1 /**********************************************************************
2 * Copyright (c) 2013, 2014 Pieter Wuille *
3 * Distributed under the MIT software license, see the accompanying *
4 * file COPYING or http://www.opensource.org/licenses/mit-license.php.*
5 **********************************************************************/
7 #ifndef SECP256K1_FIELD_REPR_IMPL_H
8 #define SECP256K1_FIELD_REPR_IMPL_H
10 #if defined HAVE_CONFIG_H
11 #include "libsecp256k1-config.h"
18 #if defined(USE_ASM_X86_64)
19 #include "field_5x52_asm_impl.h"
21 #include "field_5x52_int128_impl.h"
24 /** Implements arithmetic modulo FFFFFFFF FFFFFFFF FFFFFFFF FFFFFFFF FFFFFFFF FFFFFFFF FFFFFFFE FFFFFC2F,
25 * represented as 5 uint64_t's in base 2^52. The values are allowed to contain >52 each. In particular,
26 * each FieldElem has a 'magnitude' associated with it. Internally, a magnitude M means each element
27 * is at most M*(2^53-1), except the most significant one, which is limited to M*(2^49-1). All operations
28 * accept any input with magnitude at most M, and have different rules for propagating magnitude to their
33 static void secp256k1_fe_verify(const secp256k1_fe *a) {
34 const uint64_t *d = a->n;
35 int m = a->normalized ? 1 : 2 * a->magnitude, r = 1;
36 /* secp256k1 'p' value defined in "Standards for Efficient Cryptography" (SEC2) 2.7.1. */
37 r &= (d[0] <= 0xFFFFFFFFFFFFFULL * m);
38 r &= (d[1] <= 0xFFFFFFFFFFFFFULL * m);
39 r &= (d[2] <= 0xFFFFFFFFFFFFFULL * m);
40 r &= (d[3] <= 0xFFFFFFFFFFFFFULL * m);
41 r &= (d[4] <= 0x0FFFFFFFFFFFFULL * m);
42 r &= (a->magnitude >= 0);
43 r &= (a->magnitude <= 2048);
45 r &= (a->magnitude <= 1);
46 if (r && (d[4] == 0x0FFFFFFFFFFFFULL) && ((d[3] & d[2] & d[1]) == 0xFFFFFFFFFFFFFULL)) {
47 r &= (d[0] < 0xFFFFEFFFFFC2FULL);
54 static void secp256k1_fe_normalize(secp256k1_fe *r) {
55 uint64_t t0 = r->n[0], t1 = r->n[1], t2 = r->n[2], t3 = r->n[3], t4 = r->n[4];
57 /* Reduce t4 at the start so there will be at most a single carry from the first pass */
59 uint64_t x = t4 >> 48; t4 &= 0x0FFFFFFFFFFFFULL;
61 /* The first pass ensures the magnitude is 1, ... */
62 t0 += x * 0x1000003D1ULL;
63 t1 += (t0 >> 52); t0 &= 0xFFFFFFFFFFFFFULL;
64 t2 += (t1 >> 52); t1 &= 0xFFFFFFFFFFFFFULL; m = t1;
65 t3 += (t2 >> 52); t2 &= 0xFFFFFFFFFFFFFULL; m &= t2;
66 t4 += (t3 >> 52); t3 &= 0xFFFFFFFFFFFFFULL; m &= t3;
68 /* ... except for a possible carry at bit 48 of t4 (i.e. bit 256 of the field element) */
69 VERIFY_CHECK(t4 >> 49 == 0);
71 /* At most a single final reduction is needed; check if the value is >= the field characteristic */
72 x = (t4 >> 48) | ((t4 == 0x0FFFFFFFFFFFFULL) & (m == 0xFFFFFFFFFFFFFULL)
73 & (t0 >= 0xFFFFEFFFFFC2FULL));
75 /* Apply the final reduction (for constant-time behaviour, we do it always) */
76 t0 += x * 0x1000003D1ULL;
77 t1 += (t0 >> 52); t0 &= 0xFFFFFFFFFFFFFULL;
