Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Effective affine precomputation (by Peter Dettman) #210

Merged
merged 2 commits into from
Apr 30, 2015
Merged

Effective affine precomputation (by Peter Dettman) #210

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
4f9791a
Effective affine addition in EC multiplication
peterdettman Nov 4, 2014
2d5a186
Apply effective-affine trick to precomp
peterdettman Feb 9, 2015
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 3 additions & 3 deletions src/bench_internal.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -193,7 +193,7 @@ void bench_group_double_var(void* arg) {
bench_inv_t *data = (bench_inv_t*)arg;

for (i = 0; i < 200000; i++) {
secp256k1_gej_double_var(&data->gej_x, &data->gej_x);
secp256k1_gej_double_var(&data->gej_x, &data->gej_x, NULL);
}
}

Expand All @@ -202,7 +202,7 @@ void bench_group_add_var(void* arg) {
bench_inv_t *data = (bench_inv_t*)arg;

for (i = 0; i < 200000; i++) {
secp256k1_gej_add_var(&data->gej_x, &data->gej_x, &data->gej_y);
secp256k1_gej_add_var(&data->gej_x, &data->gej_x, &data->gej_y, NULL);
}
}

Expand All @@ -220,7 +220,7 @@ void bench_group_add_affine_var(void* arg) {
bench_inv_t *data = (bench_inv_t*)arg;

for (i = 0; i < 200000; i++) {
secp256k1_gej_add_ge_var(&data->gej_x, &data->gej_x, &data->ge_y);
secp256k1_gej_add_ge_var(&data->gej_x, &data->gej_x, &data->ge_y, NULL);
}
}

Expand Down
10 changes: 5 additions & 5 deletions src/ecmult_gen_impl.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -40,7 +40,7 @@ static void secp256k1_ecmult_gen_context_build(secp256k1_ecmult_gen_context_t *c
VERIFY_CHECK(secp256k1_ge_set_xo_var(&nums_ge, &nums_x, 0));
secp256k1_gej_set_ge(&nums_gej, &nums_ge);
/* Add G to make the bits in x uniformly distributed. */
secp256k1_gej_add_ge_var(&nums_gej, &nums_gej, &secp256k1_ge_const_g);
secp256k1_gej_add_ge_var(&nums_gej, &nums_gej, &secp256k1_ge_const_g, NULL);
}

/* compute prec. */
Expand All @@ -54,18 +54,18 @@ static void secp256k1_ecmult_gen_context_build(secp256k1_ecmult_gen_context_t *c
/* Set precj[j*16 .. j*16+15] to (numsbase, numsbase + gbase, ..., numsbase + 15*gbase). */
precj[j*16] = numsbase;
for (i = 1; i < 16; i++) {
secp256k1_gej_add_var(&precj[j*16 + i], &precj[j*16 + i - 1], &gbase);
secp256k1_gej_add_var(&precj[j*16 + i], &precj[j*16 + i - 1], &gbase, NULL);
}
/* Multiply gbase by 16. */
for (i = 0; i < 4; i++) {
secp256k1_gej_double_var(&gbase, &gbase);
secp256k1_gej_double_var(&gbase, &gbase, NULL);
}
/* Multiply numbase by 2. */
secp256k1_gej_double_var(&numsbase, &numsbase);
secp256k1_gej_double_var(&numsbase, &numsbase, NULL);
if (j == 62) {
/* In the last iteration, numsbase is (1 - 2^j) * nums instead. */
secp256k1_gej_neg(&numsbase, &numsbase);
secp256k1_gej_add_var(&numsbase, &numsbase, &nums_gej);
secp256k1_gej_add_var(&numsbase, &numsbase, &nums_gej, NULL);
}
}
secp256k1_ge_set_all_gej_var(1024, prec, precj);
Expand Down
166 changes: 112 additions & 54 deletions src/ecmult_impl.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -24,62 +24,107 @@
#define WINDOW_G 16
#endif

/** Fill a table 'pre' with precomputed odd multiples of a. W determines the size of the table.
* pre will contains the values [1*a,3*a,5*a,...,(2^(w-1)-1)*a], so it needs place for
* 2^(w-2) entries.
*
* There are two versions of this function:
* - secp256k1_ecmult_precomp_wnaf_gej, which operates on group elements in jacobian notation,
* fast to precompute, but slower to use in later additions.
* - secp256k1_ecmult_precomp_wnaf_ge, which operates on group elements in affine notations,
* (much) slower to precompute, but a bit faster to use in later additions.
* To compute a*P + b*G, we use the jacobian version for P, and the affine version for G, as
* G is constant, so it only needs to be done once in advance.
/** The number of entries a table with precomputed multiples needs to have. */
#define ECMULT_TABLE_SIZE(w) (1 << ((w)-2))

