[5]) >> 58;
637     out[6] += (((limb) (in[5] >> 64)) & bottom52bits) << 6;
638     out[7] += ((limb) (in[5] >> 64)) >> 52;
639 
640     out[7] += ((limb) in[6]) >> 58;
641     out[7] += (((limb) (in[6] >> 64)) & bottom52bits) << 6;
642     out[8] += ((limb) (in[6] >> 64)) >> 52;
643 
644     out[8] += ((limb) in[7]) >> 58;
645     out[8] += (((limb) (in[7] >> 64)) & bottom52bits) << 6;
646     /*-
647      * out[x > 1] < 2^58 + 2^6 + 2^58 + 2^12
648      *            < 2^59 + 2^13
649      */
650     overflow1 = ((limb) (in[7] >> 64)) >> 52;
651 
652     overflow1 += ((limb) in[8]) >> 58;
653     overflow1 += (((limb) (in[8] >> 64)) & bottom52bits) << 6;
654     overflow2 = ((limb) (in[8] >> 64)) >> 52;
655 
656     overflow1 <<= 1;            /* overflow1 < 2^13 + 2^7 + 2^59 */
657     overflow2 <<= 1;            /* overflow2 < 2^13 */
658 
659     out[0] += overflow1;        /* out[0] < 2^60 */
660     out[1] += overflow2;        /* out[1] < 2^59 + 2^6 + 2^13 */
661 
662     out[1] += out[0] >> 58;
663     out[0] &= bottom58bits;
664     /*-
665      * out[0] < 2^58
666      * out[1] < 2^59 + 2^6 + 2^13 + 2^2
667      *        < 2^59 + 2^14
668      */
669 }
670 
671 static void felem_square_reduce(felem out, const felem in)
672 {
673     largefelem tmp;
674     felem_square(tmp, in);
675     felem_reduce(out, tmp);
676 }
677 
678 static void felem_mul_reduce(felem out, const felem in1, const felem in2)
679 {
680     largefelem tmp;
681     felem_mul(tmp, in1, in2);
682     felem_reduce(out, tmp);
683 }
684 
685 /*-
686  * felem_inv calculates |out| = |in|^{-1}
687  *
688  * Based on Fermat's Little Theorem:
689  *   a^p = a (mod p)
690  *   a^{p-1} = 1 (mod p)
691  *   a^{p-2} = a^{-1} (mod p)
692  */
693 static void felem_inv(felem out, const felem in)
694 {
695     felem ftmp, ftmp2, ftmp3, ftmp4;
696     largefelem tmp;
697     unsigned i;
698 
699     felem_square(tmp, in);
700     felem_reduce(ftmp, tmp);    /* 2^1 */
701     felem_mul(tmp, in, ftmp);
702     felem_reduce(ftmp, tmp);    /* 2^2 - 2^0 */
703     felem_assign(ftmp2, ftmp);
704     felem_square(tmp, ftmp);
705     felem_reduce(ftmp, tmp);    /* 2^3 - 2^1 */
706     felem_mul(tmp, in, ftmp);
707     felem_reduce(ftmp, tmp);    /* 2^3 - 2^0 */
708     felem_square(tmp, ftmp);
709     felem_reduce(ftmp, tmp);    /* 2^4 - 2^1 */
710 
711     felem_square(tmp, ftmp2);
712     felem_reduce(ftmp3, tmp);   /* 2^3 - 2^1 */
713     felem_square(tmp, ftmp3);
714     felem_reduce(ftmp3, tmp);   /* 2^4 - 2^2 */
715     felem_mul(tmp, ftmp3, ftmp2);
716     felem_reduce(ftmp3, tmp);   /* 2^4 - 2^0 */
717 
718     felem_assign(ftmp2, ftmp3);
719     felem_square(tmp, ftmp3);
720     felem_reduce(ftmp3, tmp);   /* 2^5 - 2^1 */
721     felem_square(tmp, ftmp3);
722     felem_reduce(ftmp3, tmp);   /* 2^6 - 2^2 */
723     felem_square(tmp, ftmp3);
724     felem_reduce(ftmp3, tmp);   /* 2^7 - 2^3 */
725     felem_square(tmp, ftmp3);
726     felem_reduce(ftmp3, tmp);   /* 2^8 - 2^4 */
727     felem_assign(ftmp4, ftmp3);
728     felem_mul(tmp, ftmp3, ftmp);
729     felem_reduce(ftmp4, tmp);   /* 2^8 - 2^1 */
730     felem_square(tmp, ftmp4);
731     felem_reduce(ftmp4, tmp);   /* 2^9 - 2^2 */
732     felem_mul(tmp, ftmp3, ftmp2);
733     felem_reduce(ftmp3, tmp);   /* 2^8 - 2^0 */
734     felem_assign(ftmp2, ftmp3);
735 
736     for (i = 0; i < 8; i++) {
737         felem_square(tmp, ftmp3);
738         felem_reduce(ftmp3, tmp); /* 2^16 - 2^8 */
739     }
740     felem_mul(tmp, ftmp3, ftmp2);
741     felem_reduce(ftmp3, tmp);   /* 2^16 - 2^0 */
742     felem_assign(ftmp2, ftmp3);
743 
744     for (i = 0; i < 16; i++) {
745         felem_square(tmp, ftmp3);
746         felem_reduce(ftmp3, tmp); /* 2^32 - 2^16 */
747     }
748     felem_mul(tmp, ftmp3, ftmp2);
749     felem_reduce(ftmp3, tmp);   /* 2^32 - 2^0 */
750     felem_assign(ftmp2, ftmp3);
751 
752     for (i = 0; i < 32; i++) {
753         felem_square(tmp, ftmp3);
754         felem_reduce(ftmp3, tmp); /* 2^64 - 2^32 */
755     }
756     felem_mul(tmp, ftmp3, ftmp2);
757     felem_reduce(ftmp3, tmp);   /* 2^64 - 2^0 */
758     felem_assign(ftmp2, ftmp3);
759 
760     for (i = 0; i < 64; i++) {
761         felem_square(tmp, ftmp3);
762         felem_reduce(ftmp3, tmp); /* 2^128 - 2^64 */
763     }
764     felem_mul(tmp, ftmp3, ftmp2);
765     felem_reduce(ftmp3, tmp);   /* 2^128 - 2^0 */
766     felem_assign(ftmp2, ftmp3);
767 
768     for (i = 0; i < 128; i++) {
769         felem_square(tmp, ftmp3);
770         felem_reduce(ftmp3, tmp); /* 2^256 - 2^128 */
771     }
772     felem_mul(tmp, ftmp3, ftmp2);
773     felem_reduce(ftmp3, tmp);   /* 2^256 - 2^0 */
774     felem_assign(ftmp2, ftmp3);
775 
776     for (i = 0; i < 256; i++) {
777         felem_square(tmp, ftmp3);
778         felem_reduce(ftmp3, tmp); /* 2^512 - 2^256 */
779     }
780     felem_mul(tmp, ftmp3, ftmp2);
781     felem_reduce(ftmp3, tmp);   /* 2^512 - 2^0 */
782 
783     for (i = 0; i < 9; i++) {
784         felem_square(tmp, ftmp3);
785         felem_reduce(ftmp3, tmp); /* 2^521 - 2^9 */
786     }
787     felem_mul(tmp, ftmp3, ftmp4);
788     felem_reduce(ftmp3, tmp);   /* 2^512 - 2^2 */
789     felem_mul(tmp, ftmp3, in);
790     felem_reduce(out, tmp);     /* 2^512 - 3 */
791 }
792 
793 /* This is 2^521-1, expressed as an felem */
794 static const felem kPrime = {
795     0x03ffffffffffffff, 0x03ffffffffffffff, 0x03ffffffffffffff,
796     0x03ffffffffffffff, 0x03ffffffffffffff, 0x03ffffffffffffff,
797     0x03ffffffffffffff, 0x03ffffffffffffff, 0x01ffffffffffffff
798 };
799 
800 /*-
801  * felem_is_zero returns a limb with all bits set if |in| == 0 (mod p) and 0
802  * otherwise.
803  * On entry:
804  *   in[i] < 2^59 + 2^14
805  */
806 static limb felem_is_zero(const felem in)
807 {
808     felem ftmp;
809     limb is_zero, is_p;
810     felem_assign(ftmp, in);
811 
812     ftmp[0] += ftmp[8] >> 57;
813     ftmp[8] &= bottom57bits;
814     /* ftmp[8] < 2^57 */
815     ftmp[1] += ftmp[0] >> 58;
816     ftmp[0] &= bottom58bits;
817     ftmp[2] += ftmp[1] >> 58;
818     ftmp[1] &= bottom58bits;
819     ftmp[3] += ftmp[2] >> 58;
820     ftmp[2] &= bottom58bits;
821     ftmp[4] += ftmp[3] >> 58;
822     ftmp[3] &= bottom58bits;
823     ftmp[5] += ftmp[4] >> 58;
824     ftmp[4] &= bottom58bits;
825     ftmp[6] += ftmp[5] >> 58;
826     ftmp[5] &= bottom58bits;
827     ftmp[7] += ftmp[6] >> 58;
828     ftmp[6] &= bottom58bits;
829     ftmp[8] += ftmp[7] >> 58;
830     ftmp[7] &= bottom58bits;
831     /* ftmp[8] < 2^57 + 4 */
832 
833     /*
834      * The ninth limb of 2*(2^521-1) is 0x03ffffffffffffff, which is greater
835      * than our bound for ftmp[8]. Therefore we only have to check if the
836      * zero is zero or 2^521-1.
