/* Autogenerated: '../fiat-crypto/src/ExtractionOCaml/word_by_word_montgomery' --static --use-value-barrier --inline-internal np256 64 0xffffffff00000000ffffffffffffffffbce6faada7179e84f3b9cac2fc632551 mul add opp from_montgomery to_montgomery one msat divstep_precomp divstep to_bytes from_bytes selectznz */
/* curve description: np256 */
/* machine_wordsize = 64 (from "64") */
/* requested operations: mul, add, opp, from_montgomery, to_montgomery, one, msat, divstep_precomp, divstep, to_bytes, from_bytes, selectznz */
/* m = 0xffffffff00000000ffffffffffffffffbce6faada7179e84f3b9cac2fc632551 (from "0xffffffff00000000ffffffffffffffffbce6faada7179e84f3b9cac2fc632551") */
/*                                                                    */
/* NOTE: In addition to the bounds specified above each function, all */
/*   functions synthesized for this Montgomery arithmetic require the */
/*   input to be strictly less than the prime modulus (m), and also   */
/*   require the input to be in the unique saturated representation.  */
/*   All functions also ensure that these two properties are true of  */
/*   return values.                                                   */
/*  */
/* Computed values: */
/*   eval z = z[0] + (z[1] << 64) + (z[2] << 128) + (z[3] << 192) */
/*   bytes_eval z = z[0] + (z[1] << 8) + (z[2] << 16) + (z[3] << 24) + (z[4] << 32) + (z[5] << 40) + (z[6] << 48) + (z[7] << 56) + (z[8] << 64) + (z[9] << 72) + (z[10] << 80) + (z[11] << 88) + (z[12] << 96) + (z[13] << 104) + (z[14] << 112) + (z[15] << 120) + (z[16] << 128) + (z[17] << 136) + (z[18] << 144) + (z[19] << 152) + (z[20] << 160) + (z[21] << 168) + (z[22] << 176) + (z[23] << 184) + (z[24] << 192) + (z[25] << 200) + (z[26] << 208) + (z[27] << 216) + (z[28] << 224) + (z[29] << 232) + (z[30] << 240) + (z[31] << 248) */
/*   twos_complement_eval z = let x1 := z[0] + (z[1] << 64) + (z[2] << 128) + (z[3] << 192) in */
/*                            if x1 & (2^256-1) < 2^255 then x1 & (2^256-1) else (x1 & (2^256-1)) - 2^256 */

#include <stdint.h>
typedef unsigned char fiat_np256_uint1;
typedef signed char fiat_np256_int1;
#if defined(__GNUC__) || defined(__clang__)
#  define FIAT_NP256_FIAT_EXTENSION __extension__
#  define FIAT_NP256_FIAT_INLINE __inline__
#else
#  define FIAT_NP256_FIAT_EXTENSION
#  define FIAT_NP256_FIAT_INLINE
#endif

FIAT_NP256_FIAT_EXTENSION typedef signed __int128 fiat_np256_int128;
FIAT_NP256_FIAT_EXTENSION typedef unsigned __int128 fiat_np256_uint128;

/* The type fiat_np256_montgomery_domain_field_element is a field element in the Montgomery domain. */
/* Bounds: [[0x0 ~> 0xffffffffffffffff], [0x0 ~> 0xffffffffffffffff], [0x0 ~> 0xffffffffffffffff], [0x0 ~> 0xffffffffffffffff]] */
typedef uint64_t fiat_np256_montgomery_domain_field_element[4];

/* The type fiat_np256_non_montgomery_domain_field_element is a field element NOT in the Montgomery domain. */
/* Bounds: [[0x0 ~> 0xffffffffffffffff], [0x0 ~> 0xffffffffffffffff], [0x0 ~> 0xffffffffffffffff], [0x0 ~> 0xffffffffffffffff]] */
typedef uint64_t fiat_np256_non_montgomery_domain_field_element[4];

#if (-1 & 3) != 3
#error "This code only works on a two's complement system"
#endif

#if !defined(FIAT_NP256_NO_ASM) && (defined(__GNUC__) || defined(__clang__))
static __inline__ uint64_t fiat_np256_value_barrier_u64(uint64_t a) {
  __asm__("" : "+r"(a) : /* no inputs */);
  return a;
}
#else
#  define fiat_np256_value_barrier_u64(x) (x)
#endif


/*
 * The function fiat_np256_addcarryx_u64 is an addition with carry.
 *
 * Postconditions:
 *   out1 = (arg1 + arg2 + arg3) mod 2^64
 *   out2 = ⌊(arg1 + arg2 + arg3) / 2^64⌋
 *
 * Input Bounds:
 *   arg1: [0x0 ~> 0x1]
 *   arg2: [0x0 ~> 0xffffffffffffffff]
 *   arg3: [0x0 ~> 0xffffffffffffffff]
 * Output Bounds:
 *   out1: [0x0 ~> 0xffffffffffffffff]
 *   out2: [0x0 ~> 0x1]
 */
static FIAT_NP256_FIAT_INLINE void fiat_np256_addcarryx_u64(uint64_t* out1, fiat_np256_uint1* out2, fiat_np256_uint1 arg1, uint64_t arg2, uint64_t arg3) {
  fiat_np256_uint128 x1;
  uint64_t x2;
  fiat_np256_uint1 x3;
  x1 = ((arg1 + (fiat_np256_uint128)arg2) + arg3);
  x2 = (uint64_t)(x1 & UINT64_C(0xffffffffffffffff));
  x3 = (fiat_np256_uint1)(x1 >> 64);
  *out1 = x2;
  *out2 = x3;
}

/*
 * The function fiat_np256_subborrowx_u64 is a subtraction with borrow.
 *
 * Postconditions:
 *   out1 = (-arg1 + arg2 + -arg3) mod 2^64
 *   out2 = -⌊(-arg1 + arg2 + -arg3) / 2^64⌋
 *
 * Input Bounds:
 *   arg1: [0x0 ~> 0x1]
 *   arg2: [0x0 ~> 0xffffffffffffffff]
 *   arg3: [0x0 ~> 0xffffffffffffffff]
 * Output Bounds:
 *   out1: [0x0 ~> 0xffffffffffffffff]
 *   out2: [0x0 ~> 0x1]
 */
static FIAT_NP256_FIAT_INLINE void fiat_np256_subborrowx_u64(uint64_t* out1, fiat_np256_uint1* out2, fiat_np256_uint1 arg1, uint64_t arg2, uint64_t arg3) {
  fiat_np256_int128 x1;
  fiat_np256_int1 x2;
  uint64_t x3;
  x1 = ((arg2 - (fiat_np256_int128)arg1) - arg3);
  x2 = (fiat_np256_int1)(x1 >> 64);
  x3 = (uint64_t)(x1 & UINT64_C(0xffffffffffffffff));
  *out1 = x3;
  *out2 = (fiat_np256_uint1)(0x0 - x2);
}

