You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

909 lines
32 KiB

/*****************************************************************************
* Copyright (c) 2013, 2014, 2017 Pieter Wuille, Andrew Poelstra, Jonas Nick *
* Distributed under the MIT software license, see the accompanying *
* file COPYING or http://www.opensource.org/licenses/mit-license.php. *
*****************************************************************************/
10 years ago
#ifndef SECP256K1_ECMULT_IMPL_H
#define SECP256K1_ECMULT_IMPL_H
#include <string.h>
#include "group.h"
#include "scalar.h"
#include "ecmult.h"
#if defined(EXHAUSTIVE_TEST_ORDER)
/* We need to lower these values for exhaustive tests because
* the tables cannot have infinities in them (this breaks the
* affine-isomorphism stuff which tracks z-ratios) */
# if EXHAUSTIVE_TEST_ORDER > 128
# define WINDOW_A 5
# define WINDOW_G 8
# elif EXHAUSTIVE_TEST_ORDER > 8
# define WINDOW_A 4
# define WINDOW_G 4
# else
# define WINDOW_A 2
# define WINDOW_G 2
# endif
#else
/* optimal for 128-bit and 256-bit exponents. */
#define WINDOW_A 5
/** larger numbers may result in slightly better performance, at the cost of
exponentially larger precomputed tables. */
#ifdef USE_ENDOMORPHISM
/** Two tables for window size 15: 1.375 MiB. */
#define WINDOW_G 15
#else
/** One table for window size 16: 1.375 MiB. */
#define WINDOW_G 16
#endif
#endif
#ifdef USE_ENDOMORPHISM
#define WNAF_BITS 128
#else
#define WNAF_BITS 256
#endif
#define WNAF_SIZE(w) ((WNAF_BITS + (w) - 1) / (w))
/** The number of entries a table with precomputed multiples needs to have. */
#define ECMULT_TABLE_SIZE(w) (1 << ((w)-2))
/* The number of objects allocated on the scratch space for ecmult_multi algorithms */
#define PIPPENGER_SCRATCH_OBJECTS 6
#define STRAUSS_SCRATCH_OBJECTS 6
/* Minimum number of points for which pippenger_wnaf is faster than strauss wnaf */
#ifdef USE_ENDOMORPHISM
#define ECMULT_PIPPENGER_THRESHOLD 96
#else
#define ECMULT_PIPPENGER_THRESHOLD 156
#endif
/** Fill a table 'prej' with precomputed odd multiples of a. Prej will contain
* the values [1*a,3*a,...,(2*n-1)*a], so it space for n values. zr[0] will
* contain prej[0].z / a.z. The other zr[i] values = prej[i].z / prej[i-1].z.
* Prej's Z values are undefined, except for the last value.
*/
static void secp256k1_ecmult_odd_multiples_table(int n, secp256k1_gej *prej, secp256k1_fe *zr, const secp256k1_gej *a) {
secp256k1_gej d;
secp256k1_ge a_ge, d_ge;
int i;
VERIFY_CHECK(!a->infinity);
secp256k1_gej_double_var(&d, a, NULL);
/*
* Perform the additions on an isomorphism where 'd' is affine: drop the z coordinate
* of 'd', and scale the 1P starting value's x/y coordinates without changing its z.
*/
d_ge.x = d.x;
d_ge.y = d.y;
d_ge.infinity = 0;
secp256k1_ge_set_gej_zinv(&a_ge, a, &d.z);
prej[0].x = a_ge.x;
prej[0].y = a_ge.y;
prej[0].z = a->z;
prej[0].infinity = 0;
zr[0] = d.z;
for (i = 1; i < n; i++) {
secp256k1_gej_add_ge_var(&prej[i], &prej[i-1], &d_ge, &zr[i]);
}
/*
* Each point in 'prej' has a z coordinate too small by a factor of 'd.z'. Only
* the final point's z coordinate is actually used though, so just update that.