837      */
838 
839     is_zero = 0;
840     is_zero |= ftmp[0];
841     is_zero |= ftmp[1];
842     is_zero |= ftmp[2];
843     is_zero |= ftmp[3];
844     is_zero |= ftmp[4];
845     is_zero |= ftmp[5];
846     is_zero |= ftmp[6];
847     is_zero |= ftmp[7];
848     is_zero |= ftmp[8];
849 
850     is_zero--;
851     /*
852      * We know that ftmp[i] < 2^63, therefore the only way that the top bit
853      * can be set is if is_zero was 0 before the decrement.
854      */
855     is_zero = ((s64) is_zero) >> 63;
856 
857     is_p = ftmp[0] ^ kPrime[0];
858     is_p |= ftmp[1] ^ kPrime[1];
859     is_p |= ftmp[2] ^ kPrime[2];
860     is_p |= ftmp[3] ^ kPrime[3];
861     is_p |= ftmp[4] ^ kPrime[4];
862     is_p |= ftmp[5] ^ kPrime[5];
863     is_p |= ftmp[6] ^ kPrime[6];
864     is_p |= ftmp[7] ^ kPrime[7];
865     is_p |= ftmp[8] ^ kPrime[8];
866 
867     is_p--;
868     is_p = ((s64) is_p) >> 63;
869 
870     is_zero |= is_p;
871     return is_zero;
872 }
873 
874 static int felem_is_zero_int(const felem in)
875 {
876     return (int)(felem_is_zero(in) & ((limb) 1));
877 }
878 
879 /*-
880  * felem_contract converts |in| to its unique, minimal representation.
881  * On entry:
882  *   in[i] < 2^59 + 2^14
883  */
884 static void felem_contract(felem out, const felem in)
885 {
886     limb is_p, is_greater, sign;
887     static const limb two58 = ((limb) 1) << 58;
888 
889     felem_assign(out, in);
890 
891     out[0] += out[8] >> 57;
892     out[8] &= bottom57bits;
893     /* out[8] < 2^57 */
894     out[1] += out[0] >> 58;
895     out[0] &= bottom58bits;
896     out[2] += out[1] >> 58;
897     out[1] &= bottom58bits;
898     out[3] += out[2] >> 58;
899     out[2] &= bottom58bits;
900     out[4] += out[3] >> 58;
901     out[3] &= bottom58bits;
902     out[5] += out[4] >> 58;
903     out[4] &= bottom58bits;
904     out[6] += out[5] >> 58;
905     out[5] &= bottom58bits;
906     out[7] += out[6] >> 58;
907     out[6] &= bottom58bits;
908     out[8] += out[7] >> 58;
909     out[7] &= bottom58bits;
910     /* out[8] < 2^57 + 4 */
911 
912     /*
913      * If the value is greater than 2^521-1 then we have to subtract 2^521-1
914      * out. See the comments in felem_is_zero regarding why we don't test for
915      * other multiples of the prime.
916      */
917 
918     /*
919      * First, if |out| is equal to 2^521-1, we subtract it out to get zero.
920      */
921 
922     is_p = out[0] ^ kPrime[0];
923     is_p |= out[1] ^ kPrime[1];
924     is_p |= out[2] ^ kPrime[2];
925     is_p |= out[3] ^ kPrime[3];
926     is_p |= out[4] ^ kPrime[4];
927     is_p |= out[5] ^ kPrime[5];
928     is_p |= out[6] ^ kPrime[6];
929     is_p |= out[7] ^ kPrime[7];
930     is_p |= out[8] ^ kPrime[8];
931 
932     is_p--;
933     is_p &= is_p << 32;
934     is_p &= is_p << 16;
935     is_p &= is_p << 8;
936     is_p &= is_p << 4;
937     is_p &= is_p << 2;
938     is_p &= is_p << 1;
939     is_p = ((s64) is_p) >> 63;
940     is_p = ~is_p;
941 
942     /* is_p is 0 iff |out| == 2^521-1 and all ones otherwise */
943 
944     out[0] &= is_p;
945     out[1] &= is_p;
946     out[2] &= is_p;
947     out[3] &= is_p;
948     out[4] &= is_p;
949     out[5] &= is_p;
950     out[6] &= is_p;
951     out[7] &= is_p;
952     out[8] &= is_p;
953 
954     /*
955      * In order to test that |out| >= 2^521-1 we need only test if out[8] >>
956      * 57 is greater than zero as (2^521-1) + x >= 2^522
957      */
958     is_greater = out[8] >> 57;
959     is_greater |= is_greater << 32;
960     is_greater |= is_greater << 16;
961     is_greater |= is_greater << 8;
962     is_greater |= is_greater << 4;
963     is_greater |= is_greater << 2;
964     is_greater |= is_greater << 1;
965     is_greater = ((s64) is_greater) >> 63;
966 
967     out[0] -= kPrime[0] & is_greater;
968     out[1] -= kPrime[1] & is_greater;
969     out[2] -= kPrime[2] & is_greater;
970     out[3] -= kPrime[3] & is_greater;
971     out[4] -= kPrime[4] & is_greater;
972     out[5] -= kPrime[5] & is_greater;
973     out[6] -= kPrime[6] & is_greater;
974     out[7] -= kPrime[7] & is_greater;
975     out[8] -= kPrime[8] & is_greater;
976 
977     /* Eliminate negative coefficients */
978     sign = -(out[0] >> 63);
979     out[0] += (two58 & sign);
980     out[1] -= (1 & sign);
981     sign = -(out[1] >> 63);
982     out[1] += (two58 & sign);
983     out[2] -= (1 & sign);
984     sign = -(out[2] >> 63);
985     out[2] += (two58 & sign);
986     out[3] -= (1 & sign);
987     sign = -(out[3] >> 63);
988     out[3] += (two58 & sign);
989     out[4] -= (1 & sign);
990     sign = -(out[4] >> 63);
991     out[4] += (two58 & sign);
992     out[5] -= (1 & sign);
993     sign = -(out[0] >> 63);
994     out[5] += (two58 & sign);
995     out[6] -= (1 & sign);
996     sign = -(out[6] >> 63);
997     out[6] += (two58 & sign);
998     out[7] -= (1 & sign);
999     sign = -(out[7] >> 63);
1000     out[7] += (two58 & sign);
1001     out[8] -= (1 & sign);
1002     sign = -(out[5] >> 63);
1003     out[5] += (two58 & sign);
1004     out[6] -= (1 & sign);
1005     sign = -(out[6] >> 63);
1006     out[6] += (two58 & sign);
1007     out[7] -= (1 & sign);
1008     sign = -(out[7] >> 63);
1009     out[7] += (two58 & sign);
1010     out[8] -= (1 & sign);
1011 }
1012 
1013 /*-
1014  * Group operations
1015  * ----------------
1016  *
1017  * Building on top of the field operations we have the operations on the
1018  * elliptic curve group itself. Points on the curve are represented in Jacobian
1019  * coordinates */
1020 
1021 /*-
1022  * point_double calcuates 2*(x_in, y_in, z_in)
1023  *
1024  * The method is taken from:
1025  *   http://hyperelliptic.org/EFD/g1p/auto-shortw-jacobian-3.html#doubling-dbl-2001-b
1026  *
1027  * Outputs can equal corresponding inputs, i.e., x_out == x_in is allowed.