/*
 * The function fiat_np256_mulx_u64 is a multiplication, returning the full double-width result.
 *
 * Postconditions:
 *   out1 = (arg1 * arg2) mod 2^64
 *   out2 = ⌊arg1 * arg2 / 2^64⌋
 *
 * Input Bounds:
 *   arg1: [0x0 ~> 0xffffffffffffffff]
 *   arg2: [0x0 ~> 0xffffffffffffffff]
 * Output Bounds:
 *   out1: [0x0 ~> 0xffffffffffffffff]
 *   out2: [0x0 ~> 0xffffffffffffffff]
 */
static FIAT_NP256_FIAT_INLINE void fiat_np256_mulx_u64(uint64_t* out1, uint64_t* out2, uint64_t arg1, uint64_t arg2) {
  fiat_np256_uint128 x1;
  uint64_t x2;
  uint64_t x3;
  x1 = ((fiat_np256_uint128)arg1 * arg2);
  x2 = (uint64_t)(x1 & UINT64_C(0xffffffffffffffff));
  x3 = (uint64_t)(x1 >> 64);
  *out1 = x2;
  *out2 = x3;
}

/*
 * The function fiat_np256_cmovznz_u64 is a single-word conditional move.
 *
 * Postconditions:
 *   out1 = (if arg1 = 0 then arg2 else arg3)
 *
 * Input Bounds:
 *   arg1: [0x0 ~> 0x1]
 *   arg2: [0x0 ~> 0xffffffffffffffff]
 *   arg3: [0x0 ~> 0xffffffffffffffff]
 * Output Bounds:
 *   out1: [0x0 ~> 0xffffffffffffffff]
 */
static FIAT_NP256_FIAT_INLINE void fiat_np256_cmovznz_u64(uint64_t* out1, fiat_np256_uint1 arg1, uint64_t arg2, uint64_t arg3) {
  fiat_np256_uint1 x1;
  uint64_t x2;
  uint64_t x3;
  x1 = (!(!arg1));
  x2 = ((fiat_np256_int1)(0x0 - x1) & UINT64_C(0xffffffffffffffff));
  x3 = ((fiat_np256_value_barrier_u64(x2) & arg3) | (fiat_np256_value_barrier_u64((~x2)) & arg2));
  *out1 = x3;
}