*/
secp256k1_fe_mul(&prej[n-1].z, &prej[n-1].z, &d.z);
}
/** Fill a table 'pre' with precomputed odd multiples of a.
*
* There are two versions of this function:
* - secp256k1_ecmult_odd_multiples_table_globalz_windowa which brings its
* resulting point set to a single constant Z denominator, stores the X and Y
* coordinates as ge_storage points in pre, and stores the global Z in rz.
* It only operates on tables sized for WINDOW_A wnaf multiples.
* - secp256k1_ecmult_odd_multiples_table_storage_var, which converts its
* resulting point set to actually affine points, and stores those in pre.
* It operates on tables of any size, but uses heap-allocated temporaries.
*
* To compute a*P + b*G, we compute a table for P using the first function,
* and for G using the second (which requires an inverse, but it only needs to
* happen once).
*/
static void secp256k1_ecmult_odd_multiples_table_globalz_windowa(secp256k1_ge *pre, secp256k1_fe *globalz, const secp256k1_gej *a) {
secp256k1_gej prej[ECMULT_TABLE_SIZE(WINDOW_A)];
secp256k1_fe zr[ECMULT_TABLE_SIZE(WINDOW_A)];
/* Compute the odd multiples in Jacobian form. */
secp256k1_ecmult_odd_multiples_table(ECMULT_TABLE_SIZE(WINDOW_A), prej, zr, a);
/* Bring them to the same Z denominator. */
secp256k1_ge_globalz_set_table_gej(ECMULT_TABLE_SIZE(WINDOW_A), pre, globalz, prej, zr);
}
static void secp256k1_ecmult_odd_multiples_table_storage_var(int n, secp256k1_ge_storage *pre, const secp256k1_gej *a, const secp256k1_callback *cb) {
secp256k1_gej *prej = (secp256k1_gej*)checked_malloc(cb, sizeof(secp256k1_gej) * n);
secp256k1_ge *prea = (secp256k1_ge*)checked_malloc(cb, sizeof(secp256k1_ge) * n);
secp256k1_fe *zr = (secp256k1_fe*)checked_malloc(cb, sizeof(secp256k1_fe) * n);
int i;
/* Compute the odd multiples in Jacobian form. */
secp256k1_ecmult_odd_multiples_table(n, prej, zr, a);
/* Convert them in batch to affine coordinates. */
secp256k1_ge_set_table_gej_var(prea, prej, zr, n);
/* Convert them to compact storage form. */
for (i = 0; i < n; i++) {
secp256k1_ge_to_storage(&pre[i], &prea[i]);
}
free(prea);
free(prej);
free(zr);
}
/** The following two macro retrieves a particular odd multiple from a table
* of precomputed multiples. */
#define ECMULT_TABLE_GET_GE(r,pre,n,w) do { \
VERIFY_CHECK(((n) & 1) == 1); \
VERIFY_CHECK((n) >= -((1 << ((w)-1)) - 1)); \
VERIFY_CHECK((n) <= ((1 << ((w)-1)) - 1)); \
if ((n) > 0) { \
*(r) = (pre)[((n)-1)/2]; \
} else { \
secp256k1_ge_neg((r), &(pre)[(-(n)-1)/2]); \
} \
} while(0)
#define ECMULT_TABLE_GET_GE_STORAGE(r,pre,n,w) do { \
VERIFY_CHECK(((n) & 1) == 1); \
VERIFY_CHECK((n) >= -((1 << ((w)-1)) - 1)); \
VERIFY_CHECK((n) <= ((1 << ((w)-1)) - 1)); \