1028  * while x_out == y_in is not (maybe this works, but it's not tested). */
1029 static void
1030 point_double(felem x_out, felem y_out, felem z_out,
1031              const felem x_in, const felem y_in, const felem z_in)
1032 {
1033     largefelem tmp, tmp2;
1034     felem delta, gamma, beta, alpha, ftmp, ftmp2;
1035 
1036     felem_assign(ftmp, x_in);
1037     felem_assign(ftmp2, x_in);
1038 
1039     /* delta = z^2 */
1040     felem_square(tmp, z_in);
1041     felem_reduce(delta, tmp);   /* delta[i] < 2^59 + 2^14 */
1042 
1043     /* gamma = y^2 */
1044     felem_square(tmp, y_in);
1045     felem_reduce(gamma, tmp);   /* gamma[i] < 2^59 + 2^14 */
1046 
1047     /* beta = x*gamma */
1048     felem_mul(tmp, x_in, gamma);
1049     felem_reduce(beta, tmp);    /* beta[i] < 2^59 + 2^14 */
1050 
1051     /* alpha = 3*(x-delta)*(x+delta) */
1052     felem_diff64(ftmp, delta);
1053     /* ftmp[i] < 2^61 */
1054     felem_sum64(ftmp2, delta);
1055     /* ftmp2[i] < 2^60 + 2^15 */
1056     felem_scalar64(ftmp2, 3);
1057     /* ftmp2[i] < 3*2^60 + 3*2^15 */
1058     felem_mul(tmp, ftmp, ftmp2);
1059     /*-
1060      * tmp[i] < 17(3*2^121 + 3*2^76)
1061      *        = 61*2^121 + 61*2^76
1062      *        < 64*2^121 + 64*2^76
1063      *        = 2^127 + 2^82
1064      *        < 2^128
1065      */
1066     felem_reduce(alpha, tmp);
1067 
1068     /* x' = alpha^2 - 8*beta */
1069     felem_square(tmp, alpha);
1070     /*
1071      * tmp[i] < 17*2^120 < 2^125
1072      */
1073     felem_assign(ftmp, beta);
1074     felem_scalar64(ftmp, 8);
1075     /* ftmp[i] < 2^62 + 2^17 */
1076     felem_diff_128_64(tmp, ftmp);
1077     /* tmp[i] < 2^125 + 2^63 + 2^62 + 2^17 */
1078     felem_reduce(x_out, tmp);
1079 
1080     /* z' = (y + z)^2 - gamma - delta */
1081     felem_sum64(delta, gamma);
1082     /* delta[i] < 2^60 + 2^15 */
1083     felem_assign(ftmp, y_in);
1084     felem_sum64(ftmp, z_in);
1085     /* ftmp[i] < 2^60 + 2^15 */
1086     felem_square(tmp, ftmp);
1087     /*
1088      * tmp[i] < 17(2^122) < 2^127
1089      */
1090     felem_diff_128_64(tmp, delta);
1091     /* tmp[i] < 2^127 + 2^63 */
1092     felem_reduce(z_out, tmp);
1093 
1094     /* y' = alpha*(4*beta - x') - 8*gamma^2 */
1095     felem_scalar64(beta, 4);
1096     /* beta[i] < 2^61 + 2^16 */
1097     felem_diff64(beta, x_out);
1098     /* beta[i] < 2^61 + 2^60 + 2^16 */
1099     felem_mul(tmp, alpha, beta);
1100     /*-
1101      * tmp[i] < 17*((2^59 + 2^14)(2^61 + 2^60 + 2^16))
1102      *        = 17*(2^120 + 2^75 + 2^119 + 2^74 + 2^75 + 2^30)
1103      *        = 17*(2^120 + 2^119 + 2^76 + 2^74 + 2^30)
1104      *        < 2^128
1105      */
1106     felem_square(tmp2, gamma);
1107     /*-
1108      * tmp2[i] < 17*(2^59 + 2^14)^2
1109      *         = 17*(2^118 + 2^74 + 2^28)
1110      */
1111     felem_scalar128(tmp2, 8);
1112     /*-
1113      * tmp2[i] < 8*17*(2^118 + 2^74 + 2^28)
1114      *         = 2^125 + 2^121 + 2^81 + 2^77 + 2^35 + 2^31
1115      *         < 2^126
1116      */
1117     felem_diff128(tmp, tmp2);
1118     /*-
1119      * tmp[i] < 2^127 - 2^69 + 17(2^120 + 2^119 + 2^76 + 2^74 + 2^30)
1120      *        = 2^127 + 2^124 + 2^122 + 2^120 + 2^118 + 2^80 + 2^78 + 2^76 +
1121      *          2^74 + 2^69 + 2^34 + 2^30
1122      *        < 2^128
1123      */
1124     felem_reduce(y_out, tmp);
1125 }
1126 
1127 /* copy_conditional copies in to out iff mask is all ones. */
1128 static void copy_conditional(felem out, const felem in, limb mask)
1129 {
1130     unsigned i;
1131     for (i = 0; i < NLIMBS; ++i) {
1132         const limb tmp = mask & (in[i] ^ out[i]);
1133         out[i] ^= tmp;
1134     }
1135 }
1136 
1137 /*-
1138  * point_add calcuates (x1, y1, z1) + (x2, y2, z2)
1139  *
1140  * The method is taken from
1141  *   http://hyperelliptic.org/EFD/g1p/auto-shortw-jacobian-3.html#addition-add-2007-bl,
1142  * adapted for mixed addition (z2 = 1, or z2 = 0 for the point at infinity).
1143  *
1144  * This function includes a branch for checking whether the two input points
1145  * are equal (while not equal to the point at infinity). This case never
1146  * happens during single point multiplication, so there is no timing leak for
1147  * ECDH or ECDSA signing. */
1148 static void point_add(felem x3, felem y3, felem z3,
1149                       const felem x1, const felem y1, const felem z1,
1150                       const int mixed, const felem x2, const felem y2,
1151                       const felem z2)
1152 {
1153     felem ftmp, ftmp2, ftmp3, ftmp4, ftmp5, ftmp6, x_out, y_out, z_out;
1154     largefelem tmp, tmp2;
1155     limb x_equal, y_equal, z1_is_zero, z2_is_zero;
1156 
1157     z1_is_zero = felem_is_zero(z1);
1158     z2_is_zero = felem_is_zero(z2);
1159 
1160     /* ftmp = z1z1 = z1**2 */
1161     felem_square(tmp, z1);
1162     felem_reduce(ftmp, tmp);
1163 
1164     if (!mixed) {
1165         /* ftmp2 = z2z2 = z2**2 */
1166         felem_square(tmp, z2);
1167         felem_reduce(ftmp2, tmp);
1168 
1169         /* u1 = ftmp3 = x1*z2z2 */
1170         felem_mul(tmp, x1, ftmp2);
1171         felem_reduce(ftmp3, tmp);
1172 
1173         /* ftmp5 = z1 + z2 */
1174         felem_assign(ftmp5, z1);
1175         felem_sum64(ftmp5, z2);
1176         /* ftmp5[i] < 2^61 */
1177 
1178         /* ftmp5 = (z1 + z2)**2 - z1z1 - z2z2 = 2*z1z2 */
1179         felem_square(tmp, ftmp5);
1180         /* tmp[i] < 17*2^122 */
1181         felem_diff_128_64(tmp, ftmp);
1182         /* tmp[i] < 17*2^122 + 2^63 */
1183         felem_diff_128_64(tmp, ftmp2);
1184         /* tmp[i] < 17*2^122 + 2^64 */
1185         felem_reduce(ftmp5, tmp);
1186 
1187         /* ftmp2 = z2 * z2z2 */
1188         felem_mul(tmp, ftmp2, z2);
1189         felem_reduce(ftmp2, tmp);
1190 
1191         /* s1 = ftmp6 = y1 * z2**3 */
1192         felem_mul(tmp, y1, ftmp2);
1193         felem_reduce(ftmp6, tmp);
1194     } else {
1195         /*
1196          * We'll assume z2 = 1 (special case z2 = 0 is handled later)
1197          */
1198 
1199         /* u1 = ftmp3 = x1*z2z2 */
1200         felem_assign(ftmp3, x1);
1201 
1202         /* ftmp5 = 2*z1z2 */
1203         felem_scalar(ftmp5, z1, 2);
1204 
1205         /* s1 = ftmp6 = y1 * z2**3 */
1206         felem_assign(ftmp6, y1);
1207     }
1208 
1209     /* u2 = x2*z1z1 */
1210     felem_mul(tmp, x2, ftmp);
1211     /* tmp[i] < 17*2^120 */
1212 
1213     /* h = ftmp4 = u2 - u1 */
1214     felem_diff_128_64(tmp, ftmp3);
1215     /* tmp[i] < 17*2^120 + 2^63 */
1216     felem_reduce(ftmp4, tmp);
1217 
1218     x_equal = felem_is_zero(ftmp4);
1219 
1220     /* z_out = ftmp5 * h */
1221     felem_mul(tmp, ftmp5, ftmp4);
1222     felem_reduce(z_out, tmp);
1223 
1224     /* ftmp = z1 * z1z1 */
1225     felem_mul(tmp, ftmp, z1);
1226     felem_reduce(ftmp, tmp);
1227 
1228     /* s2 = tmp = y2 * z1**3 */
1229     felem_mul(tmp, y2, ftmp);
1230     /* tmp[i] < 17*2^120 */
1231 
1232     /* r = ftmp5 = (s2 - s1)*2 */
1233     felem_diff_128_64(tmp, ftmp6);
1234     /* tmp[i] < 17*2^120 + 2^63 */
1235     felem_reduce(ftmp5, tmp);
1236     y_equal = felem_is_zero(ftmp5);
1237     felem_scalar64(ftmp5, 2);
1238     /* ftmp5[i] < 2^61 */
1239 
1240     if (x_equal && y_equal && !z1_is_zero && !z2_is_zero) {
1241         point_double(x3, y3, z3, x1, y1, z1);
1242         return;
1243     }
1244 
1245     /* I = ftmp = (2h)**2 */
1246     felem_assign(ftmp, ftmp4);
1247     felem_scalar64(ftmp, 2);
1248     /* ftmp[i] < 2^61 */
1249     felem_square(tmp, ftmp);
1250     /* tmp[i] < 17*2^122 */
1251     felem_reduce(ftmp, tmp);
1252 
1253     /* J = ftmp2 = h * I */
1254     felem_mul(tmp, ftmp4, ftmp);
1255     felem_reduce(ftmp2, tmp);
1256 
1257     /* V = ftmp4 = U1 * I */
1258     felem_mul(tmp, ftmp3, ftmp);
1259     felem_reduce(ftmp4, tmp);
1260 
1261     /* x_out = r**2 - J - 2V */
1262     felem_square(tmp, ftmp5);
1263     /* tmp[i] < 17*2^122 */
1264     felem_diff_128_64(tmp, ftmp2);
1265     /* tmp[i] < 17*2^122 + 2^63 */
1266     felem_assign(ftmp3, ftmp4);
1267     felem_scalar64(ftmp4, 2);
1268     /* ftmp4[i] < 2^61 */
1269     felem_diff_128_64(tmp, ftmp4);
1270     /* tmp[i] < 17*2^122 + 2^64 */
1271     felem_reduce(x_out, tmp);
1272 
1273     /* y_out = r(V-x_out) - 2 * s1 * J */
1274     felem_diff64(ftmp3, x_out);
1275     /*
1276      * ftmp3[i] < 2^60 + 2^60 = 2^61
1277      */
1278     felem_mul(tmp, ftmp5, ftmp3);
1279     /* tmp[i] < 17*2^122 */
1280     felem_mul(tmp2, ftmp6, ftmp2);
1281     /* tmp2[i] < 17*2^120 */
1282     felem_scalar128(tmp2, 2);
1283     /* tmp2[i] < 17*2^121 */
1284     felem_diff128(tmp, tmp2);
1285         /*-
1286          * tmp[i] < 2^127 - 2^69 + 17*2^122
1287          *        = 2^126 - 2^122 - 2^6 - 2^2 - 1
1288          *        < 2^127
1289          */
1290     felem_reduce(y_out, tmp);
1291 
1292     copy_conditional(x_out, x2, z1_is_zero);
1293     copy_conditional(x_out, x1, z2_is_zero);
1294     copy_conditional(y_out, y2, z1_is_zero);
1295     copy_conditional(y_out, y1, z2_is_zero);
1296     copy_conditional(z_out, z2, z1_is_zero);
1297     copy_conditional(z_out, z1, z2_is_zero);
1298     felem_assign(x3, x_out);
1299     felem_assign(y3, y_out);
1300     felem_assign(z3, z_out);
1301 }
1302 
1303 /*-
1304  * Base point pre computation
1305  * --------------------------
1306  *
1307  * Two different sorts of precomputed tables are used in the following code.
