diff options
Diffstat (limited to 'src/scalar_4x64_impl.h')
-rw-r--r-- | src/scalar_4x64_impl.h | 252 |
1 files changed, 81 insertions, 171 deletions
diff --git a/src/scalar_4x64_impl.h b/src/scalar_4x64_impl.h index 73cbd5e18a..a1def26fca 100644 --- a/src/scalar_4x64_impl.h +++ b/src/scalar_4x64_impl.h @@ -1,12 +1,14 @@ -/********************************************************************** - * Copyright (c) 2013, 2014 Pieter Wuille * - * Distributed under the MIT software license, see the accompanying * - * file COPYING or http://www.opensource.org/licenses/mit-license.php.* - **********************************************************************/ +/*********************************************************************** + * Copyright (c) 2013, 2014 Pieter Wuille * + * Distributed under the MIT software license, see the accompanying * + * file COPYING or https://www.opensource.org/licenses/mit-license.php.* + ***********************************************************************/ #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) @@ -212,28 +214,6 @@ static int secp256k1_scalar_cond_negate(secp256k1_scalar *r, int flag) { VERIFY_CHECK(c1 >= th); \ } -/** Add 2*a*b to the number defined by (c0,c1,c2). c2 must never overflow. */ -#define muladd2(a,b) { \ - uint64_t tl, th, th2, tl2; \ - { \ - uint128_t t = (uint128_t)a * b; \ - th = t >> 64; /* at most 0xFFFFFFFFFFFFFFFE */ \ - tl = t; \ - } \ - th2 = th + th; /* at most 0xFFFFFFFFFFFFFFFE (in case th was 0x7FFFFFFFFFFFFFFF) */ \ - c2 += (th2 < th); /* never overflows by contract (verified the next line) */ \ - VERIFY_CHECK((th2 >= th) || (c2 != 0)); \ - tl2 = tl + tl; /* at most 0xFFFFFFFFFFFFFFFE (in case the lowest 63 bits of tl were 0x7FFFFFFFFFFFFFFF) */ \ - th2 += (tl2 < tl); /* at most 0xFFFFFFFFFFFFFFFF */ \ - c0 += tl2; /* overflow is handled on the next line */ \ - th2 += (c0 < tl2); /* second overflow is handled on the next line */ \ - c2 += (c0 < tl2) & (th2 == 0); /* never overflows by contract (verified the next line) */ \ - VERIFY_CHECK((c0 >= tl2) || (th2 != 0) || (c2 != 0)); \ - c1 += th2; /* overflow is handled on the next line */ \ - c2 += (c1 < th2); /* never overflows by contract (verified the next line) */ \ - VERIFY_CHECK((c1 >= th2) || (c2 != 0)); \ -} - /** Add a to the number defined by (c0,c1,c2). c2 must never overflow. */ #define sumadd(a) { \ unsigned int over; \ @@ -743,148 +723,10 @@ static void secp256k1_scalar_mul_512(uint64_t l[8], const secp256k1_scalar *a, c #endif } -static void secp256k1_scalar_sqr_512(uint64_t l[8], const secp256k1_scalar *a) { -#ifdef USE_ASM_X86_64 - __asm__ __volatile__( - /* Preload */ - "movq 0(%%rdi), %%r11\n" - "movq 8(%%rdi), %%r12\n" - "movq 16(%%rdi), %%r13\n" - "movq 24(%%rdi), %%r14\n" - /* (rax,rdx) = a0 * a0 */ - "movq %%r11, %%rax\n" - "mulq %%r11\n" - /* Extract l0 */ - "movq %%rax, 0(%%rsi)\n" - /* (r8,r9,r10) = (rdx,0) */ - "movq %%rdx, %%r8\n" - "xorq %%r9, %%r9\n" - "xorq %%r10, %%r10\n" - /* (r8,r9,r10) += 2 * a0 * a1 */ - "movq %%r11, %%rax\n" - "mulq %%r12\n" - "addq %%rax, %%r8\n" - "adcq %%rdx, %%r9\n" - "adcq $0, %%r10\n" - "addq %%rax, %%r8\n" - "adcq %%rdx, %%r9\n" - "adcq $0, %%r10\n" - /* Extract l1 */ - "movq %%r8, 8(%%rsi)\n" - "xorq %%r8, %%r8\n" - /* (r9,r10,r8) += 2 * a0 * a2 */ - "movq %%r11, %%rax\n" - "mulq %%r13\n" - "addq %%rax, %%r9\n" - "adcq %%rdx, %%r10\n" - "adcq $0, %%r8\n" - "addq %%rax, %%r9\n" - "adcq %%rdx, %%r10\n" - "adcq $0, %%r8\n" - /* (r9,r10,r8) += a1 * a1 */ - "movq %%r12, %%rax\n" - "mulq %%r12\n" - "addq %%rax, %%r9\n" - "adcq %%rdx, %%r10\n" - "adcq $0, %%r8\n" - /* Extract l2 */ - "movq %%r9, 16(%%rsi)\n" - "xorq %%r9, %%r9\n" - /* (r10,r8,r9) += 2 * a0 * a3 */ - "movq %%r11, %%rax\n" - "mulq %%r14\n" - "addq %%rax, %%r10\n" - "adcq %%rdx, %%r8\n" - "adcq $0, %%r9\n" - "addq %%rax, %%r10\n" - "adcq %%rdx, %%r8\n" - "adcq $0, %%r9\n" - /* (r10,r8,r9) += 2 * a1 * a2 */ - "movq %%r12, %%rax\n" - "mulq %%r13\n" - "addq %%rax, %%r10\n" - "adcq %%rdx, %%r8\n" - "adcq $0, %%r9\n" - "addq %%rax, %%r10\n" - "adcq %%rdx, %%r8\n" - "adcq $0, %%r9\n" - /* Extract l3 */ - "movq %%r10, 24(%%rsi)\n" - "xorq %%r10, %%r10\n" - /* (r8,r9,r10) += 2 * a1 * a3 */ - "movq %%r12, %%rax\n" - "mulq %%r14\n" - "addq %%rax, %%r8\n" - "adcq %%rdx, %%r9\n" - "adcq $0, %%r10\n" - "addq %%rax, %%r8\n" - "adcq %%rdx, %%r9\n" - "adcq $0, %%r10\n" - /* (r8,r9,r10) += a2 * a2 */ - "movq %%r13, %%rax\n" - "mulq %%r13\n" - "addq %%rax, %%r8\n" - "adcq %%rdx, %%r9\n" - "adcq $0, %%r10\n" - /* Extract l4 */ - "movq %%r8, 32(%%rsi)\n" - "xorq %%r8, %%r8\n" - /* (r9,r10,r8) += 2 * a2 * a3 */ - "movq %%r13, %%rax\n" - "mulq %%r14\n" - "addq %%rax, %%r9\n" - "adcq %%rdx, %%r10\n" - "adcq $0, %%r8\n" - "addq %%rax, %%r9\n" - "adcq %%rdx, %%r10\n" - "adcq $0, %%r8\n" - /* Extract l5 */ - "movq %%r9, 40(%%rsi)\n" - /* (r10,r8) += a3 * a3 */ - "movq %%r14, %%rax\n" - "mulq %%r14\n" - "addq %%rax, %%r10\n" - "adcq %%rdx, %%r8\n" - /* Extract l6 */ - "movq %%r10, 48(%%rsi)\n" - /* Extract l7 */ - "movq %%r8, 56(%%rsi)\n" - : - : "S"(l), "D"(a->d) - : "rax", "rdx", "r8", "r9", "r10", "r11", "r12", "r13", "r14", "cc", "memory"); -#else - /* 160 bit accumulator. */ - uint64_t c0 = 0, c1 = 0; - uint32_t c2 = 0; - - /* l[0..7] = a[0..3] * b[0..3]. */ - muladd_fast(a->d[0], a->d[0]); - extract_fast(l[0]); - muladd2(a->d[0], a->d[1]); - extract(l[1]); - muladd2(a->d[0], a->d[2]); - muladd(a->d[1], a->d[1]); - extract(l[2]); - muladd2(a->d[0], a->d[3]); - muladd2(a->d[1], a->d[2]); - extract(l[3]); - muladd2(a->d[1], a->d[3]); - muladd(a->d[2], a->d[2]); - extract(l[4]); - muladd2(a->d[2], a->d[3]); - extract(l[5]); - muladd_fast(a->d[3], a->d[3]); - extract_fast(l[6]); - VERIFY_CHECK(c1 == 0); - l[7] = c0; -#endif -} - #undef sumadd #undef sumadd_fast #undef muladd #undef muladd_fast -#undef muladd2 #undef extract #undef extract_fast @@ -906,12 +748,6 @@ static int secp256k1_scalar_shr_int(secp256k1_scalar *r, int n) { return ret; } -static void secp256k1_scalar_sqr(secp256k1_scalar *r, const secp256k1_scalar *a) { - uint64_t l[8]; - secp256k1_scalar_sqr_512(l, a); - secp256k1_scalar_reduce_512(r, l); -} - static void secp256k1_scalar_split_128(secp256k1_scalar *r1, secp256k1_scalar *r2, const secp256k1_scalar *k) { r1->d[0] = k->d[0]; r1->d[1] = k->d[1]; @@ -955,4 +791,78 @@ static SECP256K1_INLINE void secp256k1_scalar_cmov(secp256k1_scalar *r, const se r->d[3] = (r->d[3] & mask0) | (a->d[3] & mask1); } +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]; + + /* 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->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; + +#ifdef VERIFY + VERIFY_CHECK(secp256k1_scalar_check_overflow(r) == 0); +#endif +} + +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]; + +#ifdef VERIFY + VERIFY_CHECK(secp256k1_scalar_check_overflow(a) == 0); +#endif + + 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; +} + +static const secp256k1_modinv64_modinfo secp256k1_const_modinfo_scalar = { + {{0x3FD25E8CD0364141LL, 0x2ABB739ABD2280EELL, -0x15LL, 0, 256}}, + 0x34F20099AA774EC1LL +}; + +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); + +#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) { + 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 +} + +SECP256K1_INLINE static int secp256k1_scalar_is_even(const secp256k1_scalar *a) { + return !(a->d[0] & 1); +} + #endif /* SECP256K1_SCALAR_REPR_IMPL_H */ |