diff --git a/src/secp256k1/README.md b/src/secp256k1/README.md --- a/src/secp256k1/README.md +++ b/src/secp256k1/README.md @@ -38,11 +38,11 @@ * Optimized implementation of arithmetic modulo the curve's field size (2^256 - 0x1000003D1). * Using 5 52-bit limbs (including hand-optimized assembly for x86_64, by Diederik Huys). * Using 10 26-bit limbs (including hand-optimized assembly for 32-bit ARM, by Wladimir J. van der Laan). - * Field inverses and square roots using a sliding window over blocks of 1s (by Peter Dettman). * Scalar operations * Optimized implementation without data-dependent branches of arithmetic modulo the curve's order. * Using 4 64-bit limbs (relying on __int128 support in the compiler). * Using 8 32-bit limbs. +* Modular inverses (both field elements and scalars) based on [safegcd](https://gcd.cr.yp.to/index.html) with some modifications, and a variable-time variant (by Peter Dettman). * Group operations * Point addition formula specifically simplified for the curve equation (y^2 = x^3 + 7). * Use addition between points in Jacobian and affine coordinates where possible. diff --git a/src/secp256k1/src/field_10x26_impl.h b/src/secp256k1/src/field_10x26_impl.h --- a/src/secp256k1/src/field_10x26_impl.h +++ b/src/secp256k1/src/field_10x26_impl.h @@ -9,6 +9,7 @@ #include "util.h" #include "field.h" +#include "modinv32_impl.h" #ifdef VERIFY static void secp256k1_fe_verify(const secp256k1_fe *a) { @@ -1164,131 +1165,92 @@ #endif } -static void secp256k1_fe_inv(secp256k1_fe *r, const secp256k1_fe *a) { - secp256k1_fe x2, x3, x6, x9, x11, x22, x44, x88, x176, x220, x223, t1; - int j; +static void secp256k1_fe_from_signed30(secp256k1_fe *r, const secp256k1_modinv32_signed30 *a) { + const uint32_t M26 = UINT32_MAX >> 6; + const uint32_t a0 = a->v[0], a1 = a->v[1], a2 = a->v[2], a3 = a->v[3], a4 = a->v[4], + a5 = a->v[5], a6 = a->v[6], a7 = a->v[7], a8 = a->v[8]; - /** The binary representation of (p - 2) has 5 blocks of 1s, with lengths in - * { 1, 2, 22, 223 }. Use an addition chain to calculate 2^n - 1 for each block: - * [1], [2], 3, 6, 9, 11, [22], 44, 88, 176, 220, [223] + /* The output from secp256k1_modinv32{_var} should be normalized to range [0,modulus), and + * have limbs in [0,2^30). The modulus is < 2^256, so the top limb must be below 2^(256-30*8). */ + VERIFY_CHECK(a0 >> 30 == 0); + VERIFY_CHECK(a1 >> 30 == 0); + VERIFY_CHECK(a2 >> 30 == 0); + VERIFY_CHECK(a3 >> 30 == 0); + VERIFY_CHECK(a4 >> 30 == 0); + VERIFY_CHECK(a5 >> 30 == 0); + VERIFY_CHECK(a6 >> 30 == 0); + VERIFY_CHECK(a7 >> 30 == 0); + VERIFY_CHECK(a8 >> 16 == 0); + + r->n[0] = a0 & M26; + r->n[1] = (a0 >> 26 | a1 << 4) & M26; + r->n[2] = (a1 >> 22 | a2 << 8) & M26; + r->n[3] = (a2 >> 18 | a3 << 12) & M26; + r->n[4] = (a3 >> 14 | a4 << 16) & M26; + r->n[5] = (a4 >> 10 | a5 << 20) & M26; + r->n[6] = (a5 >> 6 | a6 << 24) & M26; + r->n[7] = (a6 >> 2 ) & M26; + r->n[8] = (a6 >> 28 | a7 << 2) & M26; + r->n[9] = (a7 >> 24 | a8 << 6); - secp256k1_fe_sqr(&x2, a); - secp256k1_fe_mul(&x2, &x2, a); - - secp256k1_fe_sqr(&x3, &x2); - secp256k1_fe_mul(&x3, &x3, a); - - x6 = x3; - for (j=0; j<3; j++) { - secp256k1_fe_sqr(&x6, &x6); - } - secp256k1_fe_mul(&x6, &x6, &x3); - - x9 = x6; - for (j=0; j<3; j++) { - secp256k1_fe_sqr(&x9, &x9); - } - secp256k1_fe_mul(&x9, &x9, &x3); +#ifdef VERIFY + r->magnitude = 1; + r->normalized = 1; + secp256k1_fe_verify(r); +#endif +} - x11 = x9; - for (j=0; j<2; j++) { - secp256k1_fe_sqr(&x11, &x11); - } - secp256k1_fe_mul(&x11, &x11, &x2); +static void secp256k1_fe_to_signed30(secp256k1_modinv32_signed30 *r, const secp256k1_fe *a) { + const uint32_t M30 = UINT32_MAX >> 2; + const uint64_t a0 = a->n[0], a1 = a->n[1], a2 = a->n[2], a3 = a->n[3], a4 = a->n[4], + a5 = a->n[5], a6 = a->n[6], a7 = a->n[7], a8 = a->n[8], a9 = a->n[9]; - x22 = x11; - for (j=0; j<11; j++) { - secp256k1_fe_sqr(&x22, &x22); - } - secp256k1_fe_mul(&x22, &x22, &x11); +#ifdef VERIFY + VERIFY_CHECK(a->normalized); +#endif - x44 = x22; - for (j=0; j<22; j++) { - secp256k1_fe_sqr(&x44, &x44); - } - secp256k1_fe_mul(&x44, &x44, &x22); + r->v[0] = (a0 | a1 << 26) & M30; + r->v[1] = (a1 >> 4 | a2 << 22) & M30; + r->v[2] = (a2 >> 8 | a3 << 18) & M30; + r->v[3] = (a3 >> 12 | a4 << 14) & M30; + r->v[4] = (a4 >> 16 | a5 << 10) & M30; + r->v[5] = (a5 >> 20 | a6 << 6) & M30; + r->v[6] = (a6 >> 24 | a7 << 2 + | a8 << 28) & M30; + r->v[7] = (a8 >> 2 | a9 << 24) & M30; + r->v[8] = a9 >> 6; +} - x88 = x44; - for (j=0; j<44; j++) { - secp256k1_fe_sqr(&x88, &x88); - } - secp256k1_fe_mul(&x88, &x88, &x44); +static const secp256k1_modinv32_modinfo secp256k1_const_modinfo_fe = { + {{-0x3D1, -4, 0, 0, 0, 0, 0, 0, 65536}}, + 0x2DDACACFL +}; - x176 = x88; - for (j=0; j<88; j++) { - secp256k1_fe_sqr(&x176, &x176); - } - secp256k1_fe_mul(&x176, &x176, &x88); +static void secp256k1_fe_inv(secp256k1_fe *r, const secp256k1_fe *x) { + secp256k1_fe tmp; + secp256k1_modinv32_signed30 s; - x220 = x176; - for (j=0; j<44; j++) { - secp256k1_fe_sqr(&x220, &x220); - } - secp256k1_fe_mul(&x220, &x220, &x44); + tmp = *x; + secp256k1_fe_normalize(&tmp); + secp256k1_fe_to_signed30(&s, &tmp); + secp256k1_modinv32(&s, &secp256k1_const_modinfo_fe); + secp256k1_fe_from_signed30(r, &s); - x223 = x220; - for (j=0; j<3; j++) { - secp256k1_fe_sqr(&x223, &x223); - } - secp256k1_fe_mul(&x223, &x223, &x3); + VERIFY_CHECK(secp256k1_fe_normalizes_to_zero(r) == secp256k1_fe_normalizes_to_zero(&tmp)); +} - /* The final result is then assembled using a sliding window over the blocks. */ +static void secp256k1_fe_inv_var(secp256k1_fe *r, const secp256k1_fe *x) { + secp256k1_fe tmp; + secp256k1_modinv32_signed30 s; - t1 = x223; - for (j=0; j<23; j++) { - secp256k1_fe_sqr(&t1, &t1); - } - secp256k1_fe_mul(&t1, &t1, &x22); - for (j=0; j<5; j++) { - secp256k1_fe_sqr(&t1, &t1); - } - secp256k1_fe_mul(&t1, &t1, a); - for (j=0; j<3; j++) { - secp256k1_fe_sqr(&t1, &t1); - } - secp256k1_fe_mul(&t1, &t1, &x2); - for (j=0; j<2; j++) { - secp256k1_fe_sqr(&t1, &t1); - } - secp256k1_fe_mul(r, a, &t1); -} + tmp = *x; + secp256k1_fe_normalize_var(&tmp); + secp256k1_fe_to_signed30(&s, &tmp); + secp256k1_modinv32_var(&s, &secp256k1_const_modinfo_fe); + secp256k1_fe_from_signed30(r, &s); -static void secp256k1_fe_inv_var(secp256k1_fe *r, const secp256k1_fe *a) { -#if defined(USE_FIELD_INV_BUILTIN) - secp256k1_fe_inv(r, a); -#elif defined(USE_FIELD_INV_NUM) - secp256k1_num n, m; - static const secp256k1_fe negone = SECP256K1_FE_CONST( - 0xFFFFFFFFUL, 0xFFFFFFFFUL, 0xFFFFFFFFUL, 0xFFFFFFFFUL, - 0xFFFFFFFFUL, 0xFFFFFFFFUL, 0xFFFFFFFEUL, 0xFFFFFC2EUL - ); - /* secp256k1 field prime, value p defined in "Standards for Efficient Cryptography" (SEC2) 2.7.1. */ - static const unsigned char prime[32] = { - 0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF, - 0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF, - 0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF, - 0xFF,0xFF,0xFF,0xFE,0xFF,0xFF,0xFC,0x2F - }; - unsigned char b[32]; - int res; - secp256k1_fe c = *a; - secp256k1_fe_normalize_var(&c); - secp256k1_fe_get_b32(b, &c); - secp256k1_num_set_bin(&n, b, 32); - secp256k1_num_set_bin(&m, prime, 32); - secp256k1_num_mod_inverse(&n, &n, &m); - secp256k1_num_get_bin(b, 32, &n); - res = secp256k1_fe_set_b32(r, b); - (void)res; - VERIFY_CHECK(res); - /* Verify the result is the (unique) valid inverse using non-GMP code. */ - secp256k1_fe_mul(&c, &c, r); - secp256k1_fe_add(&c, &negone); - CHECK(secp256k1_fe_normalizes_to_zero_var(&c)); -#else -#error "Please select field inverse implementation" -#endif + VERIFY_CHECK(secp256k1_fe_normalizes_to_zero(r) == secp256k1_fe_normalizes_to_zero(&tmp)); } #endif /* SECP256K1_FIELD_REPR_IMPL_H */ diff --git a/src/secp256k1/src/field_5x52_impl.h b/src/secp256k1/src/field_5x52_impl.h --- a/src/secp256k1/src/field_5x52_impl.h +++ b/src/secp256k1/src/field_5x52_impl.h @@ -13,6 +13,7 @@ #include "util.h" #include "field.h" +#include "modinv64_impl.h" #if defined(USE_ASM_X86_64) #include "field_5x52_asm_impl.h" @@ -498,130 +499,79 @@ #endif } -static void secp256k1_fe_inv(secp256k1_fe *r, const secp256k1_fe *a) { - secp256k1_fe x2, x3, x6, x9, x11, x22, x44, x88, x176, x220, x223, t1; - int j; +static void secp256k1_fe_from_signed62(secp256k1_fe *r, const secp256k1_modinv64_signed62 *a) { + const uint64_t M52 = UINT64_MAX >> 12; + const uint64_t a0 = a->v[0], a1 = a->v[1], a2 = a->v[2], a3 = a->v[3], a4 = a->v[4]; - /** The binary representation of (p - 2) has 5 blocks of 1s, with lengths in - * { 1, 2, 22, 223 }. Use an addition chain to calculate 2^n - 1 for each block: - * [1], [2], 3, 6, 9, 11, [22], 44, 88, 176, 220, [223] + /* The output from secp256k1_modinv64{_var} should be normalized to range [0,modulus), and + * have limbs in [0,2^62). The modulus is < 2^256, so the top limb must be below 2^(256-62*4). */ + VERIFY_CHECK(a0 >> 62 == 0); + VERIFY_CHECK(a1 >> 62 == 0); + VERIFY_CHECK(a2 >> 62 == 0); + VERIFY_CHECK(a3 >> 62 == 0); + VERIFY_CHECK(a4 >> 8 == 0); + + r->n[0] = a0 & M52; + r->n[1] = (a0 >> 52 | a1 << 10) & M52; + r->n[2] = (a1 >> 42 | a2 << 20) & M52; + r->n[3] = (a2 >> 32 | a3 << 30) & M52; + r->n[4] = (a3 >> 22 | a4 << 40); - secp256k1_fe_sqr(&x2, a); - secp256k1_fe_mul(&x2, &x2, a); - - secp256k1_fe_sqr(&x3, &x2); - secp256k1_fe_mul(&x3, &x3, a); - - x6 = x3; - for (j=0; j<3; j++) { - secp256k1_fe_sqr(&x6, &x6); - } - secp256k1_fe_mul(&x6, &x6, &x3); - - x9 = x6; - for (j=0; j<3; j++) { - secp256k1_fe_sqr(&x9, &x9); - } - secp256k1_fe_mul(&x9, &x9, &x3); +#ifdef VERIFY + r->magnitude = 1; + r->normalized = 1; + secp256k1_fe_verify(r); +#endif +} - x11 = x9; - for (j=0; j<2; j++) { - secp256k1_fe_sqr(&x11, &x11); - } - secp256k1_fe_mul(&x11, &x11, &x2); +static void secp256k1_fe_to_signed62(secp256k1_modinv64_signed62 *r, const secp256k1_fe *a) { + const uint64_t M62 = UINT64_MAX >> 2; + const uint64_t a0 = a->n[0], a1 = a->n[1], a2 = a->n[2], a3 = a->n[3], a4 = a->n[4]; - x22 = x11; - for (j=0; j<11; j++) { - secp256k1_fe_sqr(&x22, &x22); - } - secp256k1_fe_mul(&x22, &x22, &x11); +#ifdef VERIFY + VERIFY_CHECK(a->normalized); +#endif - x44 = x22; - for (j=0; j<22; j++) { - secp256k1_fe_sqr(&x44, &x44); - } - secp256k1_fe_mul(&x44, &x44, &x22); + r->v[0] = (a0 | a1 << 52) & M62; + r->v[1] = (a1 >> 10 | a2 << 42) & M62; + r->v[2] = (a2 >> 20 | a3 << 32) & M62; + r->v[3] = (a3 >> 30 | a4 << 22) & M62; + r->v[4] = a4 >> 40; +} - x88 = x44; - for (j=0; j<44; j++) { - secp256k1_fe_sqr(&x88, &x88); - } - secp256k1_fe_mul(&x88, &x88, &x44); +static const secp256k1_modinv64_modinfo secp256k1_const_modinfo_fe = { + {{-0x1000003D1LL, 0, 0, 0, 256}}, + 0x27C7F6E22DDACACFLL +}; - x176 = x88; - for (j=0; j<88; j++) { - secp256k1_fe_sqr(&x176, &x176); - } - secp256k1_fe_mul(&x176, &x176, &x88); +static void secp256k1_fe_inv(secp256k1_fe *r, const secp256k1_fe *x) { + secp256k1_fe tmp; + secp256k1_modinv64_signed62 s; - x220 = x176; - for (j=0; j<44; j++) { - secp256k1_fe_sqr(&x220, &x220); - } - secp256k1_fe_mul(&x220, &x220, &x44); + tmp = *x; + secp256k1_fe_normalize(&tmp); + secp256k1_fe_to_signed62(&s, &tmp); + secp256k1_modinv64(&s, &secp256k1_const_modinfo_fe); + secp256k1_fe_from_signed62(r, &s); - x223 = x220; - for (j=0; j<3; j++) { - secp256k1_fe_sqr(&x223, &x223); - } - secp256k1_fe_mul(&x223, &x223, &x3); +#ifdef VERIFY + VERIFY_CHECK(secp256k1_fe_normalizes_to_zero(r) == secp256k1_fe_normalizes_to_zero(&tmp)); +#endif +} - /* The final result is then assembled using a sliding window over the blocks. */ +static void secp256k1_fe_inv_var(secp256k1_fe *r, const secp256k1_fe *x) { + secp256k1_fe tmp; + secp256k1_modinv64_signed62 s; - t1 = x223; - for (j=0; j<23; j++) { - secp256k1_fe_sqr(&t1, &t1); - } - secp256k1_fe_mul(&t1, &t1, &x22); - for (j=0; j<5; j++) { - secp256k1_fe_sqr(&t1, &t1); - } - secp256k1_fe_mul(&t1, &t1, a); - for (j=0; j<3; j++) { - secp256k1_fe_sqr(&t1, &t1); - } - secp256k1_fe_mul(&t1, &t1, &x2); - for (j=0; j<2; j++) { - secp256k1_fe_sqr(&t1, &t1); - } - secp256k1_fe_mul(r, a, &t1); -} + tmp = *x; + secp256k1_fe_normalize_var(&tmp); + secp256k1_fe_to_signed62(&s, &tmp); + secp256k1_modinv64_var(&s, &secp256k1_const_modinfo_fe); + secp256k1_fe_from_signed62(r, &s); -static void secp256k1_fe_inv_var(secp256k1_fe *r, const secp256k1_fe *a) { -#if defined(USE_FIELD_INV_BUILTIN) - secp256k1_fe_inv(r, a); -#elif defined(USE_FIELD_INV_NUM) - secp256k1_num n, m; - static const secp256k1_fe negone = SECP256K1_FE_CONST( - 0xFFFFFFFFUL, 0xFFFFFFFFUL, 0xFFFFFFFFUL, 0xFFFFFFFFUL, - 0xFFFFFFFFUL, 0xFFFFFFFFUL, 0xFFFFFFFEUL, 0xFFFFFC2EUL - ); - /* secp256k1 field prime, value p defined in "Standards for Efficient Cryptography" (SEC2) 2.7.1. */ - static const unsigned char prime[32] = { - 0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF, - 0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF, - 0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF, - 0xFF,0xFF,0xFF,0xFE,0xFF,0xFF,0xFC,0x2F - }; - unsigned char b[32]; - int res; - secp256k1_fe c = *a; - secp256k1_fe_normalize_var(&c); - secp256k1_fe_get_b32(b, &c); - secp256k1_num_set_bin(&n, b, 32); - secp256k1_num_set_bin(&m, prime, 32); - secp256k1_num_mod_inverse(&n, &n, &m); - secp256k1_num_get_bin(b, 32, &n); - res = secp256k1_fe_set_b32(r, b); - (void)res; - VERIFY_CHECK(res); - /* Verify the result is the (unique) valid inverse using non-GMP code. */ - secp256k1_fe_mul(&c, &c, r); - secp256k1_fe_add(&c, &negone); - CHECK(secp256k1_fe_normalizes_to_zero_var(&c)); -#else -#error "Please select field inverse implementation" +#ifdef VERIFY + VERIFY_CHECK(secp256k1_fe_normalizes_to_zero(r) == secp256k1_fe_normalizes_to_zero(&tmp)); #endif } diff --git a/src/secp256k1/src/scalar_4x64_impl.h b/src/secp256k1/src/scalar_4x64_impl.h --- a/src/secp256k1/src/scalar_4x64_impl.h +++ b/src/secp256k1/src/scalar_4x64_impl.h @@ -7,6 +7,8 @@ #ifndef SECP256K1_SCALAR_REPR_IMPL_H #define SECP256K1_SCALAR_REPR_IMPL_H +#include "modinv64_impl.h" + /* Limbs of the secp256k1 order. */ #define SECP256K1_N_0 ((uint64_t)0xBFD25E8CD0364141ULL) #define SECP256K1_N_1 ((uint64_t)0xBAAEDCE6AF48A03BULL) @@ -955,178 +957,73 @@ r->d[3] = (r->d[3] & mask0) | (a->d[3] & mask1); } -static void secp256k1_scalar_inverse(secp256k1_scalar *r, const secp256k1_scalar *x) { - secp256k1_scalar *t; - int i; - /* First compute xN as x ^ (2^N - 1) for some values of N, - * and uM as x ^ M for some values of M. */ - secp256k1_scalar x2, x3, x6, x8, x14, x28, x56, x112, x126; - secp256k1_scalar u2, u5, u9, u11, u13; +static void secp256k1_scalar_from_signed62(secp256k1_scalar *r, const secp256k1_modinv64_signed62 *a) { + const uint64_t a0 = a->v[0], a1 = a->v[1], a2 = a->v[2], a3 = a->v[3], a4 = a->v[4]; - secp256k1_scalar_sqr(&u2, x); - secp256k1_scalar_mul(&x2, &u2, x); - secp256k1_scalar_mul(&u5, &u2, &x2); - secp256k1_scalar_mul(&x3, &u5, &u2); - secp256k1_scalar_mul(&u9, &x3, &u2); - secp256k1_scalar_mul(&u11, &u9, &u2); - secp256k1_scalar_mul(&u13, &u11, &u2); + /* The