1308  * Each contain various points on the curve, where each point is three field
1309  * elements (x, y, z).
1310  *
1311  * For the base point table, z is usually 1 (0 for the point at infinity).
1312  * This table has 16 elements:
1313  * index | bits    | point
1314  * ------+---------+------------------------------
1315  *     0 | 0 0 0 0 | 0G
1316  *     1 | 0 0 0 1 | 1G
1317  *     2 | 0 0 1 0 | 2^130G
1318  *     3 | 0 0 1 1 | (2^130 + 1)G
1319  *     4 | 0 1 0 0 | 2^260G
1320  *     5 | 0 1 0 1 | (2^260 + 1)G
1321  *     6 | 0 1 1 0 | (2^260 + 2^130)G
1322  *     7 | 0 1 1 1 | (2^260 + 2^130 + 1)G
1323  *     8 | 1 0 0 0 | 2^390G
1324  *     9 | 1 0 0 1 | (2^390 + 1)G
1325  *    10 | 1 0 1 0 | (2^390 + 2^130)G
1326  *    11 | 1 0 1 1 | (2^390 + 2^130 + 1)G
1327  *    12 | 1 1 0 0 | (2^390 + 2^260)G
1328  *    13 | 1 1 0 1 | (2^390 + 2^260 + 1)G
1329  *    14 | 1 1 1 0 | (2^390 + 2^260 + 2^130)G
1330  *    15 | 1 1 1 1 | (2^390 + 2^260 + 2^130 + 1)G
1331  *
1332  * The reason for this is so that we can clock bits into four different
1333  * locations when doing simple scalar multiplies against the base point.
1334  *
1335  * Tables for other points have table[i] = iG for i in 0 .. 16. */
1336 
1337 /* gmul is the table of precomputed base points */
1338 static const felem gmul[16][3] = { {{0, 0, 0, 0, 0, 0, 0, 0, 0},
1339                                     {0, 0, 0, 0, 0, 0, 0, 0, 0},
1340                                     {0, 0, 0, 0, 0, 0, 0, 0, 0}},
1341 {{0x017e7e31c2e5bd66, 0x022cf0615a90a6fe, 0x00127a2ffa8de334,
1342   0x01dfbf9d64a3f877, 0x006b4d3dbaa14b5e, 0x014fed487e0a2bd8,
1343   0x015b4429c6481390, 0x03a73678fb2d988e, 0x00c6858e06b70404},
1344  {0x00be94769fd16650, 0x031c21a89cb09022, 0x039013fad0761353,
1345   0x02657bd099031542, 0x03273e662c97ee72, 0x01e6d11a05ebef45,
1346   0x03d1bd998f544495, 0x03001172297ed0b1, 0x011839296a789a3b},
1347  {1, 0, 0, 0, 0, 0, 0, 0, 0}},
1348 {{0x0373faacbc875bae, 0x00f325023721c671, 0x00f666fd3dbde5ad,
1349   0x01a6932363f88ea7, 0x01fc6d9e13f9c47b, 0x03bcbffc2bbf734e,
1350   0x013ee3c3647f3a92, 0x029409fefe75d07d, 0x00ef9199963d85e5},
1351  {0x011173743ad5b178, 0x02499c7c21bf7d46, 0x035beaeabb8b1a58,
1352   0x00f989c4752ea0a3, 0x0101e1de48a9c1a3, 0x01a20076be28ba6c,
1353   0x02f8052e5eb2de95, 0x01bfe8f82dea117c, 0x0160074d3c36ddb7},
1354  {1, 0, 0, 0, 0, 0, 0, 0, 0}},
1355 {{0x012f3fc373393b3b, 0x03d3d6172f1419fa, 0x02adc943c0b86873,
1356   0x00d475584177952b, 0x012a4d1673750ee2, 0x00512517a0f13b0c,
1357   0x02b184671a7b1734, 0x0315b84236f1a50a, 0x00a4afc472edbdb9},
1358  {0x00152a7077f385c4, 0x03044007d8d1c2ee, 0x0065829d61d52b52,
1359   0x00494ff6b6631d0d, 0x00a11d94d5f06bcf, 0x02d2f89474d9282e,
1360   0x0241c5727c06eeb9, 0x0386928710fbdb9d, 0x01f883f727b0dfbe},
1361  {1, 0, 0, 0, 0, 0, 0, 0, 0}},
1362 {{0x019b0c3c9185544d, 0x006243a37c9d97db, 0x02ee3cbe030a2ad2,
1363   0x00cfdd946bb51e0d, 0x0271c00932606b91, 0x03f817d1ec68c561,
1364   0x03f37009806a369c, 0x03c1f30baf184fd5, 0x01091022d6d2f065},
1365  {0x0292c583514c45ed, 0x0316fca51f9a286c, 0x00300af507c1489a,
1366   0x0295f69008298cf1, 0x02c0ed8274943d7b, 0x016509b9b47a431e,
1367   0x02bc9de9634868ce, 0x005b34929bffcb09, 0x000c1a0121681524},
1368  {1, 0, 0, 0, 0, 0, 0, 0, 0}},
1369 {{0x0286abc0292fb9f2, 0x02665eee9805b3f7, 0x01ed7455f17f26d6,
1370   0x0346355b83175d13, 0x006284944cd0a097, 0x0191895bcdec5e51,
1371   0x02e288370afda7d9, 0x03b22312bfefa67a, 0x01d104d3fc0613fe},
1372  {0x0092421a12f7e47f, 0x0077a83fa373c501, 0x03bd25c5f696bd0d,
1373   0x035c41e4d5459761, 0x01ca0d1742b24f53, 0x00aaab27863a509c,
1374   0x018b6de47df73917, 0x025c0b771705cd01, 0x01fd51d566d760a7},
1375  {1, 0, 0, 0, 0, 0, 0, 0, 0}},
1376 {{0x01dd92ff6b0d1dbd, 0x039c5e2e8f8afa69, 0x0261ed13242c3b27,
1377   0x0382c6e67026e6a0, 0x01d60b10be2089f9, 0x03c15f3dce86723f,
1378   0x03c764a32d2a062d, 0x017307eac0fad056, 0x018207c0b96c5256},
1379  {0x0196a16d60e13154, 0x03e6ce74c0267030, 0x00ddbf2b4e52a5aa,
1380   0x012738241bbf31c8, 0x00ebe8dc04685a28, 0x024c2ad6d380d4a2,
1381   0x035ee062a6e62d0e, 0x0029ed74af7d3a0f, 0x00eef32aec142ebd},
1382  {1, 0, 0, 0, 0, 0, 0, 0, 0}},
1383 {{0x00c31ec398993b39, 0x03a9f45bcda68253, 0x00ac733c24c70890,
1384   0x00872b111401ff01, 0x01d178c23195eafb, 0x03bca2c816b87f74,
1385   0x0261a9af46fbad7a, 0x0324b2a8dd3d28f9, 0x00918121d8f24e23},
1386  {0x032bc8c1ca983cd7, 0x00d869dfb08fc8c6, 0x01693cb61fce1516,
1387   0x012a5ea68f4e88a8, 0x010869cab88d7ae3, 0x009081ad277ceee1,
1388   0x033a77166d064cdc, 0x03955235a1fb3a95, 0x01251a4a9b25b65e},
1389  {1, 0, 0, 0, 0, 0, 0, 0, 0}},
1390 {{0x00148a3a1b27f40b, 0x0123186df1b31fdc, 0x00026e7beaad34ce,
1391   0x01db446ac1d3dbba, 0x0299c1a33437eaec, 0x024540610183cbb7,
1392   0x0173bb0e9ce92e46, 0x02b937e43921214b, 0x01ab0436a9bf01b5},
1393  {0x0383381640d46948, 0x008dacbf0e7f330f, 0x03602122bcc3f318,
1394   0x01ee596b200620d6, 0x03bd0585fda430b3, 0x014aed77fd123a83,
1395   0x005ace749e52f742, 0x0390fe041da2b842, 0x0189a8ceb3299242},
1396  {1, 0, 0, 0, 0, 0, 0, 0, 0}},
1397 {{0x012a19d6b3282473, 0x00c0915918b423ce, 0x023a954eb94405ae,
1398   0x00529f692be26158, 0x0289fa1b6fa4b2aa, 0x0198ae4ceea346ef,
1399   0x0047d8cdfbdedd49, 0x00cc8c8953f0f6b8, 0x001424abbff49203},