/*
 * The function fiat_np256_mul multiplies two field elements in the Montgomery domain.
 *
 * Preconditions:
 *   0 ≤ eval arg1 < m
 *   0 ≤ eval arg2 < m
 * Postconditions:
 *   eval (from_montgomery out1) mod m = (eval (from_montgomery arg1) * eval (from_montgomery arg2)) mod m
 *   0 ≤ eval out1 < m
 *
 */
static void fiat_np256_mul(fiat_np256_montgomery_domain_field_element out1, const fiat_np256_montgomery_domain_field_element arg1, const fiat_np256_montgomery_domain_field_element arg2) {
  uint64_t x1;
  uint64_t x2;
  uint64_t x3;
  uint64_t x4;
  uint64_t x5;
  uint64_t x6;
  uint64_t x7;
  uint64_t x8;
  uint64_t x9;
  uint64_t x10;
  uint64_t x11;
  uint64_t x12;
  uint64_t x13;
  fiat_np256_uint1 x14;
  uint64_t x15;
  fiat_np256_uint1 x16;
  uint64_t x17;
  fiat_np256_uint1 x18;
  uint64_t x19;
  uint64_t x20;
  uint64_t x21;
  uint64_t x22;
  uint64_t x23;
  uint64_t x24;
  uint64_t x25;
  uint64_t x26;
  uint64_t x27;
  uint64_t x28;
  uint64_t x29;
  uint64_t x30;
  fiat_np256_uint1 x31;
  uint64_t x32;
  fiat_np256_uint1 x33;
  uint64_t x34;
  fiat_np256_uint1 x35;
  uint64_t x36;
  uint64_t x37;
  fiat_np256_uint1 x38;
  uint64_t x39;
  fiat_np256_uint1 x40;
  uint64_t x41;
  fiat_np256_uint1 x42;
  uint64_t x43;
  fiat_np256_uint1 x44;
  uint64_t x45;
  fiat_np256_uint1 x46;
  uint64_t x47;
  uint64_t x48;
  uint64_t x49;
  uint64_t x50;
  uint64_t x51;
  uint64_t x52;
  uint64_t x53;
  uint64_t x54;
  uint64_t x55;
  fiat_np256_uint1 x56;
  uint64_t x57;
  fiat_np256_uint1 x58;
  uint64_t x59;
  fiat_np256_uint1 x60;
  uint64_t x61;
  uint64_t x62;
  fiat_np256_uint1 x63;
  uint64_t x64;
  fiat_np256_uint1 x65;
  uint64_t x66;
  fiat_np256_uint1 x67;
  uint64_t x68;
  fiat_np256_uint1 x69;
  uint64_t x70;
  fiat_np256_uint1 x71;
  uint64_t x72;
  uint64_t x73;
  uint64_t x74;
  uint64_t x75;
  uint64_t x76;
  uint64_t x77;
  uint64_t x78;
  uint64_t x79;
  uint64_t x80;
  uint64_t x81;
  uint64_t x82;
  fiat_np256_uint1 x83;
  uint64_t x84;
  fiat_np256_uint1 x85;
  uint64_t x86;
  fiat_np256_uint1 x87;
  uint64_t x88;
  uint64_t x89;
  fiat_np256_uint1 x90;
  uint64_t x91;
  fiat_np256_uint1 x92;
  uint64_t x93;
  fiat_np256_uint1 x94;
  uint64_t x95;
  fiat_np256_uint1 x96;
  uint64_t x97;
  fiat_np256_uint1 x98;
  uint64_t x99;
  uint64_t x100;
  uint64_t x101;
  uint64_t x102;
  uint64_t x103;
  uint64_t x104;
  uint64_t x105;
  uint64_t x106;
  uint64_t x107;
  uint64_t x108;
  fiat_np256_uint1 x109;
  uint64_t x110;
  fiat_np256_uint1 x111;
  uint64_t x112;
  fiat_np256_uint1 x113;
  uint64_t x114;
  uint64_t x115;
  fiat_np256_uint1 x116;
  uint64_t x117;
  fiat_np256_uint1 x118;
  uint64_t x119;
  fiat_np256_uint1 x120;
  uint64_t x121;
  fiat_np256_uint1 x122;
  uint64_t x123;
  fiat_np256_uint1 x124;
  uint64_t x125;
  uint64_t x126;
  uint64_t x127;
  uint64_t x128;
  uint64_t x129;
  uint64_t x130;
  uint64_t x131;
  uint64_t x132;
  uint64_t x133;
  uint64_t x134;
  uint64_t x135;
  fiat_np256_uint1 x136;
  uint64_t x137;
  fiat_np256_uint1 x138;
  uint64_t x139;
  fiat_np256_uint1 x140;
  uint64_t x141;
  uint64_t x142;
  fiat_np256_uint1 x143;
  uint64_t x144;
  fiat_np256_uint1 x145;
  uint64_t x146;
  fiat_np256_uint1 x147;
  uint64_t x148;
  fiat_np256_uint1 x149;
  uint64_t x150;
  fiat_np256_uint1 x151;
  uint64_t x152;
  uint64_t x153;
  uint64_t x154;
  uint64_t x155;
  uint64_t x156;
  uint64_t x157;
  uint64_t x158;
  uint64_t x159;
  uint64_t x160;
  uint64_t x161;
  fiat_np256_uint1 x162;
  uint64_t x163;
  fiat_np256_uint1 x164;
  uint64_t x165;
  fiat_np256_uint1 x166;
  uint64_t x167;
  uint64_t x168;
  fiat_np256_uint1 x169;
  uint64_t x170;
  fiat_np256_uint1 x171;
  uint64_t x172;
  fiat_np256_uint1 x173;
  uint64_t x174;
  fiat_np256_uint1 x175;
  uint64_t x176;
  fiat_np256_uint1 x177;
  uint64_t x178;
  uint64_t x179;
  uint64_t x180;
  uint64_t x181;
  uint64_t x182;
  uint64_t x183;
  uint64_t x184;
  uint64_t x185;
  uint64_t x186;
  uint64_t x187;
  uint64_t x188;
  fiat_np256_uint1 x189;
  uint64_t x190;
  fiat_np256_uint1 x191;
  uint64_t x192;
  fiat_np256_uint1 x193;
  uint64_t x194;
  uint64_t x195;
  fiat_np256_uint1 x196;
  uint64_t x197;
  fiat_np256_uint1 x198;
  uint64_t x199;
  fiat_np256_uint1 x200;
  uint64_t x201;
  fiat_np256_uint1 x202;
  uint64_t x203;
  fiat_np256_uint1 x204;
  uint64_t x205;
  uint64_t x206;
  fiat_np256_uint1 x207;
  uint64_t x208;
  fiat_np256_uint1 x209;
  uint64_t x210;
  fiat_np256_uint1 x211;
  uint64_t x212;
  fiat_np256_uint1 x213;
  uint64_t x214;
  fiat_np256_uint1 x215;
  uint64_t x216;
  uint64_t x217;
  uint64_t x218;
  uint64_t x219;
  x1 = (arg1[1]);
  x2 = (arg1[2]);
  x3 = (arg1[3]);
  x4 = (arg1[0]);
  fiat_np256_mulx_u64(&x5, &x6, x4, (arg2[3]));
  fiat_np256_mulx_u64(&x7, &x8, x4, (arg2[2]));
  fiat_np256_mulx_u64(&x9, &x10, x4, (arg2[1]));
  fiat_np256_mulx_u64(&x11, &x12, x4, (arg2[0]));
  fiat_np256_addcarryx_u64(&x13, &x14, 0x0, x12, x9);
  fiat_np256_addcarryx_u64(&x15, &x16, x14, x10, x7);
  fiat_np256_addcarryx_u64(&x17, &x18, x16, x8, x5);
  x19 = (x18 + x6);
  fiat_np256_mulx_u64(&x20, &x21, x11, UINT64_C(0xccd1c8aaee00bc4f));
  fiat_np256_mulx_u64(&x22, &x23, x20, UINT64_C(0xffffffff00000000));
  fiat_np256_mulx_u64(&x24, &x25, x20, UINT64_C(0xffffffffffffffff));
  fiat_np256_mulx_u64(&x26, &x27, x20, UINT64_C(0xbce6faada7179e84));
  fiat_np256_mulx_u64(&x28, &x29, x20, UINT64_C(0xf3b9cac2fc632551));
  fiat_np256_addcarryx_u64(&x30, &x31, 0x0, x29, x26);
  fiat_np256_addcarryx_u64(&x32, &x33, x31, x27, x24);
  fiat_np256_addcarryx_u64(&x34, &x35, x33, x25, x22);
  x36 = (x35 + x23);
  fiat_np256_addcarryx_u64(&x37, &x38, 0x0, x11, x28);