/** 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_table_precomp_gej_var(secp256k1_gej_t *pre, const secp256k1_gej_t *a, int w) {
static void secp256k1_ecmult_odd_multiples_table(int n, secp256k1_gej_t *prej, secp256k1_fe_t *zr, const secp256k1_gej_t *a) {
secp256k1_gej_t d;
secp256k1_ge_t a_ge, d_ge;
int i;
pre[0] = *a;
secp256k1_gej_double_var(&d, &pre[0]);
for (i = 1; i < (1 << (w-2)); i++) {
secp256k1_gej_add_var(&pre[i], &d, &pre[i-1]);

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);
}

static void secp256k1_ecmult_table_precomp_ge_storage_var(secp256k1_ge_storage_t *pre, const secp256k1_gej_t *a, int w) {
secp256k1_gej_t d;
/** 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_t *pre, secp256k1_fe_t *globalz, const secp256k1_gej_t *a) {
secp256k1_gej_t prej[ECMULT_TABLE_SIZE(WINDOW_A)];
secp256k1_fe_t 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_t *pre, const secp256k1_gej_t *a) {
secp256k1_gej_t *prej = checked_malloc(sizeof(secp256k1_gej_t) * n);
secp256k1_ge_t *prea = checked_malloc(sizeof(secp256k1_ge_t) * n);
secp256k1_fe_t *zr = checked_malloc(sizeof(secp256k1_fe_t) * n);
int i;
const int table_size = 1 << (w-2);
secp256k1_gej_t *prej = (secp256k1_gej_t *)checked_malloc(sizeof(secp256k1_gej_t) * table_size);
secp256k1_ge_t *prea = (secp256k1_ge_t *)checked_malloc(sizeof(secp256k1_ge_t) * table_size);
prej[0] = *a;
secp256k1_gej_double_var(&d, a);
for (i = 1; i < table_size; i++) {
secp256k1_gej_add_var(&prej[i], &d, &prej[i-1]);
}
secp256k1_ge_set_all_gej_var(table_size, prea, prej);
for (i = 0; i < table_size; 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(n, prea, prej, zr);
/* Convert them to compact storage form. */
for (i = 0; i < n; i++) {
secp256k1_ge_to_storage(&pre[i], &prea[i]);
}
free(prej);

free(prea);
free(prej);
free(zr);
}

/** The number of entries a table with precomputed multiples needs to have. */
#define ECMULT_TABLE_SIZE(w) (1 << ((w)-2))

/** The following two macro retrieves a particular odd multiple from a table
* of precomputed multiples. */
#define ECMULT_TABLE_GET_GEJ(r,pre,n,w) do { \
#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_gej_neg((r), &(pre)[(-(n)-1)/2]); \
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)); \
Expand Down Expand Up @@ -112,7 +157,7 @@ static void secp256k1_ecmult_context_build(secp256k1_ecmult_context_t *ctx) {
ctx->pre_g = (secp256k1_ge_storage_t (*)[])checked_malloc(sizeof((*ctx->pre_g)[0]) * ECMULT_TABLE_SIZE(WINDOW_G));

/* precompute the tables with odd multiples */
secp256k1_ecmult_table_precomp_ge_storage_var(*ctx->pre_g, &gj, WINDOW_G);
secp256k1_ecmult_odd_multiples_table_storage_var(ECMULT_TABLE_SIZE(WINDOW_G), *ctx->pre_g, &gj);

