Compiler Explorer

Source code

#include <stdint.h>

#define SECP256K1_INLINE inline
#define SECP256K1_SCALAR_VERIFY
#define VERIFY_CHECK(x) (void)(x)

typedef __uint128_t secp256k1_uint128;
typedef __int128_t secp256k1_int128;

typedef __uint128_t uint128_t;
typedef __int128_t int128_t;

#if !defined(UINT128_MAX) && defined(__SIZEOF_INT128__)
# define UINT128_MAX ((uint128_t)(-1))
# define INT128_MAX ((int128_t)(UINT128_MAX >> 1))
# define INT128_MIN (-INT128_MAX - 1)
/* No (U)INT128_C macros because compilers providing __int128 do not support 128-bit literals.  */
#endif

#ifndef SECP256K1_INT128_NATIVE_IMPL_H
#define SECP256K1_INT128_NATIVE_IMPL_H

static SECP256K1_INLINE void secp256k1_u128_load(secp256k1_uint128 *r, uint64_t hi, uint64_t lo) {
    *r = (((uint128_t)hi) << 64) + lo;
}

static SECP256K1_INLINE void secp256k1_u128_mul(secp256k1_uint128 *r, uint64_t a, uint64_t b) {
   *r = (uint128_t)a * b;
}

static SECP256K1_INLINE void secp256k1_u128_accum_mul(secp256k1_uint128 *r, uint64_t a, uint64_t b) {
   *r += (uint128_t)a * b;
}

static SECP256K1_INLINE void secp256k1_u128_accum_u64(secp256k1_uint128 *r, uint64_t a) {
   *r += a;
}

static SECP256K1_INLINE void secp256k1_u128_rshift(secp256k1_uint128 *r, unsigned int n) {
   VERIFY_CHECK(n < 128);
   *r >>= n;
}

static SECP256K1_INLINE uint64_t secp256k1_u128_to_u64(const secp256k1_uint128 *a) {
   return (uint64_t)(*a);
}

static SECP256K1_INLINE uint64_t secp256k1_u128_hi_u64(const secp256k1_uint128 *a) {
   return (uint64_t)(*a >> 64);
}

static SECP256K1_INLINE void secp256k1_u128_from_u64(secp256k1_uint128 *r, uint64_t a) {
   *r = a;
}

static SECP256K1_INLINE int secp256k1_u128_check_bits(const secp256k1_uint128 *r, unsigned int n) {
   VERIFY_CHECK(n < 128);
   return (*r >> n == 0);
}

static SECP256K1_INLINE void secp256k1_i128_load(secp256k1_int128 *r, int64_t hi, uint64_t lo) {
    *r = (((uint128_t)(uint64_t)hi) << 64) + lo;
}

static SECP256K1_INLINE void secp256k1_i128_mul(secp256k1_int128 *r, int64_t a, int64_t b) {
   *r = (int128_t)a * b;
}

static SECP256K1_INLINE void secp256k1_i128_accum_mul(secp256k1_int128 *r, int64_t a, int64_t b) {
   int128_t ab = (int128_t)a * b;
   VERIFY_CHECK(0 <= ab ? *r <= INT128_MAX - ab : INT128_MIN - ab <= *r);
   *r += ab;
}

static SECP256K1_INLINE void secp256k1_i128_det(secp256k1_int128 *r, int64_t a, int64_t b, int64_t c, int64_t d) {
   int128_t ad = (int128_t)a * d;
   int128_t bc = (int128_t)b * c;
   VERIFY_CHECK(0 <= bc ? INT128_MIN + bc <= ad : ad <= INT128_MAX + bc);
   *r = ad - bc;
}

static SECP256K1_INLINE void secp256k1_i128_rshift(secp256k1_int128 *r, unsigned int n) {
   VERIFY_CHECK(n < 128);
   *r >>= n;
}

static SECP256K1_INLINE uint64_t secp256k1_i128_to_u64(const secp256k1_int128 *a) {
   return (uint64_t)*a;
}

static SECP256K1_INLINE int64_t secp256k1_i128_to_i64(const secp256k1_int128 *a) {
   VERIFY_CHECK(INT64_MIN <= *a && *a <= INT64_MAX);
   return *a;
}