  fiat_np256_addcarryx_u64(&x39, &x40, x38, x13, x30);
  fiat_np256_addcarryx_u64(&x41, &x42, x40, x15, x32);
  fiat_np256_addcarryx_u64(&x43, &x44, x42, x17, x34);
  fiat_np256_addcarryx_u64(&x45, &x46, x44, x19, x36);
  fiat_np256_mulx_u64(&x47, &x48, x1, (arg2[3]));
  fiat_np256_mulx_u64(&x49, &x50, x1, (arg2[2]));
  fiat_np256_mulx_u64(&x51, &x52, x1, (arg2[1]));
  fiat_np256_mulx_u64(&x53, &x54, x1, (arg2[0]));
  fiat_np256_addcarryx_u64(&x55, &x56, 0x0, x54, x51);
  fiat_np256_addcarryx_u64(&x57, &x58, x56, x52, x49);
  fiat_np256_addcarryx_u64(&x59, &x60, x58, x50, x47);
  x61 = (x60 + x48);
  fiat_np256_addcarryx_u64(&x62, &x63, 0x0, x39, x53);
  fiat_np256_addcarryx_u64(&x64, &x65, x63, x41, x55);
  fiat_np256_addcarryx_u64(&x66, &x67, x65, x43, x57);
  fiat_np256_addcarryx_u64(&x68, &x69, x67, x45, x59);
  fiat_np256_addcarryx_u64(&x70, &x71, x69, x46, x61);
  fiat_np256_mulx_u64(&x72, &x73, x62, UINT64_C(0xccd1c8aaee00bc4f));
  fiat_np256_mulx_u64(&x74, &x75, x72, UINT64_C(0xffffffff00000000));
  fiat_np256_mulx_u64(&x76, &x77, x72, UINT64_C(0xffffffffffffffff));
  fiat_np256_mulx_u64(&x78, &x79, x72, UINT64_C(0xbce6faada7179e84));
  fiat_np256_mulx_u64(&x80, &x81, x72, UINT64_C(0xf3b9cac2fc632551));
  fiat_np256_addcarryx_u64(&x82, &x83, 0x0, x81, x78);
  fiat_np256_addcarryx_u64(&x84, &x85, x83, x79, x76);
  fiat_np256_addcarryx_u64(&x86, &x87, x85, x77, x74);
  x88 = (x87 + x75);
  fiat_np256_addcarryx_u64(&x89, &x90, 0x0, x62, x80);
  fiat_np256_addcarryx_u64(&x91, &x92, x90, x64, x82);
  fiat_np256_addcarryx_u64(&x93, &x94, x92, x66, x84);
  fiat_np256_addcarryx_u64(&x95, &x96, x94, x68, x86);
  fiat_np256_addcarryx_u64(&x97, &x98, x96, x70, x88);
  x99 = ((uint64_t)x98 + x71);
  fiat_np256_mulx_u64(&x100, &x101, x2, (arg2[3]));
  fiat_np256_mulx_u64(&x102, &x103, x2, (arg2[2]));
  fiat_np256_mulx_u64(&x104, &x105, x2, (arg2[1]));
  fiat_np256_mulx_u64(&x106, &x107, x2, (arg2[0]));
  fiat_np256_addcarryx_u64(&x108, &x109, 0x0, x107, x104);
  fiat_np256_addcarryx_u64(&x110, &x111, x109, x105, x102);
  fiat_np256_addcarryx_u64(&x112, &x113, x111, x103, x100);
  x114 = (x113 + x101);
  fiat_np256_addcarryx_u64(&x115, &x116, 0x0, x91, x106);
  fiat_np256_addcarryx_u64(&x117, &x118, x116, x93, x108);
  fiat_np256_addcarryx_u64(&x119, &x120, x118, x95, x110);
  fiat_np256_addcarryx_u64(&x121, &x122, x120, x97, x112);
  fiat_np256_addcarryx_u64(&x123, &x124, x122, x99, x114);
  fiat_np256_mulx_u64(&x125, &x126, x115, UINT64_C(0xccd1c8aaee00bc4f));
  fiat_np256_mulx_u64(&x127, &x128, x125, UINT64_C(0xffffffff00000000));
  fiat_np256_mulx_u64(&x129, &x130, x125, UINT64_C(0xffffffffffffffff));
  fiat_np256_mulx_u64(&x131, &x132, x125, UINT64_C(0xbce6faada7179e84));
  fiat_np256_mulx_u64(&x133, &x134, x125, UINT64_C(0xf3b9cac2fc632551));
  fiat_np256_addcarryx_u64(&x135, &x136, 0x0, x134, x131);
  fiat_np256_addcarryx_u64(&x137, &x138, x136, x132, x129);
  fiat_np256_addcarryx_u64(&x139, &x140, x138, x130, x127);
  x141 = (x140 + x128);
  fiat_np256_addcarryx_u64(&x142, &x143, 0x0, x115, x133);
  fiat_np256_addcarryx_u64(&x144, &x145, x143, x117, x135);
  fiat_np256_addcarryx_u64(&x146, &x147, x145, x119, x137);
  fiat_np256_addcarryx_u64(&x148, &x149, x147, x121, x139);
  fiat_np256_addcarryx_u64(&x150, &x151, x149, x123, x141);
  x152 = ((uint64_t)x151 + x124);
  fiat_np256_mulx_u64(&x153, &x154, x3, (arg2[3]));
  fiat_np256_mulx_u64(&x155, &x156, x3, (arg2[2]));
  fiat_np256_mulx_u64(&x157, &x158, x3, (arg2[1]));
  fiat_np256_mulx_u64(&x159, &x160, x3, (arg2[0]));
  fiat_np256_addcarryx_u64(&x161, &x162, 0x0, x160, x157);
  fiat_np256_addcarryx_u64(&x163, &x164, x162, x158, x155);
  fiat_np256_addcarryx_u64(&x165, &x166, x164, x156, x153);
  x167 = (x166 + x154);
  fiat_np256_addcarryx_u64(&x168, &x169, 0x0, x144, x159);
  fiat_np256_addcarryx_u64(&x170, &x171, x169, x146, x161);
  fiat_np256_addcarryx_u64(&x172, &x173, x171, x148, x163);
  fiat_np256_addcarryx_u64(&x174, &x175, x173, x150, x165);
  fiat_np256_addcarryx_u64(&x176, &x177, x175, x152, x167);
  fiat_np256_mulx_u64(&x178, &x179, x168, UINT64_C(0xccd1c8aaee00bc4f));
  fiat_np256_mulx_u64(&x180, &x181, x178, UINT64_C(0xffffffff00000000));
  fiat_np256_mulx_u64(&x182, &x183, x178, UINT64_C(0xffffffffffffffff));
  fiat_np256_mulx_u64(&x184, &x185, x178, UINT64_C(0xbce6faada7179e84));
  fiat_np256_mulx_u64(&x186, &x187, x178, UINT64_C(0xf3b9cac2fc632551));
  fiat_np256_addcarryx_u64(&x188, &x189, 0x0, x187, x184);
  fiat_np256_addcarryx_u64(&x190, &x191, x189, x185, x182);
  fiat_np256_addcarryx_u64(&x192, &x193, x191, x183, x180);
  x194 = (x193 + x181);
  fiat_np256_addcarryx_u64(&x195, &x196, 0x0, x168, x186);
  fiat_np256_addcarryx_u64(&x197, &x198, x196, x170, x188);
  fiat_np256_addcarryx_u64(&x199, &x200, x198, x172, x190);
  fiat_np256_addcarryx_u64(&x201, &x202, x200, x174, x192);
  fiat_np256_addcarryx_u64(&x203, &x204, x202, x176, x194);
  x205 = ((uint64_t)x204 + x177);
  fiat_np256_subborrowx_u64(&x206, &x207, 0x0, x197, UINT64_C(0xf3b9cac2fc632551));
  fiat_np256_subborrowx_u64(&x208, &x209, x207, x199, UINT64_C(0xbce6faada7179e84));
  fiat_np256_subborrowx_u64(&x210, &x211, x209, x201, UINT64_C(0xffffffffffffffff));
  fiat_np256_subborrowx_u64(&x212, &x213, x211, x203, UINT64_C(0xffffffff00000000));
  fiat_np256_subborrowx_u64(&x214, &x215, x213, x205, 0x0);
  fiat_np256_cmovznz_u64(&x216, x215, x206, x197);
  fiat_np256_cmovznz_u64(&x217, x215, x208, x199);
  fiat_np256_cmovznz_u64(&x218, x215, x210, x201);
  fiat_np256_cmovznz_u64(&x219, x215, x212, x203);
  out1[0] = x216;
  out1[1] = x217;
  out1[2] = x218;
  out1[3] = x219;
}