1400  {0x0256732a1115a03a, 0x0351bc38665c6733, 0x03f7b950fb4a6447,
1401   0x000afffa94c22155, 0x025763d0a4dab540, 0x000511e92d4fc283,
1402   0x030a7e9eda0ee96c, 0x004c3cd93a28bf0a, 0x017edb3a8719217f},
1403  {1, 0, 0, 0, 0, 0, 0, 0, 0}},
1404 {{0x011de5675a88e673, 0x031d7d0f5e567fbe, 0x0016b2062c970ae5,
1405   0x03f4a2be49d90aa7, 0x03cef0bd13822866, 0x03f0923dcf774a6c,
1406   0x0284bebc4f322f72, 0x016ab2645302bb2c, 0x01793f95dace0e2a},
1407  {0x010646e13527a28f, 0x01ca1babd59dc5e7, 0x01afedfd9a5595df,
1408   0x01f15785212ea6b1, 0x0324e5d64f6ae3f4, 0x02d680f526d00645,
1409   0x0127920fadf627a7, 0x03b383f75df4f684, 0x0089e0057e783b0a},
1410  {1, 0, 0, 0, 0, 0, 0, 0, 0}},
1411 {{0x00f334b9eb3c26c6, 0x0298fdaa98568dce, 0x01c2d24843a82292,
1412   0x020bcb24fa1b0711, 0x02cbdb3d2b1875e6, 0x0014907598f89422,
1413   0x03abe3aa43b26664, 0x02cbf47f720bc168, 0x0133b5e73014b79b},
1414  {0x034aab5dab05779d, 0x00cdc5d71fee9abb, 0x0399f16bd4bd9d30,
1415   0x03582fa592d82647, 0x02be1cdfb775b0e9, 0x0034f7cea32e94cb,
1416   0x0335a7f08f56f286, 0x03b707e9565d1c8b, 0x0015c946ea5b614f},
1417  {1, 0, 0, 0, 0, 0, 0, 0, 0}},
1418 {{0x024676f6cff72255, 0x00d14625cac96378, 0x00532b6008bc3767,
1419   0x01fc16721b985322, 0x023355ea1b091668, 0x029de7afdc0317c3,
1420   0x02fc8a7ca2da037c, 0x02de1217d74a6f30, 0x013f7173175b73bf},
1421  {0x0344913f441490b5, 0x0200f9e272b61eca, 0x0258a246b1dd55d2,
1422   0x03753db9ea496f36, 0x025e02937a09c5ef, 0x030cbd3d14012692,
1423   0x01793a67e70dc72a, 0x03ec1d37048a662e, 0x006550f700c32a8d},
1424  {1, 0, 0, 0, 0, 0, 0, 0, 0}},
1425 {{0x00d3f48a347eba27, 0x008e636649b61bd8, 0x00d3b93716778fb3,
1426   0x004d1915757bd209, 0x019d5311a3da44e0, 0x016d1afcbbe6aade,
1427   0x0241bf5f73265616, 0x0384672e5d50d39b, 0x005009fee522b684},
1428  {0x029b4fab064435fe, 0x018868ee095bbb07, 0x01ea3d6936cc92b8,
1429   0x000608b00f78a2f3, 0x02db911073d1c20f, 0x018205938470100a,
1430   0x01f1e4964cbe6ff2, 0x021a19a29eed4663, 0x01414485f42afa81},
1431  {1, 0, 0, 0, 0, 0, 0, 0, 0}},
1432 {{0x01612b3a17f63e34, 0x03813992885428e6, 0x022b3c215b5a9608,
1433   0x029b4057e19f2fcb, 0x0384059a587af7e6, 0x02d6400ace6fe610,
1434   0x029354d896e8e331, 0x00c047ee6dfba65e, 0x0037720542e9d49d},
1435  {0x02ce9eed7c5e9278, 0x0374ed703e79643b, 0x01316c54c4072006,
1436   0x005aaa09054b2ee8, 0x002824000c840d57, 0x03d4eba24771ed86,
1437   0x0189c50aabc3bdae, 0x0338c01541e15510, 0x00466d56e38eed42},
1438  {1, 0, 0, 0, 0, 0, 0, 0, 0}},
1439 {{0x007efd8330ad8bd6, 0x02465ed48047710b, 0x0034c6606b215e0c,
1440   0x016ae30c53cbf839, 0x01fa17bd37161216, 0x018ead4e61ce8ab9,
1441   0x005482ed5f5dee46, 0x037543755bba1d7f, 0x005e5ac7e70a9d0f},
1442  {0x0117e1bb2fdcb2a2, 0x03deea36249f40c4, 0x028d09b4a6246cb7,
1443   0x03524b8855bcf756, 0x023d7d109d5ceb58, 0x0178e43e3223ef9c,
1444   0x0154536a0c6e966a, 0x037964d1286ee9fe, 0x0199bcd90e125055},
1445  {1, 0, 0, 0, 0, 0, 0, 0, 0}}
1446 };
1447 
1448 /*
1449  * select_point selects the |idx|th point from a precomputation table and
1450  * copies it to out.
1451  */
1452  /* pre_comp below is of the size provided in |size| */
1453 static void select_point(const limb idx, unsigned int size,
1454                          const felem pre_comp[][3], felem out[3])
1455 {
1456     unsigned i, j;
1457     limb *outlimbs = &out[0][0];
1458     memset(outlimbs, 0, 3 * sizeof(felem));
1459 
1460     for (i = 0; i < size; i++) {
1461         const limb *inlimbs = &pre_comp[i][0][0];
1462         limb mask = i ^ idx;
1463         mask |= mask >> 4;
1464         mask |= mask >> 2;
1465         mask |= mask >> 1;
1466         mask &= 1;
1467         mask--;
1468         for (j = 0; j < NLIMBS * 3; j++)
1469             outlimbs[j] |= inlimbs[j] & mask;
1470     }
1471 }
1472 
1473 /* get_bit returns the |i|th bit in |in| */
1474 static char get_bit(const felem_bytearray in, int i)
1475 {
1476     if (i < 0)
1477         return 0;
1478     return (in[i >> 3] >> (i & 7)) & 1;
1479 }
1480 
1481 /*
1482  * Interleaved point multiplication using precomputed point multiples: The
1483  * small point multiples 0*P, 1*P, ..., 16*P are in pre_comp[], the scalars
1484  * in scalars[]. If g_scalar is non-NULL, we also add this multiple of the
1485  * generator, using certain (large) precomputed multiples in g_pre_comp.
1486  * Output point (X, Y, Z) is stored in x_out, y_out, z_out
1487  */
1488 static void batch_mul(felem x_out, felem y_out, felem z_out,
1489                       const felem_bytearray scalars[],
1490                       const unsigned num_points, const u8 *g_scalar,
1491                       const int mixed, const felem pre_comp[][17][3],
1492                       const felem g_pre_comp[16][3])
1493 {
1494     int i, skip;
1495     unsigned num, gen_mul = (g_scalar != NULL);
1496     felem nq[3], tmp[4];
1497     limb bits;
1498     u8 sign, digit;
1499 
1500     /* set nq to the point at infinity */
1501     memset(nq, 0, 3 * sizeof(felem));
1502 
1503     /*
1504      * Loop over all scalars msb-to-lsb, interleaving additions of multiples
1505      * of the generator (last quarter of rounds) and additions of other
1506      * points multiples (every 5th round).