if ((n) > 0) { \
secp256k1_ge_from_storage((r), &(pre)[((n)-1)/2]); \
} else { \
secp256k1_ge_from_storage((r), &(pre)[(-(n)-1)/2]); \
secp256k1_ge_neg((r), (r)); \
} \
} while(0)
static void secp256k1_ecmult_context_init(secp256k1_ecmult_context *ctx) {
ctx->pre_g = NULL;
#ifdef USE_ENDOMORPHISM
ctx->pre_g_128 = NULL;
#endif
}
static void secp256k1_ecmult_context_build(secp256k1_ecmult_context *ctx, const secp256k1_callback *cb) {
secp256k1_gej gj;
if (ctx->pre_g != NULL) {
return;
}
/* get the generator */
secp256k1_gej_set_ge(&gj, &secp256k1_ge_const_g);
ctx->pre_g = (secp256k1_ge_storage (*)[])checked_malloc(cb, sizeof((*ctx->pre_g)[0]) * ECMULT_TABLE_SIZE(WINDOW_G));
/* precompute the tables with odd multiples */
secp256k1_ecmult_odd_multiples_table_storage_var(ECMULT_TABLE_SIZE(WINDOW_G), *ctx->pre_g, &gj, cb);
#ifdef USE_ENDOMORPHISM
{
secp256k1_gej g_128j;
int i;
ctx->pre_g_128 = (secp256k1_ge_storage (*)[])checked_malloc(cb, sizeof((*ctx->pre_g_128)[0]) * ECMULT_TABLE_SIZE(WINDOW_G));
/* calculate 2^128*generator */
g_128j = gj;
for (i = 0; i < 128; i++) {
secp256k1_gej_double_var(&g_128j, &g_128j, NULL);
}
secp256k1_ecmult_odd_multiples_table_storage_var(ECMULT_TABLE_SIZE(WINDOW_G), *ctx->pre_g_128, &g_128j, cb);
}
#endif
}
static void secp256k1_ecmult_context_clone(secp256k1_ecmult_context *dst,
const secp256k1_ecmult_context *src, const secp256k1_callback *cb) {
if (src->pre_g == NULL) {
dst->pre_g = NULL;
} else {
size_t size = sizeof((*dst->pre_g)[0]) * ECMULT_TABLE_SIZE(WINDOW_G);
dst->pre_g = (secp256k1_ge_storage (*)[])checked_malloc(cb, size);
memcpy(dst->pre_g, src->pre_g, size);
}
#ifdef USE_ENDOMORPHISM
if (src->pre_g_128 == NULL) {
dst->pre_g_128 = NULL;
} else {
size_t size = sizeof((*dst->pre_g_128)[0]) * ECMULT_TABLE_SIZE(WINDOW_G);
dst->pre_g_128 = (secp256k1_ge_storage (*)[])checked_malloc(cb, size);
memcpy(dst->pre_g_128, src->pre_g_128, size);
}
#endif
}
static int secp256k1_ecmult_context_is_built(const secp256k1_ecmult_context *ctx) {
return ctx->pre_g != NULL;
}
static void secp256k1_ecmult_context_clear(secp256k1_ecmult_context *ctx) {
free(ctx->pre_g);
#ifdef USE_ENDOMORPHISM
free(ctx->pre_g_128);
#endif
secp256k1_ecmult_context_init(ctx);
}
/** Convert a number to WNAF notation. The number becomes represented by sum(2^i * wnaf[i], i=0..bits),
* with the following guarantees:
* - each wnaf[i] is either 0, or an odd integer between -(1<<(w-1) - 1) and (1<<(w-1) - 1)
* - two non-zero entries in wnaf are separated by at least w-1 zeroes.
* - the number of set values in wnaf is returned. This number is at most 256, and at most one more
* than the number of bits in the (absolute value) of the input.