/*
 * The function fiat_np256_add adds two field elements in the Montgomery domain.
 *
 * Preconditions:
 *   0 ≤ eval arg1 < m
 *   0 ≤ eval arg2 < m
 * Postconditions:
 *   eval (from_montgomery out1) mod m = (eval (from_montgomery arg1) + eval (from_montgomery arg2)) mod m
 *   0 ≤ eval out1 < m
 *
 */
static void fiat_np256_add(fiat_np256_montgomery_domain_field_element out1, const fiat_np256_montgomery_domain_field_element arg1, const fiat_np256_montgomery_domain_field_element arg2) {
  uint64_t x1;
  fiat_np256_uint1 x2;
  uint64_t x3;
  fiat_np256_uint1 x4;
  uint64_t x5;
  fiat_np256_uint1 x6;
  uint64_t x7;
  fiat_np256_uint1 x8;
  uint64_t x9;
  fiat_np256_uint1 x10;
  uint64_t x11;
  fiat_np256_uint1 x12;
  uint64_t x13;
  fiat_np256_uint1 x14;
  uint64_t x15;
  fiat_np256_uint1 x16;
  uint64_t x17;
  fiat_np256_uint1 x18;
  uint64_t x19;
  uint64_t x20;
  uint64_t x21;
  uint64_t x22;
  fiat_np256_addcarryx_u64(&x1, &x2, 0x0, (arg1[0]), (arg2[0]));
  fiat_np256_addcarryx_u64(&x3, &x4, x2, (arg1[1]), (arg2[1]));
  fiat_np256_addcarryx_u64(&x5, &x6, x4, (arg1[2]), (arg2[2]));
  fiat_np256_addcarryx_u64(&x7, &x8, x6, (arg1[3]), (arg2[3]));
  fiat_np256_subborrowx_u64(&x9, &x10, 0x0, x1, UINT64_C(0xf3b9cac2fc632551));
  fiat_np256_subborrowx_u64(&x11, &x12, x10, x3, UINT64_C(0xbce6faada7179e84));
  fiat_np256_subborrowx_u64(&x13, &x14, x12, x5, UINT64_C(0xffffffffffffffff));
  fiat_np256_subborrowx_u64(&x15, &x16, x14, x7, UINT64_C(0xffffffff00000000));
  fiat_np256_subborrowx_u64(&x17, &x18, x16, x8, 0x0);
  fiat_np256_cmovznz_u64(&x19, x18, x9, x1);
  fiat_np256_cmovznz_u64(&x20, x18, x11, x3);
  fiat_np256_cmovznz_u64(&x21, x18, x13, x5);
  fiat_np256_cmovznz_u64(&x22, x18, x15, x7);
  out1[0] = x19;
  out1[1] = x20;
  out1[2] = x21;
  out1[3] = x22;
}