static SECP256K1_INLINE void secp256k1_i128_from_i64(secp256k1_int128 *r, int64_t a) {
   *r = a;
}

static SECP256K1_INLINE int secp256k1_i128_eq_var(const secp256k1_int128 *a, const secp256k1_int128 *b) {
   return *a == *b;
}

static SECP256K1_INLINE int secp256k1_i128_check_pow2(const secp256k1_int128 *r, unsigned int n, int sign) {
   VERIFY_CHECK(n < 127);
   VERIFY_CHECK(sign == 1 || sign == -1);
   return (*r == (int128_t)((uint128_t)sign << n));
}

#endif

/***********************************************************************
 * Copyright (c) 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_H
#define SECP256K1_SCALAR_REPR_H

#include <stdint.h>

/** A scalar modulo the group order of the secp256k1 curve. */
typedef struct {
    uint64_t d[4];
} secp256k1_scalar;

#define SECP256K1_SCALAR_CONST(d7, d6, d5, d4, d3, d2, d1, d0) {{((uint64_t)(d1)) << 32 | (d0), ((uint64_t)(d3)) << 32 | (d2), ((uint64_t)(d5)) << 32 | (d4), ((uint64_t)(d7)) << 32 | (d6)}}

#endif /* SECP256K1_SCALAR_REPR_H */

/***********************************************************************
 * 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 "checkmem.h"
//#include "int128.h"
//#include "modinv64_impl.h"
//#include "util.h"

/* Limbs of the secp256k1 order. */
#define SECP256K1_N_0 ((uint64_t)0xBFD25E8CD0364141ULL)
#define SECP256K1_N_1 ((uint64_t)0xBAAEDCE6AF48A03BULL)
#define SECP256K1_N_2 ((uint64_t)0xFFFFFFFFFFFFFFFEULL)
#define SECP256K1_N_3 ((uint64_t)0xFFFFFFFFFFFFFFFFULL)

/* Limbs of 2^256 minus the secp256k1 order. */
#define SECP256K1_N_C_0 (~SECP256K1_N_0 + 1)
#define SECP256K1_N_C_1 (~SECP256K1_N_1)
#define SECP256K1_N_C_2 (1)

/* Limbs of half the secp256k1 order. */
#define SECP256K1_N_H_0 ((uint64_t)0xDFE92F46681B20A0ULL)
#define SECP256K1_N_H_1 ((uint64_t)0x5D576E7357A4501DULL)
#define SECP256K1_N_H_2 ((uint64_t)0xFFFFFFFFFFFFFFFFULL)
#define SECP256K1_N_H_3 ((uint64_t)0x7FFFFFFFFFFFFFFFULL)

/* Inspired by the macros in OpenSSL's crypto/bn/asm/x86_64-gcc.c. */

/** Add a*b to the number defined by (c0,c1,c2). c2 must never overflow. */
#define muladd(a,b) { \
    uint64_t tl, th; \
    { \
        secp256k1_uint128 t; \
        secp256k1_u128_mul(&t, a, b); \
        th = secp256k1_u128_hi_u64(&t);  /* at most 0xFFFFFFFFFFFFFFFE */ \
        tl = secp256k1_u128_to_u64(&t); \
    } \
    c0 += tl;                 /* overflow is handled on the next line */ \
    th += (c0 < tl);          /* at most 0xFFFFFFFFFFFFFFFF */ \
    c1 += th;                 /* overflow is handled on the next line */ \
    c2 += (c1 < th);          /* never overflows by contract (verified in the next line) */ \
    VERIFY_CHECK((c1 >= th) || (c2 != 0)); \
}

/** Add a*b to the number defined by (c0,c1). c1 must never overflow. */
#define muladd_fast(a,b) { \
    uint64_t tl, th; \
    { \
        secp256k1_uint128 t; \
        secp256k1_u128_mul(&t, a, b); \
        th = secp256k1_u128_hi_u64(&t);  /* at most 0xFFFFFFFFFFFFFFFE */ \
        tl = secp256k1_u128_to_u64(&t); \
    } \
    c0 += tl;                 /* overflow is handled on the next line */ \
    th += (c0 < tl);          /* at most 0xFFFFFFFFFFFFFFFF */ \
    c1 += th;                 /* never overflows by contract (verified in the next line) */ \
    VERIFY_CHECK(c1 >= th); \
}