*/
static int secp256k1_ecmult_wnaf(int *wnaf, int len, const secp256k1_scalar *a, int w) {
secp256k1_scalar s = *a;
int last_set_bit = -1;
int bit = 0;
int sign = 1;
int carry = 0;
VERIFY_CHECK(wnaf != NULL);
VERIFY_CHECK(0 <= len && len <= 256);
VERIFY_CHECK(a != NULL);
VERIFY_CHECK(2 <= w && w <= 31);
memset(wnaf, 0, len * sizeof(wnaf[0]));
if (secp256k1_scalar_get_bits(&s, 255, 1)) {
secp256k1_scalar_negate(&s, &s);
sign = -1;
}
while (bit < len) {
int now;
int word;
if (secp256k1_scalar_get_bits(&s, bit, 1) == (unsigned int)carry) {
bit++;
continue;
}
now = w;
if (now > len - bit) {
now = len - bit;
}
word = secp256k1_scalar_get_bits_var(&s, bit, now) + carry;
carry = (word >> (w-1)) & 1;
word -= carry << w;
wnaf[bit] = sign * word;
last_set_bit = bit;
bit += now;
}
#ifdef VERIFY
CHECK(carry == 0);
while (bit < 256) {
CHECK(secp256k1_scalar_get_bits(&s, bit++, 1) == 0);
}
#endif
return last_set_bit + 1;
}
struct secp256k1_strauss_point_state {
#ifdef USE_ENDOMORPHISM
secp256k1_scalar na_1, na_lam;
int wnaf_na_1[130];
int wnaf_na_lam[130];
int bits_na_1;
int bits_na_lam;
#else
int wnaf_na[256];
int bits_na;
#endif
size_t input_pos;
};
struct secp256k1_strauss_state {
secp256k1_gej* prej;
secp256k1_fe* zr;
secp256k1_ge* pre_a;
#ifdef USE_ENDOMORPHISM
secp256k1_ge* pre_a_lam;
#endif
struct secp256k1_strauss_point_state* ps;
};
static void secp256k1_ecmult_strauss_wnaf(const secp256k1_ecmult_context *ctx, const struct secp256k1_strauss_state *state, secp256k1_gej *r, int num, const secp256k1_gej *a, const secp256k1_scalar *na, const secp256k1_scalar *ng) {
secp256k1_ge tmpa;
secp256k1_fe Z;
#ifdef USE_ENDOMORPHISM
/* Splitted G factors. */
secp256k1_scalar ng_1, ng_128;
int wnaf_ng_1[129];
int bits_ng_1 = 0;
int wnaf_ng_128[129];
int bits_ng_128 = 0;
#else
int wnaf_ng[256];
int bits_ng = 0;
#endif
int i;
int bits = 0;
int np;
int no = 0;
for (np = 0; np < num; ++np) {
if (secp256k1_scalar_is_zero(&na[np]) || secp256k1_gej_is_infinity(&a[np])) {
continue;
}
state->ps[no].input_pos = np;
#ifdef USE_ENDOMORPHISM
/* split na into na_1 and na_lam (where na = na_1 + na_lam*lambda, and na_1 and na_lam are ~128 bit) */
secp256k1_scalar_split_lambda(&state->ps[no].na_1, &state->ps[no].na_lam, &na[np]);
/* build wnaf representation for na_1 and na_lam. */
state->ps[no].bits_na_1 = secp256k1_ecmult_wnaf(state->ps[no].wnaf_na_1, 130, &state->ps[no].na_1, WINDOW_A);
state->ps[no].bits_na_lam = secp256k1_ecmult_wnaf(state->ps[no].wnaf_na_lam, 130, &state->ps[no].na_lam, WINDOW_A);
VERIFY_CHECK(state->ps[no].bits_na_1 <= 130);
VERIFY_CHECK(state->ps[no].bits_na_lam <= 130);
if (state->ps[no].bits_na_1 > bits) {
bits = state->ps[no].bits_na_1;
}
if (state->ps[no].bits_na_lam > bits) {
bits = state->ps[no].bits_na_lam;
}
#else
/* build wnaf representation for na. */
state->ps[no].bits_na = secp256k1_ecmult_wnaf(state->ps[no].wnaf_na, 256, &na[np], WINDOW_A);
if (state->ps[no].bits_na > bits) {
bits = state->ps[no].bits_na;
}
#endif
++no;
}
/* Calculate odd multiples of a.
* All multiples are brought to the same Z 'denominator', which is stored
* in Z. Due to secp256k1' isomorphism we can do all operations pretending
* that the Z coordinate was 1, use affine addition formulae, and correct
* the Z coordinate of the result once at the end.