78 t2 += (t1 >> 52); t1 &= 0xFFFFFFFFFFFFFULL;
79 t3 += (t2 >> 52); t2 &= 0xFFFFFFFFFFFFFULL;
80 t4 += (t3 >> 52); t3 &= 0xFFFFFFFFFFFFFULL;
82 /* If t4 didn't carry to bit 48 already, then it should have after any final reduction */
83 VERIFY_CHECK(t4 >> 48 == x);
85 /* Mask off the possible multiple of 2^256 from the final reduction */
86 t4 &= 0x0FFFFFFFFFFFFULL;
88 r->n[0] = t0; r->n[1] = t1; r->n[2] = t2; r->n[3] = t3; r->n[4] = t4;
93 secp256k1_fe_verify(r);
97 static void secp256k1_fe_normalize_weak(secp256k1_fe *r) {
98 uint64_t t0 = r->n[0], t1 = r->n[1], t2 = r->n[2], t3 = r->n[3], t4 = r->n[4];
100 /* Reduce t4 at the start so there will be at most a single carry from the first pass */
101 uint64_t x = t4 >> 48; t4 &= 0x0FFFFFFFFFFFFULL;
103 /* The first pass ensures the magnitude is 1, ... */
104 t0 += x * 0x1000003D1ULL;
105 t1 += (t0 >> 52); t0 &= 0xFFFFFFFFFFFFFULL;
106 t2 += (t1 >> 52); t1 &= 0xFFFFFFFFFFFFFULL;
107 t3 += (t2 >> 52); t2 &= 0xFFFFFFFFFFFFFULL;
108 t4 += (t3 >> 52); t3 &= 0xFFFFFFFFFFFFFULL;
110 /* ... except for a possible carry at bit 48 of t4 (i.e. bit 256 of the field element) */
111 VERIFY_CHECK(t4 >> 49 == 0);
113 r->n[0] = t0; r->n[1] = t1; r->n[2] = t2; r->n[3] = t3; r->n[4] = t4;
117 secp256k1_fe_verify(r);
121 static void secp256k1_fe_normalize_var(secp256k1_fe *r) {
122 uint64_t t0 = r->n[0], t1 = r->n[1], t2 = r->n[2], t3 = r->n[3], t4 = r->n[4];
124 /* Reduce t4 at the start so there will be at most a single carry from the first pass */
126 uint64_t x = t4 >> 48; t4 &= 0x0FFFFFFFFFFFFULL;
128 /* The first pass ensures the magnitude is 1, ... */
129 t0 += x * 0x1000003D1ULL;
130 t1 += (t0 >> 52); t0 &= 0xFFFFFFFFFFFFFULL;
131 t2 += (t1 >> 52); t1 &= 0xFFFFFFFFFFFFFULL; m = t1;
132 t3 += (t2 >> 52); t2 &= 0xFFFFFFFFFFFFFULL; m &= t2;
133 t4 += (t3 >> 52); t3 &= 0xFFFFFFFFFFFFFULL; m &= t3;
135 /* ... except for a possible carry at bit 48 of t4 (i.e. bit 256 of the field element) */
136 VERIFY_CHECK(t4 >> 49 == 0);
138 /* At most a single final reduction is needed; check if the value is >= the field characteristic */
139 x = (t4 >> 48) | ((t4 == 0x0FFFFFFFFFFFFULL) & (m == 0xFFFFFFFFFFFFFULL)
140 & (t0 >= 0xFFFFEFFFFFC2FULL));
143 t0 += 0x1000003D1ULL;
144 t1 += (t0 >> 52); t0 &= 0xFFFFFFFFFFFFFULL;
145 t2 += (t1 >> 52); t1 &= 0xFFFFFFFFFFFFFULL;
146 t3 += (t2 >> 52); t2 &= 0xFFFFFFFFFFFFFULL;
147 t4 += (t3 >> 52); t3 &= 0xFFFFFFFFFFFFFULL;
149 /* If t4 didn't carry to bit 48 already, then it should have after any final reduction */
150 VERIFY_CHECK(t4 >> 48 == x);
152 /* Mask off the possible multiple of 2^256 from the final reduction */
153 t4 &= 0x0FFFFFFFFFFFFULL;
156 r->n[0] = t0; r->n[1] = t1; r->n[2] = t2; r->n[3] = t3; r->n[4] = t4;
161 secp256k1_fe_verify(r);
165 static int secp256k1_fe_normalizes_to_zero(secp256k1_fe *r) {