#ifdef USE_ENDOMORPHISM
{
Expand All @@ -124,9 +169,9 @@ static void secp256k1_ecmult_context_build(secp256k1_ecmult_context_t *ctx) {
/* calculate 2^128*generator */
g_128j = gj;
for (i = 0; i < 128; i++) {
secp256k1_gej_double_var(&g_128j, &g_128j);
secp256k1_gej_double_var(&g_128j, &g_128j, NULL);
}
secp256k1_ecmult_table_precomp_ge_storage_var(*ctx->pre_g_128, &g_128j, WINDOW_G);
secp256k1_ecmult_odd_multiples_table_storage_var(ECMULT_TABLE_SIZE(WINDOW_G), *ctx->pre_g_128, &g_128j);
}
#endif
}
Expand Down Expand Up @@ -208,11 +253,11 @@ static int secp256k1_ecmult_wnaf(int *wnaf, const secp256k1_scalar_t *a, int w)
}

static void secp256k1_ecmult(const secp256k1_ecmult_context_t *ctx, secp256k1_gej_t *r, const secp256k1_gej_t *a, const secp256k1_scalar_t *na, const secp256k1_scalar_t *ng) {
secp256k1_gej_t tmpj;
secp256k1_gej_t pre_a[ECMULT_TABLE_SIZE(WINDOW_A)];
secp256k1_ge_t pre_a[ECMULT_TABLE_SIZE(WINDOW_A)];
secp256k1_ge_t tmpa;
secp256k1_fe_t Z;
#ifdef USE_ENDOMORPHISM
secp256k1_gej_t pre_a_lam[ECMULT_TABLE_SIZE(WINDOW_A)];
secp256k1_ge_t pre_a_lam[ECMULT_TABLE_SIZE(WINDOW_A)];
secp256k1_scalar_t na_1, na_lam;
/* Splitted G factors. */
secp256k1_scalar_t ng_1, ng_128;
Expand Down Expand Up @@ -252,12 +297,21 @@ static void secp256k1_ecmult(const secp256k1_ecmult_context_t *ctx, secp256k1_ge
bits = bits_na;
#endif

/* calculate odd multiples of a */
secp256k1_ecmult_table_precomp_gej_var(pre_a, a, WINDOW_A);
/* 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.
*/
secp256k1_ecmult_odd_multiples_table_globalz_windowa(pre_a, &Z, a);

#ifdef USE_ENDOMORPHISM
for (i = 0; i < ECMULT_TABLE_SIZE(WINDOW_A); i++) {
secp256k1_gej_mul_lambda(&pre_a_lam[i], &pre_a[i]);
secp256k1_ge_mul_lambda(&pre_a_lam[i], &pre_a[i]);
}

/* 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) */
Expand All @@ -281,37 +335,41 @@ static void secp256k1_ecmult(const secp256k1_ecmult_context_t *ctx, secp256k1_ge

secp256k1_gej_set_infinity(r);

for (i = bits-1; i >= 0; i--) {
for (i = bits - 1; i >= 0; i--) {
int n;
secp256k1_gej_double_var(r, r);
secp256k1_gej_double_var(r, r, NULL);
#ifdef USE_ENDOMORPHISM
if (i < bits_na_1 && (n = wnaf_na_1[i])) {
ECMULT_TABLE_GET_GEJ(&tmpj, pre_a, n, WINDOW_A);
secp256k1_gej_add_var(r, r, &tmpj);
ECMULT_TABLE_GET_GE(&tmpa, pre_a, n, WINDOW_A);
secp256k1_gej_add_ge_var(r, r, &tmpa, NULL);
}
if (i < bits_na_lam && (n = wnaf_na_lam[i])) {
ECMULT_TABLE_GET_GEJ(&tmpj, pre_a_lam, n, WINDOW_A);
secp256k1_gej_add_var(r, r, &tmpj);
ECMULT_TABLE_GET_GE(&tmpa, pre_a_lam, 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_ge_var(r, r, &tmpa);
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_ge_var(r, r, &tmpa);
secp256k1_gej_add_zinv_var(r, r, &tmpa, &Z);
}
#else
if (i < bits_na && (n = wnaf_na[i])) {
ECMULT_TABLE_GET_GEJ(&tmpj, pre_a, n, WINDOW_A);
secp256k1_gej_add_var(r, r, &tmpj);
ECMULT_TABLE_GET_GE(&tmpa, pre_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_ge_var(r, r, &tmpa);
secp256k1_gej_add_zinv_var(r, r, &tmpa, &Z);
}
#endif
}

if (!r->infinity) {
secp256k1_fe_mul(&r->z, &r->z, &Z);
}
}

#endif
28 changes: 21 additions & 7 deletions src/group.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -62,6 +62,17 @@ static void secp256k1_ge_set_gej(secp256k1_ge_t *r, secp256k1_gej_t *a);
/** Set a batch of group elements equal to the inputs given in jacobian coordinates */
static void secp256k1_ge_set_all_gej_var(size_t len, secp256k1_ge_t *r, const secp256k1_gej_t *a);