/*
 * The function fiat_np256_opp negates a field element in the Montgomery domain.
 *
 * Preconditions:
 *   0 ≤ eval arg1 < m
 * Postconditions:
 *   eval (from_montgomery out1) mod m = -eval (from_montgomery arg1) mod m
 *   0 ≤ eval out1 < m
 *
 */
static void fiat_np256_opp(fiat_np256_montgomery_domain_field_element out1, const fiat_np256_montgomery_domain_field_element arg1) {
  uint64_t x1;
  fiat_np256_uint1 x2;
  uint64_t x3;
  fiat_np256_uint1 x4;
  uint64_t x5;
  fiat_np256_uint1 x6;
  uint64_t x7;
  fiat_np256_uint1 x8;
  uint64_t x9;
  uint64_t x10;
  fiat_np256_uint1 x11;
  uint64_t x12;
  fiat_np256_uint1 x13;
  uint64_t x14;
  fiat_np256_uint1 x15;
  uint64_t x16;
  fiat_np256_uint1 x17;
  fiat_np256_subborrowx_u64(&x1, &x2, 0x0, 0x0, (arg1[0]));
  fiat_np256_subborrowx_u64(&x3, &x4, x2, 0x0, (arg1[1]));
  fiat_np256_subborrowx_u64(&x5, &x6, x4, 0x0, (arg1[2]));
  fiat_np256_subborrowx_u64(&x7, &x8, x6, 0x0, (arg1[3]));
  fiat_np256_cmovznz_u64(&x9, x8, 0x0, UINT64_C(0xffffffffffffffff));
  fiat_np256_addcarryx_u64(&x10, &x11, 0x0, x1, (x9 & UINT64_C(0xf3b9cac2fc632551)));
  fiat_np256_addcarryx_u64(&x12, &x13, x11, x3, (x9 & UINT64_C(0xbce6faada7179e84)));
  fiat_np256_addcarryx_u64(&x14, &x15, x13, x5, x9);
  fiat_np256_addcarryx_u64(&x16, &x17, x15, x7, (x9 & UINT64_C(0xffffffff00000000)));
  out1[0] = x10;
  out1[1] = x12;
  out1[2] = x14;
  out1[3] = x16;
}