166 uint64_t t0 = r->n[0], t1 = r->n[1], t2 = r->n[2], t3 = r->n[3], t4 = r->n[4];
168 /* z0 tracks a possible raw value of 0, z1 tracks a possible raw value of P */
171 /* Reduce t4 at the start so there will be at most a single carry from the first pass */
172 uint64_t x = t4 >> 48; t4 &= 0x0FFFFFFFFFFFFULL;
174 /* The first pass ensures the magnitude is 1, ... */
175 t0 += x * 0x1000003D1ULL;
176 t1 += (t0 >> 52); t0 &= 0xFFFFFFFFFFFFFULL; z0 = t0; z1 = t0 ^ 0x1000003D0ULL;
177 t2 += (t1 >> 52); t1 &= 0xFFFFFFFFFFFFFULL; z0 |= t1; z1 &= t1;
178 t3 += (t2 >> 52); t2 &= 0xFFFFFFFFFFFFFULL; z0 |= t2; z1 &= t2;
179 t4 += (t3 >> 52); t3 &= 0xFFFFFFFFFFFFFULL; z0 |= t3; z1 &= t3;
180 z0 |= t4; z1 &= t4 ^ 0xF000000000000ULL;
182 /* ... except for a possible carry at bit 48 of t4 (i.e. bit 256 of the field element) */
183 VERIFY_CHECK(t4 >> 49 == 0);
185 return (z0 == 0) | (z1 == 0xFFFFFFFFFFFFFULL);
188 static int secp256k1_fe_normalizes_to_zero_var(secp256k1_fe *r) {
189 uint64_t t0, t1, t2, t3, t4;
196 /* Reduce t4 at the start so there will be at most a single carry from the first pass */
199 /* The first pass ensures the magnitude is 1, ... */
200 t0 += x * 0x1000003D1ULL;
202 /* z0 tracks a possible raw value of 0, z1 tracks a possible raw value of P */
203 z0 = t0 & 0xFFFFFFFFFFFFFULL;
204 z1 = z0 ^ 0x1000003D0ULL;
206 /* Fast return path should catch the majority of cases */
207 if ((z0 != 0ULL) & (z1 != 0xFFFFFFFFFFFFFULL)) {
215 t4 &= 0x0FFFFFFFFFFFFULL;
218 t2 += (t1 >> 52); t1 &= 0xFFFFFFFFFFFFFULL; z0 |= t1; z1 &= t1;
219 t3 += (t2 >> 52); t2 &= 0xFFFFFFFFFFFFFULL; z0 |= t2; z1 &= t2;
220 t4 += (t3 >> 52); t3 &= 0xFFFFFFFFFFFFFULL; z0 |= t3; z1 &= t3;
221 z0 |= t4; z1 &= t4 ^ 0xF000000000000ULL;
223 /* ... except for a possible carry at bit 48 of t4 (i.e. bit 256 of the field element) */
224 VERIFY_CHECK(t4 >> 49 == 0);
226 return (z0 == 0) | (z1 == 0xFFFFFFFFFFFFFULL);
229 SECP256K1_INLINE static void secp256k1_fe_set_int(secp256k1_fe *r, int a) {
231 r->n[1] = r->n[2] = r->n[3] = r->n[4] = 0;
235 secp256k1_fe_verify(r);
239 SECP256K1_INLINE static int secp256k1_fe_is_zero(const secp256k1_fe *a) {
240 const uint64_t *t = a->n;
242 VERIFY_CHECK(a->normalized);
243 secp256k1_fe_verify(a);
245 return (t[0] | t[1] | t[2] | t[3] | t[4]) == 0;
248 SECP256K1_INLINE static int secp256k1_fe_is_odd(const secp256k1_fe *a) {
250 VERIFY_CHECK(a->normalized);
251 secp256k1_fe_verify(a);
256 SECP256K1_INLINE static void secp256k1_fe_clear(secp256k1_fe *a) {
262 for (i=0; i<5; i++) {
267 static int secp256k1_fe_cmp_var(const secp256k1_fe *a, const secp256k1_fe *b) {
270 VERIFY_CHECK(a->normalized);
271 VERIFY_CHECK(b->normalized);
272 secp256k1_fe_verify(a);
273 secp256k1_fe_verify(b);
275 for (i = 4; i >= 0; i--) {
276 if (a->n[i] > b->n[i]) {
279 if (a->n[i] < b->n[i]) {