/** Add a to the number defined by (c0,c1,c2). c2 must never overflow. */
#define sumadd(a) { \
    unsigned int over; \
    c0 += (a);                  /* overflow is handled on the next line */ \
    over = (c0 < (a));         \
    c1 += over;                 /* overflow is handled on the next line */ \
    c2 += (c1 < over);          /* never overflows by contract */ \
}

/** Add a to the number defined by (c0,c1). c1 must never overflow, c2 must be zero. */
#define sumadd_fast(a) { \
    c0 += (a);                 /* overflow is handled on the next line */ \
    c1 += (c0 < (a));          /* never overflows by contract (verified the next line) */ \
    VERIFY_CHECK((c1 != 0) | (c0 >= (a))); \
    VERIFY_CHECK(c2 == 0); \
}

/** Extract the lowest 64 bits of (c0,c1,c2) into n, and left shift the number 64 bits. */
#define extract(n) { \
    (n) = c0; \
    c0 = c1; \
    c1 = c2; \
    c2 = 0; \
}

/** Extract the lowest 64 bits of (c0,c1,c2) into n, and left shift the number 64 bits. c2 is required to be zero. */
#define extract_fast(n) { \
    (n) = c0; \
    c0 = c1; \
    c1 = 0; \
    VERIFY_CHECK(c2 == 0); \
}

void secp256k1_scalar_reduce_512(secp256k1_scalar *r, const uint64_t *l) {
#ifdef USE_ASM_X86_64
    /* Reduce 512 bits into 385. */
    uint64_t m0, m1, m2, m3, m4, m5, m6;
    uint64_t p0, p1, p2, p3, p4;
    uint64_t c;