* The exception is the precomputed G table points, which are actually
* affine. Compared to the base used for other points, they have a Z ratio
* of 1/Z, so we can use secp256k1_gej_add_zinv_var, which uses the same
* isomorphism to efficiently add with a known Z inverse.
*/
if (no > 0) {
/* Compute the odd multiples in Jacobian form. */
secp256k1_ecmult_odd_multiples_table(ECMULT_TABLE_SIZE(WINDOW_A), state->prej, state->zr, &a[state->ps[0].input_pos]);
for (np = 1; np < no; ++np) {
secp256k1_gej tmp = a[state->ps[np].input_pos];
#ifdef VERIFY
secp256k1_fe_normalize_var(&(state->prej[(np - 1) * ECMULT_TABLE_SIZE(WINDOW_A) + ECMULT_TABLE_SIZE(WINDOW_A) - 1].z));
#endif
secp256k1_gej_rescale(&tmp, &(state->prej[(np - 1) * ECMULT_TABLE_SIZE(WINDOW_A) + ECMULT_TABLE_SIZE(WINDOW_A) - 1].z));
secp256k1_ecmult_odd_multiples_table(ECMULT_TABLE_SIZE(WINDOW_A), state->prej + np * ECMULT_TABLE_SIZE(WINDOW_A), state->zr + np * ECMULT_TABLE_SIZE(WINDOW_A), &tmp);
secp256k1_fe_mul(state->zr + np * ECMULT_TABLE_SIZE(WINDOW_A), state->zr + np * ECMULT_TABLE_SIZE(WINDOW_A), &(a[state->ps[np].input_pos].z));
}
/* Bring them to the same Z denominator. */
secp256k1_ge_globalz_set_table_gej(ECMULT_TABLE_SIZE(WINDOW_A) * no, state->pre_a, &Z, state->prej, state->zr);
} else {
secp256k1_fe_set_int(&Z, 1);
}
#ifdef USE_ENDOMORPHISM
for (np = 0; np < no; ++np) {
for (i = 0; i < ECMULT_TABLE_SIZE(WINDOW_A); i++) {
secp256k1_ge_mul_lambda(&state->pre_a_lam[np * ECMULT_TABLE_SIZE(WINDOW_A) + i], &state->pre_a[np * ECMULT_TABLE_SIZE(WINDOW_A) + i]);
}
}
if (ng) {
/* split ng into ng_1 and ng_128 (where gn = gn_1 + gn_128*2^128, and gn_1 and gn_128 are ~128 bit) */
secp256k1_scalar_split_128(&ng_1, &ng_128, ng);
/* Build wnaf representation for ng_1 and ng_128 */
bits_ng_1 = secp256k1_ecmult_wnaf(wnaf_ng_1, 129, &ng_1, WINDOW_G);
bits_ng_128 = secp256k1_ecmult_wnaf(wnaf_ng_128, 129, &ng_128, WINDOW_G);
if (bits_ng_1 > bits) {
bits = bits_ng_1;
}
if (bits_ng_128 > bits) {
bits = bits_ng_128;
}
}
#else
if (ng) {
bits_ng = secp256k1_ecmult_wnaf(wnaf_ng, 256, ng, WINDOW_G);
if (bits_ng > bits) {
bits = bits_ng;
}
}
#endif
secp256k1_gej_set_infinity(r);
for (i = bits - 1; i >= 0; i--) {
int n;
secp256k1_gej_double_var(r, r, NULL);
#ifdef USE_ENDOMORPHISM
for (np = 0; np < no; ++np) {
if (i < state->ps[np].bits_na_1 && (n = state->ps[np].wnaf_na_1[i])) {
ECMULT_TABLE_GET_GE(&tmpa, state->pre_a + np * ECMULT_TABLE_SIZE(WINDOW_A), n, WINDOW_A);
secp256k1_gej_add_ge_var(r, r, &tmpa, NULL);
}
if (i < state->ps[np].bits_na_lam && (n = state->ps[np].wnaf_na_lam[i])) {
ECMULT_TABLE_GET_GE(&tmpa, state->pre_a_lam + np * ECMULT_TABLE_SIZE(WINDOW_A), n, WINDOW_A);
secp256k1_gej_add_ge_var(r, r, &tmpa, NULL);
}
}
if (i < bits_ng_1 && (n = wnaf_ng_1[i])) {
ECMULT_TABLE_GET_GE_STORAGE(&tmpa, *ctx->pre_g, n, WINDOW_G);
secp256k1_gej_add_zinv_var(r, r, &tmpa, &Z);
}
if (i < bits_ng_128 && (n = wnaf_ng_128[i])) {