/** Set a batch of group elements equal to the inputs given in jacobian
* coordinates (with known z-ratios). zr must contain the known z-ratios such
* that mul(a[i].z, zr[i+1]) == a[i+1].z. zr[0] is ignored. */
static void secp256k1_ge_set_table_gej_var(size_t len, secp256k1_ge_t *r, const secp256k1_gej_t *a, const secp256k1_fe_t *zr);

/** Bring a batch inputs given in jacobian coordinates (with known z-ratios) to
* the same global z "denominator". zr must contain the known z-ratios such
* that mul(a[i].z, zr[i+1]) == a[i+1].z. zr[0] is ignored. The x and y
* coordinates of the result are stored in r, the common z coordinate is
* stored in globalz. */
static void secp256k1_ge_globalz_set_table_gej(size_t len, secp256k1_ge_t *r, secp256k1_fe_t *globalz, const secp256k1_gej_t *a, const secp256k1_fe_t *zr);

/** Set a group element (jacobian) equal to the point at infinity. */
static void secp256k1_gej_set_infinity(secp256k1_gej_t *r);
Expand All @@ -81,23 +92,26 @@ static void secp256k1_gej_neg(secp256k1_gej_t *r, const secp256k1_gej_t *a);
/** Check whether a group element is the point at infinity. */
static int secp256k1_gej_is_infinity(const secp256k1_gej_t *a);

/** Set r equal to the double of a. */
static void secp256k1_gej_double_var(secp256k1_gej_t *r, const secp256k1_gej_t *a);
/** Set r equal to the double of a. If rzr is not-NULL, r->z = a->z * *rzr (where infinity means an implicit z = 0). */
static void secp256k1_gej_double_var(secp256k1_gej_t *r, const secp256k1_gej_t *a, secp256k1_fe_t *rzr);
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think the doccomment needs to be updated to reflect what happens to rzr. ("Is it r.z/a.z or its recipricol?" is not obvious just from the name. I also don't think the behaviour for infinity is obvious since then rzr ought be be "0/0".)


/** Set r equal to the sum of a and b. */
static void secp256k1_gej_add_var(secp256k1_gej_t *r, const secp256k1_gej_t *a, const secp256k1_gej_t *b);
/** Set r equal to the sum of a and b. If rzr is non-NULL, r->z = a->z * *rzr (a cannot be infinity in that case). */
static void secp256k1_gej_add_var(secp256k1_gej_t *r, const secp256k1_gej_t *a, const secp256k1_gej_t *b, secp256k1_fe_t *rzr);
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Ditto here.


/** Set r equal to the sum of a and b (with b given in affine coordinates, and not infinity). */
static void secp256k1_gej_add_ge(secp256k1_gej_t *r, const secp256k1_gej_t *a, const secp256k1_ge_t *b);

/** Set r equal to the sum of a and b (with b given in affine coordinates). This is more efficient
than secp256k1_gej_add_var. It is identical to secp256k1_gej_add_ge but without constant-time
guarantee, and b is allowed to be infinity. */
static void secp256k1_gej_add_ge_var(secp256k1_gej_t *r, const secp256k1_gej_t *a, const secp256k1_ge_t *b);
guarantee, and b is allowed to be infinity. If rzr is non-NULL, r->z = a->z * *rzr (a cannot be infinity in that case). */
static void secp256k1_gej_add_ge_var(secp256k1_gej_t *r, const secp256k1_gej_t *a, const secp256k1_ge_t *b, secp256k1_fe_t *rzr);
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

and here.


/** Set r equal to the sum of a and b (with the inverse of b's Z coordinate passed as bzinv). */
static void secp256k1_gej_add_zinv_var(secp256k1_gej_t *r, const secp256k1_gej_t *a, const secp256k1_ge_t *b, const secp256k1_fe_t *bzinv);

#ifdef USE_ENDOMORPHISM
/** Set r to be equal to lambda times a, where lambda is chosen in a way such that this is very fast. */
static void secp256k1_gej_mul_lambda(secp256k1_gej_t *r, const secp256k1_gej_t *a);
static void secp256k1_ge_mul_lambda(secp256k1_ge_t *r, const secp256k1_ge_t *a);
#endif

/** Clear a secp256k1_gej_t to prevent leaking sensitive information. */
Expand Down
Loading