__asm__ __volatile__(
    /* Preload. */
    "movq 32(%%rsi), %%r11\n"
    "movq 40(%%rsi), %%r12\n"
    "movq 48(%%rsi), %%r13\n"
    "movq 56(%%rsi), %%r14\n"
    /* Initialize r8,r9,r10 */
    "movq 0(%%rsi), %%r8\n"
    "xorq %%r9, %%r9\n"
    "xorq %%r10, %%r10\n"
    /* (r8,r9) += n0 * c0 */
    "movq %8, %%rax\n"
    "mulq %%r11\n"
    "addq %%rax, %%r8\n"
    "adcq %%rdx, %%r9\n"
    /* extract m0 */
    "movq %%r8, %q0\n"
    "xorq %%r8, %%r8\n"
    /* (r9,r10) += l1 */
    "addq 8(%%rsi), %%r9\n"
    "adcq $0, %%r10\n"
    /* (r9,r10,r8) += n1 * c0 */
    "movq %8, %%rax\n"
    "mulq %%r12\n"
    "addq %%rax, %%r9\n"
    "adcq %%rdx, %%r10\n"
    "adcq $0, %%r8\n"
    /* (r9,r10,r8) += n0 * c1 */
    "movq %9, %%rax\n"
    "mulq %%r11\n"
    "addq %%rax, %%r9\n"
    "adcq %%rdx, %%r10\n"
    "adcq $0, %%r8\n"
    /* extract m1 */
    "movq %%r9, %q1\n"
    "xorq %%r9, %%r9\n"
    /* (r10,r8,r9) += l2 */
    "addq 16(%%rsi), %%r10\n"
    "adcq $0, %%r8\n"
    "adcq $0, %%r9\n"
    /* (r10,r8,r9) += n2 * c0 */
    "movq %8, %%rax\n"
    "mulq %%r13\n"
    "addq %%rax, %%r10\n"
    "adcq %%rdx, %%r8\n"
    "adcq $0, %%r9\n"
    /* (r10,r8,r9) += n1 * c1 */
    "movq %9, %%rax\n"
    "mulq %%r12\n"
    "addq %%rax, %%r10\n"
    "adcq %%rdx, %%r8\n"
    "adcq $0, %%r9\n"
    /* (r10,r8,r9) += n0 */
    "addq %%r11, %%r10\n"
    "adcq $0, %%r8\n"
    "adcq $0, %%r9\n"
    /* extract m2 */
    "movq %%r10, %q2\n"
    "xorq %%r10, %%r10\n"
    /* (r8,r9,r10) += l3 */
    "addq 24(%%rsi), %%r8\n"
    "adcq $0, %%r9\n"
    "adcq $0, %%r10\n"
    /* (r8,r9,r10) += n3 * c0 */
    "movq %8, %%rax\n"
    "mulq %%r14\n"
    "addq %%rax, %%r8\n"
    "adcq %%rdx, %%r9\n"
    "adcq $0, %%r10\n"
    /* (r8,r9,r10) += n2 * c1 */
    "movq %9, %%rax\n"
    "mulq %%r13\n"
    "addq %%rax, %%r8\n"
    "adcq %%rdx, %%r9\n"
    "adcq $0, %%r10\n"
    /* (r8,r9,r10) += n1 */
    "addq %%r12, %%r8\n"
    "adcq $0, %%r9\n"
    "adcq $0, %%r10\n"
    /* extract m3 */
    "movq %%r8, %q3\n"
    "xorq %%r8, %%r8\n"
    /* (r9,r10,r8) += n3 * c1 */
    "movq %9, %%rax\n"
    "mulq %%r14\n"
    "addq %%rax, %%r9\n"
    "adcq %%rdx, %%r10\n"
    "adcq $0, %%r8\n"
    /* (r9,r10,r8) += n2 */
    "addq %%r13, %%r9\n"
    "adcq $0, %%r10\n"
    "adcq $0, %%r8\n"
    /* extract m4 */
    "movq %%r9, %q4\n"
    /* (r10,r8) += n3 */
    "addq %%r14, %%r10\n"
    "adcq $0, %%r8\n"
    /* extract m5 */
    "movq %%r10, %q5\n"
    /* extract m6 */
    "movq %%r8, %q6\n"
    : "=&g"(m0), "=&g"(m1), "=&g"(m2), "=g"(m3), "=g"(m4), "=g"(m5), "=g"(m6)
    : "S"(l), "i"(SECP256K1_N_C_0), "i"(SECP256K1_N_C_1)
    : "rax", "rdx", "r8", "r9", "r10", "r11", "r12", "r13", "r14", "cc");

SECP256K1_CHECKMEM_MSAN_DEFINE(&m0, sizeof(m0));
    SECP256K1_CHECKMEM_MSAN_DEFINE(&m1, sizeof(m1));
    SECP256K1_CHECKMEM_MSAN_DEFINE(&m2, sizeof(m2));
    SECP256K1_CHECKMEM_MSAN_DEFINE(&m3, sizeof(m3));
    SECP256K1_CHECKMEM_MSAN_DEFINE(&m4, sizeof(m4));
    SECP256K1_CHECKMEM_MSAN_DEFINE(&m5, sizeof(m5));
    SECP256K1_CHECKMEM_MSAN_DEFINE(&m6, sizeof(m6));