/*
 * The function fiat_np256_from_montgomery translates a field element out of the Montgomery domain.
 *
 * Preconditions:
 *   0 ≤ eval arg1 < m
 * Postconditions:
 *   eval out1 mod m = (eval arg1 * ((2^64)⁻¹ mod m)^4) mod m
 *   0 ≤ eval out1 < m
 *
 */
static void fiat_np256_from_montgomery(fiat_np256_non_montgomery_domain_field_element out1, const fiat_np256_montgomery_domain_field_element arg1) {
  uint64_t x1;
  uint64_t x2;
  uint64_t x3;
  uint64_t x4;
  uint64_t x5;
  uint64_t x6;
  uint64_t x7;
  uint64_t x8;
  uint64_t x9;
  uint64_t x10;
  uint64_t x11;
  uint64_t x12;
  fiat_np256_uint1 x13;
  uint64_t x14;
  fiat_np256_uint1 x15;
  uint64_t x16;
  fiat_np256_uint1 x17;
  uint64_t x18;
  fiat_np256_uint1 x19;
  uint64_t x20;
  fiat_np256_uint1 x21;
  uint64_t x22;
  fiat_np256_uint1 x23;
  uint64_t x24;
  fiat_np256_uint1 x25;
  uint64_t x26;
  fiat_np256_uint1 x27;
  uint64_t x28;
  fiat_np256_uint1 x29;
  uint64_t x30;
  fiat_np256_uint1 x31;
  uint64_t x32;
  uint64_t x33;
  uint64_t x34;
  uint64_t x35;
  uint64_t x36;
  uint64_t x37;
  uint64_t x38;
  uint64_t x39;
  uint64_t x40;
  uint64_t x41;
  uint64_t x42;
  fiat_np256_uint1 x43;
  uint64_t x44;
  fiat_np256_uint1 x45;
  uint64_t x46;
  fiat_np256_uint1 x47;
  uint64_t x48;
  fiat_np256_uint1 x49;
  uint64_t x50;
  fiat_np256_uint1 x51;
  uint64_t x52;
  fiat_np256_uint1 x53;
  uint64_t x54;
  fiat_np256_uint1 x55;
  uint64_t x56;
  fiat_np256_uint1 x57;
  uint64_t x58;
  fiat_np256_uint1 x59;
  uint64_t x60;
  fiat_np256_uint1 x61;
  uint64_t x62;
  uint64_t x63;
  uint64_t x64;
  uint64_t x65;
  uint64_t x66;
  uint64_t x67;
  uint64_t x68;
  uint64_t x69;
  uint64_t x70;
  uint64_t x71;
  uint64_t x72;
  fiat_np256_uint1 x73;
  uint64_t x74;
  fiat_np256_uint1 x75;
  uint64_t x76;
  fiat_np256_uint1 x77;
  uint64_t x78;
  fiat_np256_uint1 x79;
  uint64_t x80;
  fiat_np256_uint1 x81;
  uint64_t x82;
  fiat_np256_uint1 x83;
  uint64_t x84;
  fiat_np256_uint1 x85;
  uint64_t x86;
  fiat_np256_uint1 x87;
  uint64_t x88;
  fiat_np256_uint1 x89;
  uint64_t x90;
  fiat_np256_uint1 x91;
  uint64_t x92;
  uint64_t x93;
  uint64_t x94;
  uint64_t x95;
  uint64_t x96;
  uint64_t x97;
  uint64_t x98;
  uint64_t x99;
  uint64_t x100;
  uint64_t x101;
  uint64_t x102;
  fiat_np256_uint1 x103;
  uint64_t x104;
  fiat_np256_uint1 x105;
  uint64_t x106;
  fiat_np256_uint1 x107;
  uint64_t x108;
  fiat_np256_uint1 x109;
  uint64_t x110;
  fiat_np256_uint1 x111;
  uint64_t x112;
  fiat_np256_uint1 x113;
  uint64_t x114;
  fiat_np256_uint1 x115;
  uint64_t x116;
  uint64_t x117;
  fiat_np256_uint1 x118;
  uint64_t x119;
  fiat_np256_uint1 x120;
  uint64_t x121;
  fiat_np256_uint1 x122;
  uint64_t x123;
  fiat_np256_uint1 x124;
  uint64_t x125;
  fiat_np256_uint1 x126;
  uint64_t x127;
  uint64_t x128;
  uint64_t x129;
  uint64_t x130;
  x1 = (arg1[0]);
  fiat_np256_mulx_u64(&x2, &x3, x1, UINT64_C(0xccd1c8aaee00bc4f));
  fiat_np256_mulx_u64(&x4, &x5, x2, UINT64_C(0xffffffff00000000));
  fiat_np256_mulx_u64(&x6, &x7, x2, UINT64_C(0xffffffffffffffff));
  fiat_np256_mulx_u64(&x8, &x9, x2, UINT64_C(0xbce6faada7179e84));
  fiat_np256_mulx_u64(&x10, &x11, x2, UINT64_C(0xf3b9cac2fc632551));
  fiat_np256_addcarryx_u64(&x12, &x13, 0x0, x11, x8);
  fiat_np256_addcarryx_u64(&x14, &x15, x13, x9, x6);
  fiat_np256_addcarryx_u64(&x16, &x17, x15, x7, x4);
  fiat_np256_addcarryx_u64(&x18, &x19, 0x0, x1, x10);
  fiat_np256_addcarryx_u64(&x20, &x21, x19, 0x0, x12);
  fiat_np256_addcarryx_u64(&x22, &x23, x21, 0x0, x14);
  fiat_np256_addcarryx_u64(&x24, &x25, x23, 0x0, x16);
  fiat_np256_addcarryx_u64(&x26, &x27, 0x0, x20, (arg1[1]));
  fiat_np256_addcarryx_u64(&x28, &x29, x27, x22, 0x0);
  fiat_np256_addcarryx_u64(&x30, &x31, x29, x24, 0x0);
  fiat_np256_mulx_u64(&x32, &x33, x26, UINT64_C(0xccd1c8aaee00bc4f));
  fiat_np256_mulx_u64(&x34, &x35, x32, UINT64_C(0xffffffff00000000));
  fiat_np256_mulx_u64(&x36, &x37, x32, UINT64_C(0xffffffffffffffff));
  fiat_np256_mulx_u64(&x38, &x39, x32, UINT64_C(0xbce6faada7179e84));
  fiat_np256_mulx_u64(&x40, &x41, x32, UINT64_C(0xf3b9cac2fc632551));
  fiat_np256_addcarryx_u64(&x42, &x43, 0x0, x41, x38);
  fiat_np256_addcarryx_u64(&x44, &x45, x43, x39, x36);
  fiat_np256_addcarryx_u64(&x46, &x47, x45, x37, x34);
  fiat_np256_addcarryx_u64(&x48, &x49, 0x0, x26, x40);
  fiat_np256_addcarryx_u64(&x50, &x51, x49, x28, x42);
  fiat_np256_addcarryx_u64(&x52, &x53, x51, x30, x44);
  fiat_np256_addcarryx_u64(&x54, &x55, x53, (x31 + (x25 + (x17 + x5))), x46);
  fiat_np256_addcarryx_u64(&x56, &x57, 0x0, x50, (arg1[2]));
  fiat_np256_addcarryx_u64(&x58, &x59, x57, x52, 0x0);
  fiat_np256_addcarryx_u64(&x60, &x61, x59, x54, 0x0);
  fiat_np256_mulx_u64(&x62, &x63, x56, UINT64_C(0xccd1c8aaee00bc4f));
  fiat_np256_mulx_u64(&x64, &x65, x62, UINT64_C(0xffffffff00000000));
  fiat_np256_mulx_u64(&x66, &x67, x62, UINT64_C(0xffffffffffffffff));
  fiat_np256_mulx_u64(&x68, &x69, x62, UINT64_C(0xbce6faada7179e84));
  fiat_np256_mulx_u64(&x70, &x71, x62, UINT64_C(0xf3b9cac2fc632551));
  fiat_np256_addcarryx_u64(&x72, &x73, 0x0, x71, x68);
  fiat_np256_addcarryx_u64(&x74, &x75, x73, x69, x66);
  fiat_np256_addcarryx_u64(&x76, &x77, x75, x67, x64);
  fiat_np256_addcarryx_u64(&x78, &x79, 0x0, x56, x70);
  fiat_np256_addcarryx_u64(&x80, &x81, x79, x58, x72);
  fiat_np256_addcarryx_u64(&x82, &x83, x81, x60, x74);
  fiat_np256_addcarryx_u64(&x84, &x85, x83, (x61 + (x55 + (x47 + x35))), x76);
  fiat_np256_addcarryx_u64(&x86, &x87, 0x0, x80, (arg1[3]));
  fiat_np256_addcarryx_u64(&x88, &x89, x87, x82, 0x0);
  fiat_np256_addcarryx_u64(&x90, &x91, x89, x84, 0x0);
  fiat_np256_mulx_u64(&x92, &x93, x86, UINT64_C(0xccd1c8aaee00bc4f));
  fiat_np256_mulx_u64(&x94, &x95, x92, UINT64_C(0xffffffff00000000));
  fiat_np256_mulx_u64(&x96, &x97, x92, UINT64_C(0xffffffffffffffff));
  fiat_np256_mulx_u64(&x98, &x99, x92, UINT64_C(0xbce6faada7179e84));
  fiat_np256_mulx_u64(&x100, &x101, x92, UINT64_C(0xf3b9cac2fc632551));
  fiat_np256_addcarryx_u64(&x102, &x103, 0x0, x101, x98);
  fiat_np256_addcarryx_u64(&x104, &x105, x103, x99, x96);
  fiat_np256_addcarryx_u64(&x106, &x107, x105, x97, x94);
  fiat_np256_addcarryx_u64(&x108, &x109, 0x0, x86, x100);
  fiat_np256_addcarryx_u64(&x110, &x111, x109, x88, x102);
  fiat_np256_addcarryx_u64(&x112, &x113, x111, x90, x104);
  fiat_np256_addcarryx_u64(&x114, &x115, x113, (x91 + (x85 + (x77 + x65))), x106);
  x116 = (x115 + (x107 + x95));
  fiat_np256_subborrowx_u64(&x117, &x118, 0x0, x110, UINT64_C(0xf3b9cac2fc632551));
  fiat_np256_subborrowx_u64(&x119, &x120, x118, x112, UINT64_C(0xbce6faada7179e84));
  fiat_np256_subborrowx_u64(&x121, &x122, x120, x114, UINT64_C(0xffffffffffffffff));
  fiat_np256_subborrowx_u64(&x123, &x124, x122, x116, UINT64_C(0xffffffff00000000));
  fiat_np256_subborrowx_u64(&x125, &x126, x124, 0x0, 0x0);
  fiat_np256_cmovznz_u64(&x127, x126, x117, x110);
  fiat_np256_cmovznz_u64(&x128, x126, x119, x112);
  fiat_np256_cmovznz_u64(&x129, x126, x121, x114);
  fiat_np256_cmovznz_u64(&x130, x126, x123, x116);
  out1[0] = x127;
  out1[1] = x128;
  out1[2] = x129;
  out1[3] = x130;
}