1507      */
1508     skip = 1;                   /* save two point operations in the first
1509                                  * round */
1510     for (i = (num_points ? 520 : 130); i >= 0; --i) {
1511         /* double */
1512         if (!skip)
1513             point_double(nq[0], nq[1], nq[2], nq[0], nq[1], nq[2]);
1514 
1515         /* add multiples of the generator */
1516         if (gen_mul && (i <= 130)) {
1517             bits = get_bit(g_scalar, i + 390) << 3;
1518             if (i < 130) {
1519                 bits |= get_bit(g_scalar, i + 260) << 2;
1520                 bits |= get_bit(g_scalar, i + 130) << 1;
1521                 bits |= get_bit(g_scalar, i);
1522             }
1523             /* select the point to add, in constant time */
1524             select_point(bits, 16, g_pre_comp, tmp);
1525             if (!skip) {
1526                 /* The 1 argument below is for "mixed" */
1527                 point_add(nq[0], nq[1], nq[2],
1528                           nq[0], nq[1], nq[2], 1, tmp[0], tmp[1], tmp[2]);
1529             } else {
1530                 memcpy(nq, tmp, 3 * sizeof(felem));
1531                 skip = 0;
1532             }
1533         }
1534 
1535         /* do other additions every 5 doublings */
1536         if (num_points && (i % 5 == 0)) {
1537             /* loop over all scalars */
1538             for (num = 0; num < num_points; ++num) {
1539                 bits = get_bit(scalars[num], i + 4) << 5;
1540                 bits |= get_bit(scalars[num], i + 3) << 4;
1541                 bits |= get_bit(scalars[num], i + 2) << 3;
1542                 bits |= get_bit(scalars[num], i + 1) << 2;
1543                 bits |= get_bit(scalars[num], i) << 1;
1544                 bits |= get_bit(scalars[num], i - 1);
1545                 ec_GFp_nistp_recode_scalar_bits(&sign, &digit, bits);
1546 
1547                 /*
1548                  * select the point to add or subtract, in constant time
1549                  */
1550                 select_point(digit, 17, pre_comp[num], tmp);
1551                 felem_neg(tmp[3], tmp[1]); /* (X, -Y, Z) is the negative
1552                                             * point */
1553                 copy_conditional(tmp[1], tmp[3], (-(limb) sign));
1554 
1555                 if (!skip) {
1556                     point_add(nq[0], nq[1], nq[2],
1557                               nq[0], nq[1], nq[2],
1558                               mixed, tmp[0], tmp[1], tmp[2]);
1559                 } else {
1560                     memcpy(nq, tmp, 3 * sizeof(felem));
1561                     skip = 0;
1562                 }
1563             }
1564         }
1565     }
1566     felem_assign(x_out, nq[0]);
1567     felem_assign(y_out, nq[1]);
1568     felem_assign(z_out, nq[2]);
1569 }
1570 
1571 /* Precomputation for the group generator. */
1572 typedef struct {
1573     felem g_pre_comp[16][3];
1574     int references;
1575 } NISTP521_PRE_COMP;
1576 
1577 const EC_METHOD *EC_GFp_nistp521_method(void)
1578 {
1579     static const EC_METHOD ret = {
1580         EC_FLAGS_DEFAULT_OCT,
1581         NID_X9_62_prime_field,
1582         ec_GFp_nistp521_group_init,
1583         ec_GFp_simple_group_finish,
1584         ec_GFp_simple_group_clear_finish,
1585         ec_GFp_nist_group_copy,
1586         ec_GFp_nistp521_group_set_curve,
1587         ec_GFp_simple_group_get_curve,
1588         ec_GFp_simple_group_get_degree,
1589         ec_GFp_simple_group_check_discriminant,
1590         ec_GFp_simple_point_init,
1591         ec_GFp_simple_point_finish,
1592         ec_GFp_simple_point_clear_finish,
1593         ec_GFp_simple_point_copy,
1594         ec_GFp_simple_point_set_to_infinity,
1595         ec_GFp_simple_set_Jprojective_coordinates_GFp,
1596         ec_GFp_simple_get_Jprojective_coordinates_GFp,
1597         ec_GFp_simple_point_set_affine_coordinates,
1598         ec_GFp_nistp521_point_get_affine_coordinates,
1599         0 /* point_set_compressed_coordinates */ ,
1600         0 /* point2oct */ ,
1601         0 /* oct2point */ ,
1602         ec_GFp_simple_add,
1603         ec_GFp_simple_dbl,
1604         ec_GFp_simple_invert,
1605         ec_GFp_simple_is_at_infinity,
1606         ec_GFp_simple_is_on_curve,
1607         ec_GFp_simple_cmp,
1608         ec_GFp_simple_make_affine,
1609         ec_GFp_simple_points_make_affine,
1610         ec_GFp_nistp521_points_mul,
1611         ec_GFp_nistp521_precompute_mult,
1612         ec_GFp_nistp521_have_precompute_mult,
1613         ec_GFp_nist_field_mul,
1614         ec_GFp_nist_field_sqr,
1615         0 /* field_div */ ,
1616         0 /* field_encode */ ,
1617         0 /* field_decode */ ,
1618         0                       /* field_set_to_one */
1619     };
1620 
1621     return &ret;
1622 }
1623 
1624 /******************************************************************************/
1625 /*
1626  * FUNCTIONS TO MANAGE PRECOMPUTATION
1627  */
1628 
1629 static NISTP521_PRE_COMP *nistp521_pre_comp_new()
1630 {
1631     NISTP521_PRE_COMP *ret = NULL;
1632     ret = (NISTP521_PRE_COMP *) OPENSSL_malloc(sizeof(NISTP521_PRE_COMP));
1633     if (!ret) {
1634         ECerr(EC_F_NISTP521_PRE_COMP_NEW, ERR_R_MALLOC_FAILURE);
1635         return ret;
1636     }
1637     memset(ret->g_pre_comp, 0, sizeof(ret->g_pre_comp));
1638     ret->references = 1;
1639     return ret;
1640 }
1641 
1642 static void *nistp521_pre_comp_dup(void *src_)
1643 {
1644     NISTP521_PRE_COMP *src = src_;
1645 
1646     /* no need to actually copy, these objects never change! */
1647     CRYPTO_add(&src->references, 1, CRYPTO_LOCK_EC_PRE_COMP);
1648 
1649     return src_;
1650 }
1651 
1652 static void nistp521_pre_comp_free(void *pre_)
1653 {
1654     int i;
1655     NISTP521_PRE_COMP *pre = pre_;
1656 
1657     if (!pre)
1658         return;
1659 
1660     i = CRYPTO_add(&pre->references, -1, CRYPTO_LOCK_EC_PRE_COMP);
1661     if (i > 0)
1662         return;
1663 
1664     OPENSSL_free(pre);
1665 }
1666 
1667 static void nistp521_pre_comp_clear_free(void *pre_)
1668 {
1669     int i;
1670     NISTP521_PRE_COMP *pre = pre_;
1671 
1672     if (!pre)
1673         return;
1674 
1675     i = CRYPTO_add(&pre->references, -1, CRYPTO_LOCK_EC_PRE_COMP);
1676     if (i > 0)
1677         return;
1678 
1679     OPENSSL_cleanse(pre, sizeof(*pre));
1680     OPENSSL_free(pre);
1681 }
1682 
1683 /******************************************************************************/
1684 /*
1685  * OPENSSL EC_METHOD FUNCTIONS
1686  */
1687 
1688 int ec_GFp_nistp521_group_init(EC_GROUP *group)
1689 {
1690     int ret;
1691     ret = ec_GFp_simple_group_init(group);
1692     group->a_is_minus3 = 1;
1693     return ret;
1694 }
1695 
1696 int ec_GFp_nistp521_group_set_curve(EC_GROUP *group, const BIGNUM *p,
1697                                     const BIGNUM *a, const BIGNUM *b,
1698                                     BN_CTX *ctx)
1699 {
1700     int ret = 0;
1701     BN_CTX *new_ctx = NULL;
1702     BIGNUM *curve_p, *curve_a, *curve_b;
1703 
1704     if (ctx == NULL)
1705         if ((ctx = new_ctx = BN_CTX_new()) == NULL)
1706             return 0;
1707     BN_CTX_start(ctx);
1708     if (((curve_p = BN_CTX_get(ctx)) == NULL) ||
1709         ((curve_a = BN_CTX_get(ctx)) == NULL) ||
1710         ((curve_b = BN_CTX_get(ctx)) == NULL))
1711         goto err;
1712     BN_bin2bn(nistp521_curve_params[0], sizeof(felem_bytearray), curve_p);
1713     BN_bin2bn(nistp521_curve_params[1], sizeof(felem_bytearray), curve_a);
1714     BN_bin2bn(nistp521_curve_params[2], sizeof(felem_bytearray), curve_b);
1715     if ((BN_cmp(curve_p, p)) || (BN_cmp(curve_a, a)) || (BN_cmp(curve_b, b))) {
1716         ECerr(EC_F_EC_GFP_NISTP521_GROUP_SET_CURVE,
1717               EC_R_WRONG_CURVE_PARAMETERS);
1718         goto err;
1719     }
1720     group->field_mod_func = BN_nist_mod_521;
1721     ret = ec_GFp_simple_group_set_curve(group, p, a, b, ctx);
1722  err:
1723     BN_CTX_end(ctx);
1724     if (new_ctx != NULL)
1725         BN_CTX_free(new_ctx);
1726     return ret;
1727 }
1728 
1729 /*
1730  * Takes the Jacobian coordinates (X, Y, Z) of a point and returns (X', Y') =
1731  * (X/Z^2, Y/Z^3)
1732  */
1733 int ec_GFp_nistp521_point_get_affine_coordinates(const EC_GROUP *group,
1734                                                  const EC_POINT *point,
1735                                                  BIGNUM *x, BIGNUM *y,
1736                                                  BN_CTX *ctx)
1737 {
1738     felem z1, z2, x_in, y_in, x_out, y_out;
1739     largefelem tmp;
1740 
1741     if (EC_POINT_is_at_infinity(group, point)) {
1742         ECerr(EC_F_EC_GFP_NISTP521_POINT_GET_AFFINE_COORDINATES,
1743               EC_R_POINT_AT_INFINITY);
1744         return 0;
1745     }
1746     if ((!BN_to_felem(x_in, &point->X)) || (!BN_to_felem(y_in, &point->Y)) ||
1747         (!BN_to_felem(z1, &point->Z)))
1748         return 0;
1749     felem_inv(z2, z1);
1750     felem_square(tmp, z2);
1751     felem_reduce(z1, tmp);
1752     felem_mul(tmp, x_in, z1);
1753     felem_reduce(x_in, tmp);
1754     felem_contract(x_out, x_in);
1755     if (x != NULL) {
1756         if (!felem_to_BN(x, x_out)) {
1757             ECerr(EC_F_EC_GFP_NISTP521_POINT_GET_AFFINE_COORDINATES,
1758                   ERR_R_BN_LIB);
1759             return 0;
1760         }
1761     }
1762     felem_mul(tmp, z1, z2);
1763     felem_reduce(z1, tmp);
1764     felem_mul(tmp, y_in, z1);
1765     felem_reduce(y_in, tmp);
1766     felem_contract(y_out, y_in);
1767     if (y != NULL) {
1768         if (!felem_to_BN(y, y_out)) {
1769             ECerr(EC_F_EC_GFP_NISTP521_POINT_GET_AFFINE_COORDINATES,
1770                   ERR_R_BN_LIB);
1771             return 0;
1772         }
1773     }
1774     return 1;
1775 }
1776 
1777 /* points below is of size |num|, and tmp_felems is of size |num+1/ */
1778 static void make_points_affine(size_t num, felem points[][3],
1779                                felem tmp_felems[])
1780 {
1781     /*
1782      * Runs in constant time, unless an input is the point at infinity (which
1783      * normally shouldn't happen).