ECMULT_TABLE_GET_GE_STORAGE(&tmpa, *ctx->pre_g_128, n, WINDOW_G);
secp256k1_gej_add_zinv_var(r, r, &tmpa, &Z);
}
#else
for (np = 0; np < no; ++np) {
if (i < state->ps[np].bits_na && (n = state->ps[np].wnaf_na[i])) {
ECMULT_TABLE_GET_GE(&tmpa, state->pre_a + np * ECMULT_TABLE_SIZE(WINDOW_A), n, WINDOW_A);
secp256k1_gej_add_ge_var(r, r, &tmpa, NULL);
}
}
if (i < bits_ng && (n = wnaf_ng[i])) {
ECMULT_TABLE_GET_GE_STORAGE(&tmpa, *ctx->pre_g, n, WINDOW_G);
secp256k1_gej_add_zinv_var(r, r, &tmpa, &Z);
}
#endif
}
if (!r->infinity) {
secp256k1_fe_mul(&r->z, &r->z, &Z);
}
}
static void secp256k1_ecmult(const secp256k1_ecmult_context *ctx, secp256k1_gej *r, const secp256k1_gej *a, const secp256k1_scalar *na, const secp256k1_scalar *ng) {
secp256k1_gej prej[ECMULT_TABLE_SIZE(WINDOW_A)];
secp256k1_fe zr[ECMULT_TABLE_SIZE(WINDOW_A)];
secp256k1_ge pre_a[ECMULT_TABLE_SIZE(WINDOW_A)];
struct secp256k1_strauss_point_state ps[1];
#ifdef USE_ENDOMORPHISM
secp256k1_ge pre_a_lam[ECMULT_TABLE_SIZE(WINDOW_A)];
#endif
struct secp256k1_strauss_state state;
state.prej = prej;
state.zr = zr;
state.pre_a = pre_a;
#ifdef USE_ENDOMORPHISM
state.pre_a_lam = pre_a_lam;
#endif
state.ps = ps;
secp256k1_ecmult_strauss_wnaf(ctx, &state, r, 1, a, na, ng);
}
static size_t secp256k1_strauss_scratch_size(size_t n_points) {
#ifdef USE_ENDOMORPHISM
static const size_t point_size = (2 * sizeof(secp256k1_ge) + sizeof(secp256k1_gej) + sizeof(secp256k1_fe)) * ECMULT_TABLE_SIZE(WINDOW_A) + sizeof(struct secp256k1_strauss_point_state) + sizeof(secp256k1_gej) + sizeof(secp256k1_scalar);
#else
static const size_t point_size = (sizeof(secp256k1_ge) + sizeof(secp256k1_gej) + sizeof(secp256k1_fe)) * ECMULT_TABLE_SIZE(WINDOW_A) + sizeof(struct secp256k1_strauss_point_state) + sizeof(secp256k1_gej) + sizeof(secp256k1_scalar);
#endif
return n_points*point_size;
}
static int secp256k1_ecmult_strauss_batch(const secp256k1_ecmult_context *ctx, secp256k1_scratch *scratch, secp256k1_gej *r, const secp256k1_scalar *inp_g_sc, secp256k1_ecmult_multi_callback cb, void *cbdata, size_t n_points, size_t cb_offset) {
secp256k1_gej* points;
secp256k1_scalar* scalars;
struct secp256k1_strauss_state state;
size_t i;
secp256k1_gej_set_infinity(r);
if (inp_g_sc == NULL && n_points == 0) {
return 1;
}
if (!secp256k1_scratch_resize(scratch, secp256k1_strauss_scratch_size(n_points), STRAUSS_SCRATCH_OBJECTS)) {
return 0;
}
secp256k1_scratch_reset(scratch);
points = (secp256k1_gej*)secp256k1_scratch_alloc(scratch, n_points * sizeof(secp256k1_gej));
scalars = (secp256k1_scalar*)secp256k1_scratch_alloc(scratch, n_points * sizeof(secp256k1_scalar));
state.prej = (secp256k1_gej*)secp256k1_scratch_alloc(scratch, n_points * ECMULT_TABLE_SIZE(WINDOW_A) * sizeof(secp256k1_gej));
state.zr = (secp256k1_fe*)secp256k1_scratch_alloc(scratch, n_points * ECMULT_TABLE_SIZE(WINDOW_A) * sizeof(secp256k1_fe));
#ifdef USE_ENDOMORPHISM