286 static int secp256k1_fe_set_b32(secp256k1_fe *r, const unsigned char *a) {
287 r->n[0] = (uint64_t)a[31]
288 | ((uint64_t)a[30] << 8)
289 | ((uint64_t)a[29] << 16)
290 | ((uint64_t)a[28] << 24)
291 | ((uint64_t)a[27] << 32)
292 | ((uint64_t)a[26] << 40)
293 | ((uint64_t)(a[25] & 0xF) << 48);
294 r->n[1] = (uint64_t)((a[25] >> 4) & 0xF)
295 | ((uint64_t)a[24] << 4)
296 | ((uint64_t)a[23] << 12)
297 | ((uint64_t)a[22] << 20)
298 | ((uint64_t)a[21] << 28)
299 | ((uint64_t)a[20] << 36)
300 | ((uint64_t)a[19] << 44);
301 r->n[2] = (uint64_t)a[18]
302 | ((uint64_t)a[17] << 8)
303 | ((uint64_t)a[16] << 16)
304 | ((uint64_t)a[15] << 24)
305 | ((uint64_t)a[14] << 32)
306 | ((uint64_t)a[13] << 40)
307 | ((uint64_t)(a[12] & 0xF) << 48);
308 r->n[3] = (uint64_t)((a[12] >> 4) & 0xF)
309 | ((uint64_t)a[11] << 4)
310 | ((uint64_t)a[10] << 12)
311 | ((uint64_t)a[9] << 20)
312 | ((uint64_t)a[8] << 28)
313 | ((uint64_t)a[7] << 36)
314 | ((uint64_t)a[6] << 44);
315 r->n[4] = (uint64_t)a[5]
316 | ((uint64_t)a[4] << 8)
317 | ((uint64_t)a[3] << 16)
318 | ((uint64_t)a[2] << 24)
319 | ((uint64_t)a[1] << 32)
320 | ((uint64_t)a[0] << 40);
321 if (r->n[4] == 0x0FFFFFFFFFFFFULL && (r->n[3] & r->n[2] & r->n[1]) == 0xFFFFFFFFFFFFFULL && r->n[0] >= 0xFFFFEFFFFFC2FULL) {
327 secp256k1_fe_verify(r);
332 /** Convert a field element to a 32-byte big endian value. Requires the input to be normalized */
333 static void secp256k1_fe_get_b32(unsigned char *r, const secp256k1_fe *a) {
335 VERIFY_CHECK(a->normalized);
336 secp256k1_fe_verify(a);
338 r[0] = (a->n[4] >> 40) & 0xFF;
339 r[1] = (a->n[4] >> 32) & 0xFF;
340 r[2] = (a->n[4] >> 24) & 0xFF;
341 r[3] = (a->n[4] >> 16) & 0xFF;
342 r[4] = (a->n[4] >> 8) & 0xFF;
343 r[5] = a->n[4] & 0xFF;
344 r[6] = (a->n[3] >> 44) & 0xFF;
345 r[7] = (a->n[3] >> 36) & 0xFF;
346 r[8] = (a->n[3] >> 28) & 0xFF;
347 r[9] = (a->n[3] >> 20) & 0xFF;
348 r[10] = (a->n[3] >> 12) & 0xFF;
349 r[11] = (a->n[3] >> 4) & 0xFF;
350 r[12] = ((a->n[2] >> 48) & 0xF) | ((a->n[3] & 0xF) << 4);
351 r[13] = (a->n[2] >> 40) & 0xFF;
352 r[14] = (a->n[2] >> 32) & 0xFF;
353 r[15] = (a->n[2] >> 24) & 0xFF;
354 r[16] = (a->n[2] >> 16) & 0xFF;
355 r[17] = (a->n[2] >> 8) & 0xFF;
356 r[18] = a->n[2] & 0xFF;
357 r[19] = (a->n[1] >> 44) & 0xFF;
358 r[20] = (a->n[1] >> 36) & 0xFF;
359 r[21] = (a->n[1] >> 28) & 0xFF;
360 r[22] = (a->n[1] >> 20) & 0xFF;
361 r[23] = (a->n[1] >> 12) & 0xFF;
362 r[24] = (a->n[1] >> 4) & 0xFF;
363 r[25] = ((a->n[0] >> 48) & 0xF) | ((a->n[1] & 0xF) << 4);
364 r[26] = (a->n[0] >> 40) & 0xFF;
365 r[27] = (a->n[0] >> 32) & 0xFF;
366 r[28] = (a->n[0] >> 24) & 0xFF;
367 r[29] = (a->n[0] >> 16) & 0xFF;
368 r[30] = (a->n[0] >> 8) & 0xFF;
369 r[31] = a->n[0] & 0xFF;
372 SECP256K1_INLINE static void secp256k1_fe_negate(secp256k1_fe *r, const secp256k1_fe *a, int m) {
374 VERIFY_CHECK(a->magnitude <= m);
375 secp256k1_fe_verify(a);