1784      */
1785     ec_GFp_nistp_points_make_affine_internal(num,
1786                                              points,
1787                                              sizeof(felem),
1788                                              tmp_felems,
1789                                              (void (*)(void *))felem_one,
1790                                              (int (*)(const void *))
1791                                              felem_is_zero_int,
1792                                              (void (*)(void *, const void *))
1793                                              felem_assign,
1794                                              (void (*)(void *, const void *))
1795                                              felem_square_reduce, (void (*)
1796                                                                    (void *,
1797                                                                     const void
1798                                                                     *,
1799                                                                     const void
1800                                                                     *))
1801                                              felem_mul_reduce,
1802                                              (void (*)(void *, const void *))
1803                                              felem_inv,
1804                                              (void (*)(void *, const void *))
1805                                              felem_contract);
1806 }
1807 
1808 /*
1809  * Computes scalar*generator + \sum scalars[i]*points[i], ignoring NULL
1810  * values Result is stored in r (r can equal one of the inputs).
1811  */
1812 int ec_GFp_nistp521_points_mul(const EC_GROUP *group, EC_POINT *r,
1813                                const BIGNUM *scalar, size_t num,
1814                                const EC_POINT *points[],
1815                                const BIGNUM *scalars[], BN_CTX *ctx)
1816 {
1817     int ret = 0;
1818     int j;
1819     int mixed = 0;
1820     BN_CTX *new_ctx = NULL;
1821     BIGNUM *x, *y, *z, *tmp_scalar;
1822     felem_bytearray g_secret;
1823     felem_bytearray *secrets = NULL;
1824     felem(*pre_comp)[17][3] = NULL;
1825     felem *tmp_felems = NULL;
1826     felem_bytearray tmp;
1827     unsigned i, num_bytes;
1828     int have_pre_comp = 0;
1829     size_t num_points = num;
1830     felem x_in, y_in, z_in, x_out, y_out, z_out;
1831     NISTP521_PRE_COMP *pre = NULL;
1832     felem(*g_pre_comp)[3] = NULL;
1833     EC_POINT *generator = NULL;
1834     const EC_POINT *p = NULL;
1835     const BIGNUM *p_scalar = NULL;
1836 
1837     if (ctx == NULL)
1838         if ((ctx = new_ctx = BN_CTX_new()) == NULL)
1839             return 0;
1840     BN_CTX_start(ctx);
1841     if (((x = BN_CTX_get(ctx)) == NULL) ||
1842         ((y = BN_CTX_get(ctx)) == NULL) ||
1843         ((z = BN_CTX_get(ctx)) == NULL) ||
1844         ((tmp_scalar = BN_CTX_get(ctx)) == NULL))
1845         goto err;
1846 
1847     if (scalar != NULL) {
1848         pre = EC_EX_DATA_get_data(group->extra_data,
1849                                   nistp521_pre_comp_dup,
1850                                   nistp521_pre_comp_free,
1851                                   nistp521_pre_comp_clear_free);
1852         if (pre)
1853             /* we have precomputation, try to use it */
1854             g_pre_comp = &pre->g_pre_comp[0];
1855         else
1856             /* try to use the standard precomputation */
1857             g_pre_comp = (felem(*)[3]) gmul;
1858         generator = EC_POINT_new(group);
1859         if (generator == NULL)
1860             goto err;
1861         /* get the generator from precomputation */
1862         if (!felem_to_BN(x, g_pre_comp[1][0]) ||
1863             !felem_to_BN(y, g_pre_comp[1][1]) ||
1864             !felem_to_BN(z, g_pre_comp[1][2])) {
1865             ECerr(EC_F_EC_GFP_NISTP521_POINTS_MUL, ERR_R_BN_LIB);
1866             goto err;
1867         }
1868         if (!EC_POINT_set_Jprojective_coordinates_GFp(group,
1869                                                       generator, x, y, z,
1870                                                       ctx))
1871             goto err;
1872         if (0 == EC_POINT_cmp(group, generator, group->generator, ctx))
1873             /* precomputation matches generator */
1874             have_pre_comp = 1;
1875         else
1876             /*
1877              * we don't have valid precomputation: treat the generator as a
1878              * random point
1879              */
1880             num_points++;
1881     }
1882 
1883     if (num_points > 0) {
1884         if (num_points >= 2) {
1885             /*
1886              * unless we precompute multiples for just one point, converting
1887              * those into affine form is time well spent
1888              */
1889             mixed = 1;
1890         }
1891         secrets = OPENSSL_malloc(num_points * sizeof(felem_bytearray));
1892         pre_comp = OPENSSL_malloc(num_points * 17 * 3 * sizeof(felem));
1893         if (mixed)
1894             tmp_felems =
1895                 OPENSSL_malloc((num_points * 17 + 1) * sizeof(felem));
1896         if ((secrets == NULL) || (pre_comp == NULL)
1897             || (mixed && (tmp_felems == NULL))) {
1898             ECerr(EC_F_EC_GFP_NISTP521_POINTS_MUL, ERR_R_MALLOC_FAILURE);
1899             goto err;
1900         }
1901 
1902         /*
1903          * we treat NULL scalars as 0, and NULL points as points at infinity,
1904          * i.e., they contribute nothing to the linear combination
1905          */
1906         memset(secrets, 0, num_points * sizeof(felem_bytearray));
1907         memset(pre_comp, 0, num_points * 17 * 3 * sizeof(felem));
1908         for (i = 0; i < num_points; ++i) {
1909             if (i == num)
1910                 /*
1911                  * we didn't have a valid precomputation, so we pick the
1912                  * generator
1913                  */
1914             {
1915                 p = EC_GROUP_get0_generator(group);
1916                 p_scalar = scalar;
1917             } else
1918                 /* the i^th point */
1919             {
1920                 p = points[i];
1921                 p_scalar = scalars[i];
1922             }
1923             if ((p_scalar != NULL) && (p != NULL)) {
1924                 /* reduce scalar to 0 <= scalar < 2^521 */
1925                 if ((BN_num_bits(p_scalar) > 521)
1926                     || (BN_is_negative(p_scalar))) {
1927                     /*
1928                      * this is an unusual input, and we don't guarantee
1929                      * constant-timeness
1930                      */
1931                     if (!BN_nnmod(tmp_scalar, p_scalar, &group->order, ctx)) {
1932                         ECerr(EC_F_EC_GFP_NISTP521_POINTS_MUL, ERR_R_BN_LIB);
1933                         goto err;
1934                     }
1935                     num_bytes = BN_bn2bin(tmp_scalar, tmp);
1936                 } else
1937                     num_bytes = BN_bn2bin(p_scalar, tmp);
1938                 flip_endian(secrets[i], tmp, num_bytes);
1939                 /* precompute multiples */
1940                 if ((!BN_to_felem(x_out, &p->X)) ||
1941                     (!BN_to_felem(y_out, &p->Y)) ||
1942                     (!BN_to_felem(z_out, &p->Z)))
1943                     goto err;
1944                 memcpy(pre_comp[i][1][0], x_out, sizeof(felem));
1945                 memcpy(pre_comp[i][1][1], y_out, sizeof(felem));
1946                 memcpy(pre_comp[i][1][2], z_out, sizeof(felem));
1947                 for (j = 2; j <= 16; ++j) {
1948                     if (j & 1) {
1949                         point_add(pre_comp[i][j][0], pre_comp[i][j][1],
1950                                   pre_comp[i][j][2], pre_comp[i][1][0],
1951                                   pre_comp[i][1][1], pre_comp[i][1][2], 0,
1952                                   pre_comp[i][j - 1][0],
1953                                   pre_comp[i][j - 1][1],
1954                                   pre_comp[i][j - 1][2]);
1955                     } else {
1956                         point_double(pre_comp[i][j][0], pre_comp[i][j][1],