/* Reduce 385 bits into 258. */
    __asm__ __volatile__(
    /* Preload */
    "movq %q9, %%r11\n"
    "movq %q10, %%r12\n"
    "movq %q11, %%r13\n"
    /* Initialize (r8,r9,r10) */
    "movq %q5, %%r8\n"
    "xorq %%r9, %%r9\n"
    "xorq %%r10, %%r10\n"
    /* (r8,r9) += m4 * c0 */
    "movq %12, %%rax\n"
    "mulq %%r11\n"
    "addq %%rax, %%r8\n"
    "adcq %%rdx, %%r9\n"
    /* extract p0 */
    "movq %%r8, %q0\n"
    "xorq %%r8, %%r8\n"
    /* (r9,r10) += m1 */
    "addq %q6, %%r9\n"
    "adcq $0, %%r10\n"
    /* (r9,r10,r8) += m5 * c0 */
    "movq %12, %%rax\n"
    "mulq %%r12\n"
    "addq %%rax, %%r9\n"
    "adcq %%rdx, %%r10\n"
    "adcq $0, %%r8\n"
    /* (r9,r10,r8) += m4 * c1 */
    "movq %13, %%rax\n"
    "mulq %%r11\n"
    "addq %%rax, %%r9\n"
    "adcq %%rdx, %%r10\n"
    "adcq $0, %%r8\n"
    /* extract p1 */
    "movq %%r9, %q1\n"
    "xorq %%r9, %%r9\n"
    /* (r10,r8,r9) += m2 */
    "addq %q7, %%r10\n"
    "adcq $0, %%r8\n"
    "adcq $0, %%r9\n"
    /* (r10,r8,r9) += m6 * c0 */
    "movq %12, %%rax\n"
    "mulq %%r13\n"
    "addq %%rax, %%r10\n"
    "adcq %%rdx, %%r8\n"
    "adcq $0, %%r9\n"
    /* (r10,r8,r9) += m5 * c1 */
    "movq %13, %%rax\n"
    "mulq %%r12\n"
    "addq %%rax, %%r10\n"
    "adcq %%rdx, %%r8\n"
    "adcq $0, %%r9\n"
    /* (r10,r8,r9) += m4 */
    "addq %%r11, %%r10\n"
    "adcq $0, %%r8\n"
    "adcq $0, %%r9\n"
    /* extract p2 */
    "movq %%r10, %q2\n"
    /* (r8,r9) += m3 */
    "addq %q8, %%r8\n"
    "adcq $0, %%r9\n"
    /* (r8,r9) += m6 * c1 */
    "movq %13, %%rax\n"
    "mulq %%r13\n"
    "addq %%rax, %%r8\n"
    "adcq %%rdx, %%r9\n"
    /* (r8,r9) += m5 */
    "addq %%r12, %%r8\n"
    "adcq $0, %%r9\n"
    /* extract p3 */
    "movq %%r8, %q3\n"
    /* (r9) += m6 */
    "addq %%r13, %%r9\n"
    /* extract p4 */
    "movq %%r9, %q4\n"
    : "=&g"(p0), "=&g"(p1), "=&g"(p2), "=g"(p3), "=g"(p4)
    : "g"(m0), "g"(m1), "g"(m2), "g"(m3), "g"(m4), "g"(m5), "g"(m6), "i"(SECP256K1_N_C_0), "i"(SECP256K1_N_C_1)
    : "rax", "rdx", "r8", "r9", "r10", "r11", "r12", "r13", "cc");

SECP256K1_CHECKMEM_MSAN_DEFINE(&p0, sizeof(p0));
    SECP256K1_CHECKMEM_MSAN_DEFINE(&p1, sizeof(p1));
    SECP256K1_CHECKMEM_MSAN_DEFINE(&p2, sizeof(p2));
    SECP256K1_CHECKMEM_MSAN_DEFINE(&p3, sizeof(p3));
    SECP256K1_CHECKMEM_MSAN_DEFINE(&p4, sizeof(p4));