output from secp256k1_modinv64{_var} should be normalized to range [0,modulus), and + * have limbs in [0,2^62). The modulus is < 2^256, so the top limb must be below 2^(256-62*4). + */ + VERIFY_CHECK(a0 >> 62 == 0); + VERIFY_CHECK(a1 >> 62 == 0); + VERIFY_CHECK(a2 >> 62 == 0); + VERIFY_CHECK(a3 >> 62 == 0); + VERIFY_CHECK(a4 >> 8 == 0); - secp256k1_scalar_sqr(&x6, &u13); - secp256k1_scalar_sqr(&x6, &x6); - secp256k1_scalar_mul(&x6, &x6, &u11); + r->d[0] = a0 | a1 << 62; + r->d[1] = a1 >> 2 | a2 << 60; + r->d[2] = a2 >> 4 | a3 << 58; + r->d[3] = a3 >> 6 | a4 << 56; - secp256k1_scalar_sqr(&x8, &x6); - secp256k1_scalar_sqr(&x8, &x8); - secp256k1_scalar_mul(&x8, &x8, &x2); +#ifdef VERIFY + VERIFY_CHECK(secp256k1_scalar_check_overflow(r) == 0); +#endif +} - secp256k1_scalar_sqr(&x14, &x8); - for (i = 0; i < 5; i++) { - secp256k1_scalar_sqr(&x14, &x14); - } - secp256k1_scalar_mul(&x14, &x14, &x6); +static void secp256k1_scalar_to_signed62(secp256k1_modinv64_signed62 *r, const secp256k1_scalar *a) { + const uint64_t M62 = UINT64_MAX >> 2; + const uint64_t a0 = a->d[0], a1 = a->d[1], a2 = a->d[2], a3 = a->d[3]; - secp256k1_scalar_sqr(&x28, &x14); - for (i = 0; i < 13; i++) { - secp256k1_scalar_sqr(&x28, &x28); - } - secp256k1_scalar_mul(&x28, &x28, &x14); +#ifdef VERIFY + VERIFY_CHECK(secp256k1_scalar_check_overflow(a) == 0); +#endif - secp256k1_scalar_sqr(&x56, &x28); - for (i = 0; i < 27; i++) { - secp256k1_scalar_sqr(&x56, &x56); - } - secp256k1_scalar_mul(&x56, &x56, &x28); + r->v[0] = a0 & M62; + r->v[1] = (a0 >> 62 | a1 << 2) & M62; + r->v[2] = (a1 >> 60 | a2 << 4) & M62; + r->v[3] = (a2 >> 58 | a3 << 6) & M62; + r->v[4] = a3 >> 56; +} - secp256k1_scalar_sqr(&x112, &x56); - for (i = 0; i < 55; i++) { - secp256k1_scalar_sqr(&x112, &x112); - } - secp256k1_scalar_mul(&x112, &x112, &x56); +static const secp256k1_modinv64_modinfo secp256k1_const_modinfo_scalar = { + {{0x3FD25E8CD0364141LL, 0x2ABB739ABD2280EELL, -0x15LL, 0, 256}}, + 0x34F20099AA774EC1LL +}; - secp256k1_scalar_sqr(&x126, &x112); - for (i = 0; i < 13; i++) { - secp256k1_scalar_sqr(&x126, &x126); - } - secp256k1_scalar_mul(&x126, &x126, &x14); +static void secp256k1_scalar_inverse(secp256k1_scalar *r, const secp256k1_scalar *x) { + secp256k1_modinv64_signed62 s; +#ifdef VERIFY + int zero_in = secp256k1_scalar_is_zero(x); +#endif + secp256k1_scalar_to_signed62(&s, x); + secp256k1_modinv64(&s, &secp256k1_const_modinfo_scalar); + secp256k1_scalar_from_signed62(r, &s); - /* Then accumulate the final result (t starts at x126). */ - t = &x126; - for (i = 0; i < 3; i++) { - secp256k1_scalar_sqr(t, t); - } - secp256k1_scalar_mul(t, t, &u5); /* 101 */ - for (i = 0; i < 4; i++) { /* 0 */ - secp256k1_scalar_sqr(t, t); - } - secp256k1_scalar_mul(t, t, &x3); /* 111 */ - for (i = 0; i < 4; i++) { /* 0 */ - secp256k1_scalar_sqr(t, t); - } - secp256k1_scalar_mul(t, t, &u5); /* 101 */ - for (i = 0; i < 5; i++) { /* 0 */ - secp256k1_scalar_sqr(t, t); - } - secp256k1_scalar_mul(t, t, &u11); /* 1011 */ - for (i = 0; i < 4; i++) { - secp256k1_scalar_sqr(t, t); - } - secp256k1_scalar_mul(t, t, &u11); /* 1011 */ - for (i = 0; i < 4; i++) { /* 0 */ - secp256k1_scalar_sqr(t, t); - } - secp256k1_scalar_mul(t, t, &x3); /* 111 */ - for (i = 0; i < 5; i++) { /* 00 */ - secp256k1_scalar_sqr(t, t); - } - secp256k1_scalar_mul(t, t, &x3); /* 111 */ - for (i = 0; i < 6; i++) { /* 00 */ - secp256k1_scalar_sqr(t, t); - } - secp256k1_scalar_mul(t, t, &u13); /* 1101 */ - for (i = 0; i < 4; i++) { /* 0 */ - secp256k1_scalar_sqr(t, t); - } - secp256k1_scalar_mul(t, t, &u5); /* 101 */ - for (i = 0; i < 3; i++) { - secp256k1_scalar_sqr(t, t); - } - secp256k1_scalar_mul(t, t, &x3); /* 111 */ - for (i = 0; i < 5; i++) { /* 0 */ - secp256k1_scalar_sqr(t, t); - } - secp256k1_scalar_mul(t, t, &u9); /* 1001 */ - for (i = 0; i < 6; i++) { /* 000 */ - secp256k1_scalar_sqr(t, t); - } - secp256k1_scalar_mul(t, t, &u5); /* 101 */ - for (i = 0; i < 10; i++) { /* 0000000 */ - secp256k1_scalar_sqr(t, t); - } - secp256k1_scalar_mul(t, t, &x3); /* 111 */ - for (i = 0; i < 4; i++) { /* 0 */ - secp256k1_scalar_sqr(t, t); - } - secp256k1_scalar_mul(t, t, &x3); /* 111 */ - for (i = 0; i < 9; i++) { /* 0 */ - secp256k1_scalar_sqr(t, t); - } - secp256k1_scalar_mul(t, t, &x8); /* 11111111 */ - for (i = 0; i < 5; i++) { /* 0 */ - secp256k1_scalar_sqr(t, t); - } - secp256k1_scalar_mul(t, t, &u9); /* 1001 */ - for (i = 0; i < 6; i++) { /* 00 */ - secp256k1_scalar_sqr(t, t); - } - secp256k1_scalar_mul(t, t, &u11); /* 1011 */ - for (i = 0; i < 4; i++) { - secp256k1_scalar_sqr(t, t); - } - secp256k1_scalar_mul(t, t, &u13); /* 1101 */ - for (i = 0; i < 5; i++) { - secp256k1_scalar_sqr(t, t); - } - secp256k1_scalar_mul(t, t, &x2); /* 11 */ - for (i = 0; i < 6; i++) { /* 00 */ - secp256k1_scalar_sqr(t, t); - } - secp256k1_scalar_mul(t, t, &u13); /* 1101 */ - for (i = 0; i < 10; i++) { /* 000000 */ - secp256k1_scalar_sqr(t, t); - } - secp256k1_scalar_mul(t, t, &u13); /* 1101 */ - for (i = 0; i < 4; i++) { - secp256k1_scalar_sqr(t, t); - } - secp256k1_scalar_mul(t, t, &u9); /* 1001 */ - for (i = 0; i < 6; i++) { /* 00000 */ - secp256k1_scalar_sqr(t, t); - } - secp256k1_scalar_mul(t, t, x); /* 1 */ - for (i = 0; i < 8; i++) { /* 00 */ - secp256k1_scalar_sqr(t, t); - } - secp256k1_scalar_mul(r, t, &x6); /* 111111 */ +#ifdef VERIFY + VERIFY_CHECK(secp256k1_scalar_is_zero(r) == zero_in); +#endif } static void secp256k1_scalar_inverse_var(secp256k1_scalar *r, const secp256k1_scalar *x) { -#if defined(USE_SCALAR_INV_BUILTIN) - secp256k1_scalar_inverse(r, x); -#elif defined(USE_SCALAR_INV_NUM) - unsigned char b[32]; - secp256k1_num n, m; - secp256k1_scalar t = *x; - secp256k1_scalar_get_b32(b, &t); - secp256k1_num_set_bin(&n, b, 32); - secp256k1_scalar_order_get_num(&m); - secp256k1_num_mod_inverse(&n, &n, &m); - secp256k1_num_get_bin(b, 32, &n); - secp256k1_scalar_set_b32(r, b, NULL); - /* Verify that the inverse was computed correctly, without GMP code. */ - secp256k1_scalar_mul(&t, &t, r); - CHECK(secp256k1_scalar_is_one(&t)); -#else -#error "Please select scalar inverse implementation" + secp256k1_modinv64_signed62 s; +#ifdef VERIFY + int zero_in = secp256k1_scalar_is_zero(x); +#endif + secp256k1_scalar_to_signed62(&s, x); + secp256k1_modinv64_var(&s, &secp256k1_const_modinfo_scalar); + secp256k1_scalar_from_signed62(r, &s); + +#ifdef VERIFY + VERIFY_CHECK(secp256k1_scalar_is_zero(r) == zero_in); #endif } diff --git a/src/secp256k1/src/scalar_8x32_impl.h b/src/secp256k1/src/scalar_8x32_impl.h --- a/src/secp256k1/src/scalar_8x32_impl.h +++ b/src/secp256k1/src/scalar_8x32_impl.h @@ -7,6 +7,8 @@ #ifndef SECP256K1_SCALAR_REPR_IMPL_H #define SECP256K1_SCALAR_REPR_IMPL_H +#include "modinv32_impl.h" + /* Limbs of the secp256k1 order. */ #define SECP256K1_N_0 ((uint32_t)0xD0364141UL) #define SECP256K1_N_1 ((uint32_t)0xBFD25E8CUL) @@ -731,178 +733,87 @@ r->d[7] = (r->d[7] & mask0) | (a->d[7] & mask1); } -static void secp256k1_scalar_inverse(secp256k1_scalar *r, const secp256k1_scalar *x) { - secp256k1_scalar *t; - int i; - /* First compute xN as x ^ (2^N - 1) for some values of N, - * and uM as x ^ M for some values of M. */ - secp256k1_scalar x2, x3, x6, x8, x14, x28, x56, x112, x126; - secp256k1_scalar