state.pre_a = (secp256k1_ge*)secp256k1_scratch_alloc(scratch, n_points * 2 * ECMULT_TABLE_SIZE(WINDOW_A) * sizeof(secp256k1_ge));
state.pre_a_lam = state.pre_a + n_points * ECMULT_TABLE_SIZE(WINDOW_A);
#else
state.pre_a = (secp256k1_ge*)secp256k1_scratch_alloc(scratch, n_points * ECMULT_TABLE_SIZE(WINDOW_A) * sizeof(secp256k1_ge));
#endif
state.ps = (struct secp256k1_strauss_point_state*)secp256k1_scratch_alloc(scratch, n_points * sizeof(struct secp256k1_strauss_point_state));
for (i = 0; i < n_points; i++) {
secp256k1_ge point;
if (!cb(&scalars[i], &point, i+cb_offset, cbdata)) return 0;
secp256k1_gej_set_ge(&points[i], &point);
}
secp256k1_ecmult_strauss_wnaf(ctx, &state, r, n_points, points, scalars, inp_g_sc);
return 1;
}
/* Wrapper for secp256k1_ecmult_multi_func interface */
static int secp256k1_ecmult_strauss_batch_single(const secp256k1_ecmult_context *actx, secp256k1_scratch *scratch, secp256k1_gej *r, const secp256k1_scalar *inp_g_sc, secp256k1_ecmult_multi_callback cb, void *cbdata, size_t n) {
return secp256k1_ecmult_strauss_batch(actx, scratch, r, inp_g_sc, cb, cbdata, n, 0);
}
/** Convert a number to WNAF notation.
* The number becomes represented by sum(2^{wi} * wnaf[i], i=0..WNAF_SIZE(w)+1) - return_val.
* It has the following guarantees:
* - each wnaf[i] is either 0 or an odd integer between -(1 << w) and (1 << w)
* - the number of words set is always WNAF_SIZE(w)
* - the returned skew is 0 without endomorphism, or 0 or 1 with endomorphism
*/
static int secp256k1_wnaf_fixed(int *wnaf, const secp256k1_scalar *s, int w) {
int sign = 0;
int skew = 0;
int pos = 1;
#ifndef USE_ENDOMORPHISM
secp256k1_scalar neg_s;
#endif
const secp256k1_scalar *work = s;
if (secp256k1_scalar_is_zero(s)) {
while (pos * w < WNAF_BITS) {
wnaf[pos] = 0;
++pos;
}
return 0;
}
if (secp256k1_scalar_is_even(s)) {
#ifdef USE_ENDOMORPHISM
skew = 1;
#else
secp256k1_scalar_negate(&neg_s, s);
work = &neg_s;
sign = -1;
#endif
}
wnaf[0] = (secp256k1_scalar_get_bits_var(work, 0, w) + skew + sign) ^ sign;
while (pos * w < WNAF_BITS) {
int now = w;
int val;
if (now + pos * w > WNAF_BITS) {
now = WNAF_BITS - pos * w;
}
val = secp256k1_scalar_get_bits_var(work, pos * w, now);
if ((val & 1) == 0) {
wnaf[pos - 1] -= ((1 << w) + sign) ^ sign;
wnaf[pos] = (val + 1 + sign) ^ sign;
} else {
wnaf[pos] = (val + sign) ^ sign;
}
++pos;
}
VERIFY_CHECK(pos == WNAF_SIZE(w));
return skew;
}
struct secp256k1_pippenger_point_state {
int skew_na;
size_t input_pos;
};
struct secp256k1_pippenger_state {
int *wnaf_na;
struct secp256k1_pippenger_point_state* ps;
};
/*
* pippenger_wnaf computes the result of a multi-point multiplication as
* follows: The scalars are brought into wnaf with n_wnaf elements each. Then
* for every i < n_wnaf, first each point is added to a "bucket" corresponding
* to the point's wnaf[i]. Second, the buckets are added together such that
* r += 1*bucket[0] + 3*bucket[1] + 5*bucket[2] + ...