/* Reduce 258 bits into 256. */
    __asm__ __volatile__(
    /* Preload */
    "movq %q5, %%r10\n"
    /* (rax,rdx) = p4 * c0 */
    "movq %7, %%rax\n"
    "mulq %%r10\n"
    /* (rax,rdx) += p0 */
    "addq %q1, %%rax\n"
    "adcq $0, %%rdx\n"
    /* extract r0 */
    "movq %%rax, 0(%q6)\n"
    /* Move to (r8,r9) */
    "movq %%rdx, %%r8\n"
    "xorq %%r9, %%r9\n"
    /* (r8,r9) += p1 */
    "addq %q2, %%r8\n"
    "adcq $0, %%r9\n"
    /* (r8,r9) += p4 * c1 */
    "movq %8, %%rax\n"
    "mulq %%r10\n"
    "addq %%rax, %%r8\n"
    "adcq %%rdx, %%r9\n"
    /* Extract r1 */
    "movq %%r8, 8(%q6)\n"
    "xorq %%r8, %%r8\n"
    /* (r9,r8) += p4 */
    "addq %%r10, %%r9\n"
    "adcq $0, %%r8\n"
    /* (r9,r8) += p2 */
    "addq %q3, %%r9\n"
    "adcq $0, %%r8\n"
    /* Extract r2 */
    "movq %%r9, 16(%q6)\n"
    "xorq %%r9, %%r9\n"
    /* (r8,r9) += p3 */
    "addq %q4, %%r8\n"
    "adcq $0, %%r9\n"
    /* Extract r3 */
    "movq %%r8, 24(%q6)\n"
    /* Extract c */
    "movq %%r9, %q0\n"
    : "=g"(c)
    : "g"(p0), "g"(p1), "g"(p2), "g"(p3), "g"(p4), "D"(r), "i"(SECP256K1_N_C_0), "i"(SECP256K1_N_C_1)
    : "rax", "rdx", "r8", "r9", "r10", "cc", "memory");

SECP256K1_CHECKMEM_MSAN_DEFINE(r, sizeof(*r));
    SECP256K1_CHECKMEM_MSAN_DEFINE(&c, sizeof(c));

#else
    secp256k1_uint128 c128;
    uint64_t c, c0, c1, c2;
    uint64_t n0 = l[4], n1 = l[5], n2 = l[6], n3 = l[7];
    uint64_t m0, m1, m2, m3, m4, m5;
    uint32_t m6;
    uint64_t p0, p1, p2, p3;
    uint32_t p4;

/* Reduce 512 bits into 385. */
    /* m[0..6] = l[0..3] + n[0..3] * SECP256K1_N_C. */
    c0 = l[0]; c1 = 0; c2 = 0;
    muladd_fast(n0, SECP256K1_N_C_0);
    extract_fast(m0);
    sumadd_fast(l[1]);
    muladd(n1, SECP256K1_N_C_0);
    muladd(n0, SECP256K1_N_C_1);
    extract(m1);
    sumadd(l[2]);
    muladd(n2, SECP256K1_N_C_0);
    muladd(n1, SECP256K1_N_C_1);
    sumadd(n0);
    extract(m2);
    sumadd(l[3]);
    muladd(n3, SECP256K1_N_C_0);
    muladd(n2, SECP256K1_N_C_1);
    sumadd(n1);
    extract(m3);
    muladd(n3, SECP256K1_N_C_1);
    sumadd(n2);
    extract(m4);
    sumadd_fast(n3);
    extract_fast(m5);
    VERIFY_CHECK(c0 <= 1);
    m6 = c0;

/* Reduce 385 bits into 258. */
    /* p[0..4] = m[0..3] + m[4..6] * SECP256K1_N_C. */
    c0 = m0; c1 = 0; c2 = 0;
    muladd_fast(m4, SECP256K1_N_C_0);
    extract_fast(p0);
    sumadd_fast(m1);
    muladd(m5, SECP256K1_N_C_0);
    muladd(m4, SECP256K1_N_C_1);
    extract(p1);
    sumadd(m2);
    muladd(m6, SECP256K1_N_C_0);
    muladd(m5, SECP256K1_N_C_1);
    sumadd(m4);
    extract(p2);
    sumadd_fast(m3);
    muladd_fast(m6, SECP256K1_N_C_1);
    sumadd_fast(m5);
    extract_fast(p3);
    p4 = c0 + m6;
    VERIFY_CHECK(p4 <= 2);

/* Reduce 258 bits into 256. */
    /* r[0..3] = p[0..3] + p[4] * SECP256K1_N_C. */
    secp256k1_u128_from_u64(&c128, p0);
    secp256k1_u128_accum_mul(&c128, SECP256K1_N_C_0, p4);
    r->d[0] = secp256k1_u128_to_u64(&c128); secp256k1_u128_rshift(&c128, 64);
    secp256k1_u128_accum_u64(&c128, p1);
    secp256k1_u128_accum_mul(&c128, SECP256K1_N_C_1, p4);
    r->d[1] = secp256k1_u128_to_u64(&c128); secp256k1_u128_rshift(&c128, 64);
    secp256k1_u128_accum_u64(&c128, p2);
    secp256k1_u128_accum_u64(&c128, p4);
    r->d[2] = secp256k1_u128_to_u64(&c128); secp256k1_u128_rshift(&c128, 64);
    secp256k1_u128_accum_u64(&c128, p3);
    r->d[3] = secp256k1_u128_to_u64(&c128);
    c = secp256k1_u128_hi_u64(&c128);
#endif