u2, u5, u9, u11, u13; - - secp256k1_scalar_sqr(&u2, x); - secp256k1_scalar_mul(&x2, &u2, x); - secp256k1_scalar_mul(&u5, &u2, &x2); - secp256k1_scalar_mul(&x3, &u5, &u2); - secp256k1_scalar_mul(&u9, &x3, &u2); - secp256k1_scalar_mul(&u11, &u9, &u2); - secp256k1_scalar_mul(&u13, &u11, &u2); - - secp256k1_scalar_sqr(&x6, &u13); - secp256k1_scalar_sqr(&x6, &x6); - secp256k1_scalar_mul(&x6, &x6, &u11); - - secp256k1_scalar_sqr(&x8, &x6); - secp256k1_scalar_sqr(&x8, &x8); - secp256k1_scalar_mul(&x8, &x8, &x2); - - secp256k1_scalar_sqr(&x14, &x8); - for (i = 0; i < 5; i++) { - secp256k1_scalar_sqr(&x14, &x14); - } - secp256k1_scalar_mul(&x14, &x14, &x6); +static void secp256k1_scalar_from_signed30(secp256k1_scalar *r, const secp256k1_modinv32_signed30 *a) { + const uint32_t a0 = a->v[0], a1 = a->v[1], a2 = a->v[2], a3 = a->v[3], a4 = a->v[4], + a5 = a->v[5], a6 = a->v[6], a7 = a->v[7], a8 = a->v[8]; + + /* The output from secp256k1_modinv32{_var} should be normalized to range [0,modulus), and + * have limbs in [0,2^30). The modulus is < 2^256, so the top limb must be below 2^(256-30*8). + */ + VERIFY_CHECK(a0 >> 30 == 0); + VERIFY_CHECK(a1 >> 30 == 0); + VERIFY_CHECK(a2 >> 30 == 0); + VERIFY_CHECK(a3 >> 30 == 0); + VERIFY_CHECK(a4 >> 30 == 0); + VERIFY_CHECK(a5 >> 30 == 0); + VERIFY_CHECK(a6 >> 30 == 0); + VERIFY_CHECK(a7 >> 30 == 0); + VERIFY_CHECK(a8 >> 16 == 0); + + r->d[0] = a0 | a1 << 30; + r->d[1] = a1 >> 2 | a2 << 28; + r->d[2] = a2 >> 4 | a3 << 26; + r->d[3] = a3 >> 6 | a4 << 24; + r->d[4] = a4 >> 8 | a5 << 22; + r->d[5] = a5 >> 10 | a6 << 20; + r->d[6] = a6 >> 12 | a7 << 18; + r->d[7] = a7 >> 14 | a8 << 16; - secp256k1_scalar_sqr(&x28, &x14); - for (i = 0; i < 13; i++) { - secp256k1_scalar_sqr(&x28, &x28); - } - secp256k1_scalar_mul(&x28, &x28, &x14); +#ifdef VERIFY + VERIFY_CHECK(secp256k1_scalar_check_overflow(r) == 0); +#endif +} - secp256k1_scalar_sqr(&x56, &x28); - for (i = 0; i < 27; i++) { - secp256k1_scalar_sqr(&x56, &x56); - } - secp256k1_scalar_mul(&x56, &x56, &x28); +static void secp256k1_scalar_to_signed30(secp256k1_modinv32_signed30 *r, const secp256k1_scalar *a) { + const uint32_t M30 = UINT32_MAX >> 2; + const uint32_t a0 = a->d[0], a1 = a->d[1], a2 = a->d[2], a3 = a->d[3], + a4 = a->d[4], a5 = a->d[5], a6 = a->d[6], a7 = a->d[7]; - secp256k1_scalar_sqr(&x112, &x56); - for (i = 0; i < 55; i++) { - secp256k1_scalar_sqr(&x112, &x112); - } - secp256k1_scalar_mul(&x112, &x112, &x56); +#ifdef VERIFY + VERIFY_CHECK(secp256k1_scalar_check_overflow(a) == 0); +#endif - secp256k1_scalar_sqr(&x126, &x112); - for (i = 0; i < 13; i++) { - secp256k1_scalar_sqr(&x126, &x126); - } - secp256k1_scalar_mul(&x126, &x126, &x14); + r->v[0] = a0 & M30; + r->v[1] = (a0 >> 30 | a1 << 2) & M30; + r->v[2] = (a1 >> 28 | a2 << 4) & M30; + r->v[3] = (a2 >> 26 | a3 << 6) & M30; + r->v[4] = (a3 >> 24 | a4 << 8) & M30; + r->v[5] = (a4 >> 22 | a5 << 10) & M30; + r->v[6] = (a5 >> 20 | a6 << 12) & M30; + r->v[7] = (a6 >> 18 | a7 << 14) & M30; + r->v[8] = a7 >> 16; +} - /* Then accumulate the final result (t starts at x126). */ - t = &x126; - for (i = 0; i < 3; i++) { - secp256k1_scalar_sqr(t, t); - } - secp256k1_scalar_mul(t, t, &u5); /* 101 */ - for (i = 0; i < 4; i++) { /* 0 */ - secp256k1_scalar_sqr(t, t); - } - secp256k1_scalar_mul(t, t, &x3); /* 111 */ - for (i = 0; i < 4; i++) { /* 0 */ - secp256k1_scalar_sqr(t, t); - } - secp256k1_scalar_mul(t, t, &u5); /* 101 */ - for (i = 0; i < 5; i++) { /* 0 */ - secp256k1_scalar_sqr(t, t); - } - secp256k1_scalar_mul(t, t, &u11); /* 1011 */ - for (i = 0; i < 4; i++) { - secp256k1_scalar_sqr(t, t); - } - secp256k1_scalar_mul(t, t, &u11); /* 1011 */ - for (i = 0; i < 4; i++) { /* 0 */ - secp256k1_scalar_sqr(t, t); - } - secp256k1_scalar_mul(t, t, &x3); /* 111 */ - for (i = 0; i < 5; i++) { /* 00 */ - secp256k1_scalar_sqr(t, t); - } - secp256k1_scalar_mul(t, t, &x3); /* 111 */ - for (i = 0; i < 6; i++) { /* 00 */ - secp256k1_scalar_sqr(t, t); - } - secp256k1_scalar_mul(t, t, &u13); /* 1101 */ - for (i = 0; i < 4; i++) { /* 0 */ - secp256k1_scalar_sqr(t, t); - } - secp256k1_scalar_mul(t, t, &u5); /* 101 */ - for (i = 0; i < 3; i++) { - secp256k1_scalar_sqr(t, t); - } - secp256k1_scalar_mul(t, t, &x3); /* 111 */ - for (i = 0; i < 5; i++) { /* 0 */ - secp256k1_scalar_sqr(t, t); - } - secp256k1_scalar_mul(t, t, &u9); /* 1001 */ - for (i = 0; i < 6; i++) { /* 000 */ - secp256k1_scalar_sqr(t, t); - } - secp256k1_scalar_mul(t, t, &u5); /* 101 */ - for (i = 0; i < 10; i++) { /* 0000000 */ - secp256k1_scalar_sqr(t, t); - } - secp256k1_scalar_mul(t, t, &x3); /* 111 */ - for (i = 0; i < 4; i++) { /* 0 */ - secp256k1_scalar_sqr(t, t); - } - secp256k1_scalar_mul(t, t, &x3); /* 111 */ - for (i = 0; i < 9; i++) { /* 0 */ - secp256k1_scalar_sqr(t, t); - } - secp256k1_scalar_mul(t, t, &x8); /* 11111111 */ - for (i = 0; i < 5; i++) { /* 0 */ - secp256k1_scalar_sqr(t, t); - } - secp256k1_scalar_mul(t, t, &u9); /* 1001 */ - for (i = 0; i < 6; i++) { /* 00 */ - secp256k1_scalar_sqr(t, t); - } - secp256k1_scalar_mul(t, t, &u11); /* 1011 */ - for (i = 0; i < 4; i++) { - secp256k1_scalar_sqr(t, t); - } - secp256k1_scalar_mul(t, t, &u13); /* 1101 */ - for (i = 0; i < 5; i++) { - secp256k1_scalar_sqr(t, t); - } - secp256k1_scalar_mul(t, t, &x2); /* 11 */ - for (i = 0; i < 6; i++) { /* 00 */ - secp256k1_scalar_sqr(t, t); - } - secp256k1_scalar_mul(t, t, &u13); /* 1101 */ - for (i = 0; i < 10; i++) { /* 000000 */ - secp256k1_scalar_sqr(t, t); - } - secp256k1_scalar_mul(t, t, &u13); /* 1101 */ - for (i = 0; i < 4; i++) { - secp256k1_scalar_sqr(t, t); - } - secp256k1_scalar_mul(t, t, &u9); /* 1001 */ - for (i = 0; i < 6; i++) { /* 00000 */ - secp256k1_scalar_sqr(t, t); - } - secp256k1_scalar_mul(t, t, x); /* 1 */ - for (i = 0; i < 8; i++) { /* 00 */ - secp256k1_scalar_sqr(t, t); - } - secp256k1_scalar_mul(r, t, &x6); /* 111111 */ +static const secp256k1_modinv32_modinfo secp256k1_const_modinfo_scalar = { + {{0x10364141L, 0x3F497A33L, 0x348A03BBL, 0x2BB739ABL, -0x146L, 0, 0, 0, 65536}}, + 0x2A774EC1L +}; + +static void secp256k1_scalar_inverse(secp256k1_scalar *r, const secp256k1_scalar *x) { + secp256k1_modinv32_signed30 s; +#ifdef VERIFY + int zero_in = secp256k1_scalar_is_zero(x); +#endif + secp256k1_scalar_to_signed30(&s, x); + secp256k1_modinv32(&s, &secp256k1_const_modinfo_scalar); + secp256k1_scalar_from_signed30(r, &s); + +#ifdef VERIFY + VERIFY_CHECK(secp256k1_scalar_is_zero(r) == zero_in); +#endif } static void secp256k1_scalar_inverse_var(secp256k1_scalar *r, const secp256k1_scalar *x) { -#if defined(USE_SCALAR_INV_BUILTIN) - secp256k1_scalar_inverse(r, x); -#elif defined(USE_SCALAR_INV_NUM) - unsigned char b[32]; - secp256k1_num n, m; - secp256k1_scalar t = *x; - secp256k1_scalar_get_b32(b, &t); - secp256k1_num_set_bin(&n, b, 32); - secp256k1_scalar_order_get_num(&m); - secp256k1_num_mod_inverse(&n, &n, &m); - secp256k1_num_get_bin(b, 32, &n); - secp256k1_scalar_set_b32(r, b, NULL); - /* Verify that the inverse was computed correctly, without GMP code. */ - secp256k1_scalar_mul(&t, &t, r); - CHECK(secp256k1_scalar_is_one(&t)); -#else -#error "Please select scalar inverse implementation" + secp256k1_modinv32_signed30 s; +#ifdef VERIFY + int zero_in = secp256k1_scalar_is_zero(x); +#endif + secp256k1_scalar_to_signed30(&s, x); + secp256k1_modinv32_var(&s, &secp256k1_const_modinfo_scalar); + secp256k1_scalar_from_signed30(r, &s); + +#ifdef VERIFY + VERIFY_CHECK(secp256k1_scalar_is_zero(r) == zero_in); #endif }