1957                                      pre_comp[i][j][2], pre_comp[i][j / 2][0],
1958                                      pre_comp[i][j / 2][1],
1959                                      pre_comp[i][j / 2][2]);
1960                     }
1961                 }
1962             }
1963         }
1964         if (mixed)
1965             make_points_affine(num_points * 17, pre_comp[0], tmp_felems);
1966     }
1967 
1968     /* the scalar for the generator */
1969     if ((scalar != NULL) && (have_pre_comp)) {
1970         memset(g_secret, 0, sizeof(g_secret));
1971         /* reduce scalar to 0 <= scalar < 2^521 */
1972         if ((BN_num_bits(scalar) > 521) || (BN_is_negative(scalar))) {
1973             /*
1974              * this is an unusual input, and we don't guarantee
1975              * constant-timeness
1976              */
1977             if (!BN_nnmod(tmp_scalar, scalar, &group->order, ctx)) {
1978                 ECerr(EC_F_EC_GFP_NISTP521_POINTS_MUL, ERR_R_BN_LIB);
1979                 goto err;
1980             }
1981             num_bytes = BN_bn2bin(tmp_scalar, tmp);
1982         } else
1983             num_bytes = BN_bn2bin(scalar, tmp);
1984         flip_endian(g_secret, tmp, num_bytes);
1985         /* do the multiplication with generator precomputation */
1986         batch_mul(x_out, y_out, z_out,
1987                   (const felem_bytearray(*))secrets, num_points,
1988                   g_secret,
1989                   mixed, (const felem(*)[17][3])pre_comp,
1990                   (const felem(*)[3])g_pre_comp);
1991     } else
1992         /* do the multiplication without generator precomputation */
1993         batch_mul(x_out, y_out, z_out,
1994                   (const felem_bytearray(*))secrets, num_points,
1995                   NULL, mixed, (const felem(*)[17][3])pre_comp, NULL);
1996     /* reduce the output to its unique minimal representation */
1997     felem_contract(x_in, x_out);
1998     felem_contract(y_in, y_out);
1999     felem_contract(z_in, z_out);
2000     if ((!felem_to_BN(x, x_in)) || (!felem_to_BN(y, y_in)) ||
2001         (!felem_to_BN(z, z_in))) {
2002         ECerr(EC_F_EC_GFP_NISTP521_POINTS_MUL, ERR_R_BN_LIB);
2003         goto err;
2004     }
2005     ret = EC_POINT_set_Jprojective_coordinates_GFp(group, r, x, y, z, ctx);
2006 
2007  err:
2008     BN_CTX_end(ctx);
2009     if (generator != NULL)
2010         EC_POINT_free(generator);
2011     if (new_ctx != NULL)
2012         BN_CTX_free(new_ctx);
2013     if (secrets != NULL)
2014         OPENSSL_free(secrets);
2015     if (pre_comp != NULL)
2016         OPENSSL_free(pre_comp);
2017     if (tmp_felems != NULL)
2018         OPENSSL_free(tmp_felems);
2019     return ret;
2020 }
2021 
2022 int ec_GFp_nistp521_precompute_mult(EC_GROUP *group, BN_CTX *ctx)
2023 {
2024     int ret = 0;
2025     NISTP521_PRE_COMP *pre = NULL;
2026     int i, j;
2027     BN_CTX *new_ctx = NULL;
2028     BIGNUM *x, *y;
2029     EC_POINT *generator = NULL;
2030     felem tmp_felems[16];
2031 
2032     /* throw away old precomputation */
2033     EC_EX_DATA_free_data(&group->extra_data, nistp521_pre_comp_dup,
2034                          nistp521_pre_comp_free,
2035                          nistp521_pre_comp_clear_free);
2036     if (ctx == NULL)
2037         if ((ctx = new_ctx = BN_CTX_new()) == NULL)
2038             return 0;
2039     BN_CTX_start(ctx);
2040     if (((x = BN_CTX_get(ctx)) == NULL) || ((y = BN_CTX_get(ctx)) == NULL))
2041         goto err;
2042     /* get the generator */
2043     if (group->generator == NULL)
2044         goto err;
2045     generator = EC_POINT_new(group);
2046     if (generator == NULL)
2047         goto err;
2048     BN_bin2bn(nistp521_curve_params[3], sizeof(felem_bytearray), x);
2049     BN_bin2bn(nistp521_curve_params[4], sizeof(felem_bytearray), y);
2050     if (!EC_POINT_set_affine_coordinates_GFp(group, generator, x, y, ctx))
2051         goto err;
2052     if ((pre = nistp521_pre_comp_new()) == NULL)
2053         goto err;
2054     /*
2055      * if the generator is the standard one, use built-in precomputation
2056      */
2057     if (0 == EC_POINT_cmp(group, generator, group->generator, ctx)) {
2058         memcpy(pre->g_pre_comp, gmul, sizeof(pre->g_pre_comp));
2059         ret = 1;
2060         goto err;
2061     }
2062     if ((!BN_to_felem(pre->g_pre_comp[1][0], &group->generator->X)) ||
2063         (!BN_to_felem(pre->g_pre_comp[1][1], &group->generator->Y)) ||
2064         (!BN_to_felem(pre->g_pre_comp[1][2], &group->generator->Z)))
2065         goto err;
2066     /* compute 2^130*G, 2^260*G, 2^390*G */
2067     for (i = 1; i <= 4; i <<= 1) {
2068         point_double(pre->g_pre_comp[2 * i][0], pre->g_pre_comp[2 * i][1],
2069                      pre->g_pre_comp[2 * i][2], pre->g_pre_comp[i][0],
2070                      pre->g_pre_comp[i][1], pre->g_pre_comp[i][2]);
2071         for (j = 0; j < 129; ++j) {
2072             point_double(pre->g_pre_comp[2 * i][0],
2073                          pre->g_pre_comp[2 * i][1],
2074                          pre->g_pre_comp[2 * i][2],
2075                          pre->g_pre_comp[2 * i][0],
2076                          pre->g_pre_comp[2 * i][1],
2077                          pre->g_pre_comp[2 * i][2]);
2078         }
2079     }
2080     /* g_pre_comp[0] is the point at infinity */
2081     memset(pre->g_pre_comp[0], 0, sizeof(pre->g_pre_comp[0]));
2082     /* the remaining multiples */
2083     /* 2^130*G + 2^260*G */
2084     point_add(pre->g_pre_comp[6][0], pre->g_pre_comp[6][1],
2085               pre->g_pre_comp[6][2], pre->g_pre_comp[4][0],
2086               pre->g_pre_comp[4][1], pre->g_pre_comp[4][2],
2087               0, pre->g_pre_comp[2][0], pre->g_pre_comp[2][1],
2088               pre->g_pre_comp[2][2]);
2089     /* 2^130*G + 2^390*G */
2090     point_add(pre->g_pre_comp[10][0], pre->g_pre_comp[10][1],
2091               pre->g_pre_comp[10][2], pre->g_pre_comp[8][0],
2092               pre->g_pre_comp[8][1], pre->g_pre_comp[8][2],
2093               0, pre->g_pre_comp[2][0], pre->g_pre_comp[2][1],
2094               pre->g_pre_comp[2][2]);
2095     /* 2^260*G + 2^390*G */
2096     point_add(pre->g_pre_comp[12][0], pre->g_pre_comp[12][1],
2097               pre->g_pre_comp[12][2], pre->g_pre_comp[8][0],
2098               pre->g_pre_comp[8][1], pre->g_pre_comp[8][2],
2099               0, pre->g_pre_comp[4][0], pre->g_pre_comp[4][1],
2100               pre->g_pre_comp[4][2]);
2101     /* 2^130*G + 2^260*G + 2^390*G */
2102     point_add(pre->g_pre_comp[14][0], pre->g_pre_comp[14][1],
2103               pre->g_pre_comp[14][2], pre->g_pre_comp[12][0],
2104               pre->g_pre_comp[12][1], pre->g_pre_comp[12][2],
2105               0, pre->g_pre_comp[2][0], pre->g_pre_comp[2][1],
2106               pre->g_pre_comp[2][2]);
2107     for (i = 1; i < 8; ++i) {
2108         /* odd multiples: add G */
2109         point_add(pre->g_pre_comp[2 * i + 1][0],
2110                   pre->g_pre_comp[2 * i + 1][1],
2111                   pre->g_pre_comp[2 * i + 1][2], pre->g_pre_comp[2 * i][0],
2112                   pre->g_pre_comp[2 * i][1], pre->g_pre_comp[2 * i][2], 0,
2113                   pre->g_pre_comp[1][0], pre->g_pre_comp[1][1],
2114                   pre->g_pre_comp[1][2]);
2115     }
2116     make_points_affine(15, &(pre->g_pre_comp[1]), tmp_felems);
2117 
2118     if (!EC_EX_DATA_set_data(&group->extra_data, pre, nistp521_pre_comp_dup,
2119                              nistp521_pre_comp_free,
2120                              nistp521_pre_comp_clear_free))
2121         goto err;
2122     ret = 1;
2123     pre = NULL;
2124  err:
2125     BN_CTX_end(ctx);
2126     if (generator != NULL)
2127         EC_POINT_free(generator);
2128     if (new_ctx != NULL)
2129         BN_CTX_free(new_ctx);
2130     if (pre)
2131         nistp521_pre_comp_free(pre);
2132     return ret;
2133 }
2134 
2135 int ec_GFp_nistp521_have_precompute_mult(const EC_GROUP *group)
2136 {
2137     if (EC_EX_DATA_get_data(group->extra_data, nistp521_pre_comp_dup,
2138                             nistp521_pre_comp_free,
2139                             nistp521_pre_comp_clear_free)
2140         != NULL)
2141         return 1;
2142     else
2143         return 0;
2144 }
2145 
2146 #else
2147 static void *dummy = &dummy;
2148 #endif
2149