/* Final reduction of r. */
    //secp256k1_scalar_reduce(r, c + secp256k1_scalar_check_overflow(r));
}

void secp256k1_scalar_mul_512(uint64_t *l8, const secp256k1_scalar *a, const secp256k1_scalar *b) {
#ifdef USE_ASM_X86_64
    const uint64_t *pb = b->d;
    __asm__ __volatile__(
    /* Preload */
    "movq 0(%%rdi), %%r15\n"
    "movq 8(%%rdi), %%rbx\n"
    "movq 16(%%rdi), %%rcx\n"
    "movq 0(%%rdx), %%r11\n"
    "movq 8(%%rdx), %%r12\n"
    "movq 16(%%rdx), %%r13\n"
    "movq 24(%%rdx), %%r14\n"
    /* (rax,rdx) = a0 * b0 */
    "movq %%r15, %%rax\n"
    "mulq %%r11\n"
    /* Extract l8[0] */
    "movq %%rax, 0(%%rsi)\n"
    /* (r8,r9,r10) = (rdx) */
    "movq %%rdx, %%r8\n"
    "xorq %%r9, %%r9\n"
    "xorq %%r10, %%r10\n"
    /* (r8,r9,r10) += a0 * b1 */
    "movq %%r15, %%rax\n"
    "mulq %%r12\n"
    "addq %%rax, %%r8\n"
    "adcq %%rdx, %%r9\n"
    "adcq $0, %%r10\n"
    /* (r8,r9,r10) += a1 * b0 */
    "movq %%rbx, %%rax\n"
    "mulq %%r11\n"
    "addq %%rax, %%r8\n"
    "adcq %%rdx, %%r9\n"
    "adcq $0, %%r10\n"
    /* Extract l8[1] */
    "movq %%r8, 8(%%rsi)\n"
    "xorq %%r8, %%r8\n"
    /* (r9,r10,r8) += a0 * b2 */
    "movq %%r15, %%rax\n"
    "mulq %%r13\n"
    "addq %%rax, %%r9\n"
    "adcq %%rdx, %%r10\n"
    "adcq $0, %%r8\n"
    /* (r9,r10,r8) += a1 * b1 */
    "movq %%rbx, %%rax\n"
    "mulq %%r12\n"
    "addq %%rax, %%r9\n"
    "adcq %%rdx, %%r10\n"
    "adcq $0, %%r8\n"
    /* (r9,r10,r8) += a2 * b0 */
    "movq %%rcx, %%rax\n"
    "mulq %%r11\n"
    "addq %%rax, %%r9\n"
    "adcq %%rdx, %%r10\n"
    "adcq $0, %%r8\n"
    /* Extract l8[2] */
    "movq %%r9, 16(%%rsi)\n"
    "xorq %%r9, %%r9\n"
    /* (r10,r8,r9) += a0 * b3 */
    "movq %%r15, %%rax\n"
    "mulq %%r14\n"
    "addq %%rax, %%r10\n"
    "adcq %%rdx, %%r8\n"
    "adcq $0, %%r9\n"
    /* Preload a3 */
    "movq 24(%%rdi), %%r15\n"
    /* (r10,r8,r9) += a1 * b2 */
    "movq %%rbx, %%rax\n"
    "mulq %%r13\n"
    "addq %%rax, %%r10\n"
    "adcq %%rdx, %%r8\n"
    "adcq $0, %%r9\n"
    /* (r10,r8,r9) += a2 * b1 */
    "movq %%rcx, %%rax\n"
    "mulq %%r12\n"
    "addq %%rax, %%r10\n"
    "adcq %%rdx, %%r8\n"
    "adcq $0, %%r9\n"
    /* (r10,r8,r9) += a3 * b0 */
    "movq %%r15, %%rax\n"
    "mulq %%r11\n"
    "addq %%rax, %%r10\n"
    "adcq %%rdx, %%r8\n"
    "adcq $0, %%r9\n"
    /* Extract l8[3] */
    "movq %%r10, 24(%%rsi)\n"
    "xorq %%r10, %%r10\n"
    /* (r8,r9,r10) += a1 * b3 */
    "movq %%rbx, %%rax\n"
    "mulq %%r14\n"
    "addq %%rax, %%r8\n"
    "adcq %%rdx, %%r9\n"
    "adcq $0, %%r10\n"
    /* (r8,r9,r10) += a2 * b2 */
    "movq %%rcx, %%rax\n"
    "mulq %%r13\n"
    "addq %%rax, %%r8\n"
    "adcq %%rdx, %%r9\n"
    "adcq $0, %%r10\n"
    /* (r8,r9,r10) += a3 * b1 */
    "movq %%r15, %%rax\n"
    "mulq %%r12\n"
    "addq %%rax, %%r8\n"
    "adcq %%rdx, %%r9\n"
    "adcq $0, %%r10\n"
    /* Extract l8[4] */
    "movq %%r8, 32(%%rsi)\n"
    "xorq %%r8, %%r8\n"
    /* (r9,r10,r8) += a2 * b3 */
    "movq %%rcx, %%rax\n"
    "mulq %%r14\n"
    "addq %%rax, %%r9\n"
    "adcq %%rdx, %%r10\n"
    "adcq $0, %%r8\n"
    /* (r9,r10,r8) += a3 * b2 */
    "movq %%r15, %%rax\n"
    "mulq %%r13\n"
    "addq %%rax, %%r9\n"
    "adcq %%rdx, %%r10\n"
    "adcq $0, %%r8\n"
    /* Extract l8[5] */
    "movq %%r9, 40(%%rsi)\n"
    /* (r10,r8) += a3 * b3 */
    "movq %%r15, %%rax\n"
    "mulq %%r14\n"
    "addq %%rax, %%r10\n"
    "adcq %%rdx, %%r8\n"
    /* Extract l8[6] */
    "movq %%r10, 48(%%rsi)\n"
    /* Extract l8[7] */
    "movq %%r8, 56(%%rsi)\n"
    : "+d"(pb)
    : "S"(l8), "D"(a->d)
    : "rax", "rbx", "rcx", "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15", "cc", "memory");