*/
static int secp256k1_ecmult_pippenger_wnaf(secp256k1_gej *buckets, int bucket_window, struct secp256k1_pippenger_state *state, secp256k1_gej *r, secp256k1_scalar *sc, secp256k1_ge *pt, size_t num) {
size_t n_wnaf = WNAF_SIZE(bucket_window+1);
size_t np;
size_t no = 0;
int i;
int j;
for (np = 0; np < num; ++np) {
if (secp256k1_scalar_is_zero(&sc[np]) || secp256k1_ge_is_infinity(&pt[np])) {
continue;
}
state->ps[no].input_pos = np;
state->ps[no].skew_na = secp256k1_wnaf_fixed(&state->wnaf_na[no*n_wnaf], &sc[np], bucket_window+1);
no++;
}
secp256k1_gej_set_infinity(r);
if (no == 0) {
return 1;
}
for (i = n_wnaf - 1; i >= 0; i--) {
secp256k1_gej running_sum;
secp256k1_gej walking_sum;
for(j = 0; j < ECMULT_TABLE_SIZE(bucket_window+2); j++) {
secp256k1_gej_set_infinity(&buckets[j]);
}
for(j = 0; j < bucket_window+1; j++) {
secp256k1_gej_double_var(r, r, NULL);
}
for (np = 0; np < no; ++np) {
int n = state->wnaf_na[np*n_wnaf + i];
struct secp256k1_pippenger_point_state point_state = state->ps[np];
secp256k1_ge tmp;
int idx;
#ifdef USE_ENDOMORPHISM
if (i == 0) {
/* correct for wnaf skew */
int skew = point_state.skew_na;
if (skew) {
secp256k1_ge_neg(&tmp, &pt[point_state.input_pos]);
secp256k1_gej_add_ge_var(&buckets[0], &buckets[0], &tmp, NULL);
}
}
#endif
if (n > 0) {
idx = (n - 1)/2;
secp256k1_gej_add_ge_var(&buckets[idx], &buckets[idx], &pt[point_state.input_pos], NULL);
} else if (n < 0) {
idx = -(n + 1)/2;
secp256k1_ge_neg(&tmp, &pt[point_state.input_pos]);
secp256k1_gej_add_ge_var(&buckets[idx], &buckets[idx], &tmp, NULL);
}
}
secp256k1_gej_set_infinity(&running_sum);
secp256k1_gej_set_infinity(&walking_sum);
/* Compute walking_sum as bucket[0] + 3*bucket[1] + 5*bucket[2] + ...
* by first setting
* running_sum = bucket[0] + bucket[1] + bucket[2] + ...
* walking_sum = bucket[0] + 2*bucket[1] + 3*bucket[2] + ...
* and then computing
* walking_sum = 2*walking_sum - running_sum
*/
for(j = ECMULT_TABLE_SIZE(bucket_window+2) - 1; j >