377 r->n[0] = 0xFFFFEFFFFFC2FULL * 2 * (m + 1) - a->n[0];
378 r->n[1] = 0xFFFFFFFFFFFFFULL * 2 * (m + 1) - a->n[1];
379 r->n[2] = 0xFFFFFFFFFFFFFULL * 2 * (m + 1) - a->n[2];
380 r->n[3] = 0xFFFFFFFFFFFFFULL * 2 * (m + 1) - a->n[3];
381 r->n[4] = 0x0FFFFFFFFFFFFULL * 2 * (m + 1) - a->n[4];
383 r->magnitude = m + 1;
385 secp256k1_fe_verify(r);
389 SECP256K1_INLINE static void secp256k1_fe_mul_int(secp256k1_fe *r, int a) {
398 secp256k1_fe_verify(r);
402 SECP256K1_INLINE static void secp256k1_fe_add(secp256k1_fe *r, const secp256k1_fe *a) {
404 secp256k1_fe_verify(a);
412 r->magnitude += a->magnitude;
414 secp256k1_fe_verify(r);
418 static void secp256k1_fe_mul(secp256k1_fe *r, const secp256k1_fe *a, const secp256k1_fe * SECP256K1_RESTRICT b) {
420 VERIFY_CHECK(a->magnitude <= 8);
421 VERIFY_CHECK(b->magnitude <= 8);
422 secp256k1_fe_verify(a);
423 secp256k1_fe_verify(b);
424 VERIFY_CHECK(r != b);
426 secp256k1_fe_mul_inner(r->n, a->n, b->n);
430 secp256k1_fe_verify(r);
434 static void secp256k1_fe_sqr(secp256k1_fe *r, const secp256k1_fe *a) {
436 VERIFY_CHECK(a->magnitude <= 8);
437 secp256k1_fe_verify(a);
439 secp256k1_fe_sqr_inner(r->n, a->n);
443 secp256k1_fe_verify(r);
447 static SECP256K1_INLINE void secp256k1_fe_cmov(secp256k1_fe *r, const secp256k1_fe *a, int flag) {
448 uint64_t mask0, mask1;
449 mask0 = flag + ~((uint64_t)0);
451 r->n[0] = (r->n[0] & mask0) | (a->n[0] & mask1);
452 r->n[1] = (r->n[1] & mask0) | (a->n[1] & mask1);
453 r->n[2] = (r->n[2] & mask0) | (a->n[2] & mask1);
454 r->n[3] = (r->n[3] & mask0) | (a->n[3] & mask1);
455 r->n[4] = (r->n[4] & mask0) | (a->n[4] & mask1);
457 if (a->magnitude > r->magnitude) {
458 r->magnitude = a->magnitude;
460 r->normalized &= a->normalized;
464 static SECP256K1_INLINE void secp256k1_fe_storage_cmov(secp256k1_fe_storage *r, const secp256k1_fe_storage *a, int flag) {
465 uint64_t mask0, mask1;
466 mask0 = flag + ~((uint64_t)0);
468 r->n[0] = (r->n[0] & mask0) | (a->n[0] & mask1);
469 r->n[1] = (r->n[1] & mask0) | (a->n[1] & mask1);
470 r->n[2] = (r->n[2] & mask0) | (a->n[2] & mask1);
471 r->n[3] = (r->n[3] & mask0) | (a->n[3] & mask1);
474 static void secp256k1_fe_to_storage(secp256k1_fe_storage *r, const secp256k1_fe *a) {
476 VERIFY_CHECK(a->normalized);
478 r->n[0] = a->n[0] | a->n[1] << 52;
479 r->n[1] = a->n[1] >> 12 | a->n[2] << 40;
480 r->n[2] = a->n[2] >> 24 | a->n[3] << 28;
481 r->n[3] = a->n[3] >> 36 | a->n[4] << 16;
484 static SECP256K1_INLINE void secp256k1_fe_from_storage(secp256k1_fe *r, const secp256k1_fe_storage *a) {
485 r->n[0] = a->n[0] & 0xFFFFFFFFFFFFFULL;
486 r->n[1] = a->n[0] >> 52 | ((a->n[1] << 12) & 0xFFFFFFFFFFFFFULL);
487 r->n[2] = a->n[1] >> 40 | ((a->n[2] << 24) & 0xFFFFFFFFFFFFFULL);
488 r->n[3] = a->n[2] >> 28 | ((a->n[3] << 36) & 0xFFFFFFFFFFFFFULL);
489 r->n[4] = a->n[3] >> 16;
496 #endif /* SECP256K1_FIELD_REPR_IMPL_H */