SECP256K1_CHECKMEM_MSAN_DEFINE(l8, sizeof(*l8) * 8);

#else
    /* 160 bit accumulator. */
    uint64_t c0 = 0, c1 = 0;
    uint32_t c2 = 0;

/* l8[0..7] = a[0..3] * b[0..3]. */
    muladd_fast(a->d[0], b->d[0]);
    extract_fast(l8[0]);
    muladd(a->d[0], b->d[1]);
    muladd(a->d[1], b->d[0]);
    extract(l8[1]);
    muladd(a->d[0], b->d[2]);
    muladd(a->d[1], b->d[1]);
    muladd(a->d[2], b->d[0]);
    extract(l8[2]);
    muladd(a->d[0], b->d[3]);
    muladd(a->d[1], b->d[2]);
    muladd(a->d[2], b->d[1]);
    muladd(a->d[3], b->d[0]);
    extract(l8[3]);
    muladd(a->d[1], b->d[3]);
    muladd(a->d[2], b->d[2]);
    muladd(a->d[3], b->d[1]);
    extract(l8[4]);
    muladd(a->d[2], b->d[3]);
    muladd(a->d[3], b->d[2]);
    extract(l8[5]);
    muladd_fast(a->d[3], b->d[3]);
    extract_fast(l8[6]);
    VERIFY_CHECK(c1 == 0);
    l8[7] = c0;
#endif
}

#undef sumadd
#undef sumadd_fast
#undef muladd
#undef muladd_fast
#undef extract
#undef extract_fast

#endif /* SECP256K1_SCALAR_REPR_IMPL_H */