/*
 * The function fiat_np256_to_montgomery translates a field element into the Montgomery domain.
 *
 * Preconditions:
 *   0 ≤ eval arg1 < m
 * Postconditions:
 *   eval (from_montgomery out1) mod m = eval arg1 mod m
 *   0 ≤ eval out1 < m
 *
 */
static void fiat_np256_to_montgomery(fiat_np256_montgomery_domain_field_element out1, const fiat_np256_non_montgomery_domain_field_element arg1) {
  uint64_t x1;
  uint64_t x2;
  uint64_t x3;
  uint64_t x4;
  uint64_t x5;
  uint64_t x6;
  uint64_t x7;
  uint64_t x8;
  uint64_t x9;
  uint64_t x10;
  uint64_t x11;
  uint64_t x12;
  uint64_t x13;
  fiat_np256_uint1 x14;
  uint64_t x15;
  fiat_np256_uint1 x16;
  uint64_t x17;
  fiat_np256_uint1 x18;
  uint64_t x19;
  uint64_t x20;
  uint64_t x21;
  uint64_t x22;
  uint64_t x23;
  uint64_t x24;
  uint64_t x25;
  uint64_t x26;
  uint64_t x27;
  uint64_t x28;
  uint64_t x29;
  fiat_np256_uint1 x30;
  uint64_t x31;
  fiat_np256_uint1 x32;
  uint64_t x33;
  fiat_np256_uint1 x34;
  uint64_t x35;
  fiat_np256_uint1 x36;
  uint64_t x37;
  fiat_np256_uint1 x38;
  uint64_t x39;
  fiat_np256_uint1 x40;
  uint64_t x41;
  fiat_np256_uint1 x42;
  uint64_t x43;
  fiat_np256_uint1 x44;
  uint64_t x45;
  uint64_t x46;
  uint64_t x47;
  uint64_t x48;
  uint64_t x49;
  uint64_t x50;
  uint64_t x51;
  uint64_t x52;
  uint64_t x53;
  fiat_np256_uint1 x54;
  uint64_t x55;
  fiat_np256_uint1 x56;
  uint64_t x57;
  fiat_np256_uint1 x58;
  uint64_t x59;
  fiat_np256_uint1 x60;
  uint64_t x61;
  fiat_np256_uint1 x62;
  uint64_t x63;
  fiat_np256_uint1 x64;
  uint64_t x65;
  fiat_np256_uint1 x66;
  uint64_t x67;
  uint64_t x68;
  uint64_t x69;
  uint64_t x70;
  uint64_t x71;
  uint64_t x72;
  uint64_t x73;
  uint64_t x74;
  uint64_t x75;
  uint64_t x76;
  uint64_t x77;
  fiat_np256_uint1 x78;
  uint64_t x79;
  fiat_np256_uint1 x80;
  uint64_t x81;
  fiat_np256_uint1 x82;
  uint64_t x83;
  fiat_np256_uint1 x84;
  uint64_t x85;
  fiat_np256_uint1 x86;
  uint64_t x87;
  fiat_np256_uint1 x88;
  uint64_t x89;
  fiat_np256_uint1 x90;
  uint64_t x91;
  fiat_np256_uint1 x92;
  uint64_t x93;
  uint64_t x94;
  uint64_t x95;
  uint64_t x96;
  uint64_t x97;
  uint64_t x98;
  uint64_t x99;
  uint64_t x100;
  uint64_t x101;
  fiat_np256_uint1 x102;
  uint64_t x103;
  fiat_np256_uint1 x104;
  uint64_t x105;
  fiat_np256_uint1 x106;
  uint64_t x107;
  fiat_np256_uint1 x108;
  uint64_t x109;
  fiat_np256_uint1 x110;
  uint64_t x111;
  fiat_np256_uint1 x112;
  uint64_t x113;
  fiat_np256_uint1 x114;
  uint64_t x115;
  uint64_t x116;
  uint64_t x117;
  uint64_t x118;
  uint64_t x119;
  uint64_t x120;
  uint64_t x121;
  uint64_t x122;
  uint64_t x123;
  uint64_t x124;
  uint64_t x125;
  fiat_np256_uint1 x126;
  uint64_t x127;
  fiat_np256_uint1 x128;
  uint64_t x129;
  fiat_np256_uint1 x130;
  uint64_t x131;
  fiat_np256_uint1 x132;
  uint64_t x133;
  fiat_np256_uint1 x134;
  uint64_t x135;
  fiat_np256_uint1 x136;
  uint64_t x137;
  fiat_np256_uint1 x138;
  uint64_t x139;
  fiat_np256_uint1 x140;
  uint64_t x141;
  uint64_t x142;
  uint64_t x143;
  uint64_t x144;
  uint64_t x145;
  uint64_t x146;
  uint64_t x147;
  uint64_t x148;
  uint64_t x149;
  fiat_np256_uint1 x150;
  uint64_t x151;
  fiat_np256_uint1 x152;
  uint64_t x153;
  fiat_np256_uint1 x154;
  uint64_t x155;
  fiat_np256_uint1 x156;
  uint64_t x157;
  fiat_np256_uint1 x158;
  uint64_t x159;
  fiat_np256_uint1 x160;
  uint64_t x161;
  fiat_np256_uint1 x162;
  uint64_t x163;
  uint64_t x164;
  uint64_t x165;
  uint64_t x166;
  uint64_t x167;
  uint64_t x168;
  uint64_t x169;
  uint64_t x170;
  uint64_t x171;
  uint64_t x172;
  uint64_t x173;
  fiat_np256_uint1 x174;
  uint64_t x175;
  fiat_np256_uint1 x176;
  uint64_t x177;
  fiat_np256_uint1 x178;
  uint64_t x179;
  fiat_np256_uint1 x180;
  uint64_t x181;
  fiat_np256_uint1 x182;
  uint64_t x183;
  fiat_np256_uint1 x184;
  uint64_t x185;
  fiat_np256_uint1 x186;
  uint64_t x187;
  fiat_np256_uint1 x188;
  uint64_t x189;
  fiat_np256_uint1 x190;
  uint64_t x191;
  fiat_np256_uint1 x192;
  uint64_t x193;
  fiat_np256_uint1 x194;
  uint64_t x195;
  fiat_np256_uint1 x196;
  uint64_t x197;
  fiat_np256_uint1 x198;
  uint64_t x199;
  uint64_t x200;
  uint64_t x201;
  uint64_t x202;
  x1 = (arg1[1]);
  x2 = (arg1[2]);
  x3 = (arg1[3]);
  x4 = (arg1[0]);
  fiat_np256_mulx_u64(&x5, &x6, x4, UINT64_C(0x66e12d94f3d95620));
  fiat_np256_mulx_u64(&x7, &x8, x4, UINT64_C(0x2845b2392b6bec59));
  fiat_np256_mulx_u64(&x9, &x10, x4, UINT64_C(0x4699799c49bd6fa6));
  fiat_np256_mulx_u64(&x11, &x12, x4, UINT64_C(0x83244c95be79eea2));
  fiat_np256_addcarryx_u64(&x13, &x14, 0x0, x12, x9);
  fiat_np256_addcarryx_u64(&x15, &x16, x14, x10, x7);
  fiat_np256_addcarryx_u64(&x17, &x18, x16, x8, x5);
  fiat_np256_mulx_u64(&x19, &x20, x11, UINT64_C(0xccd1c8aaee00bc4f));
  fiat_np256_mulx_u64(&x21, &x22, x19, UINT64_C(0xffffffff00000000));
  fiat_np256_mulx_u64(&x23, &x24, x19, UINT64_C(0xffffffffffffffff));
  fiat_np256_mulx_u64(&x25, &x26, x19, UINT64_C(0xbce6faada7179e84));
  fiat_np256_mulx_u64(&x27, &x28, x19, UINT64_C(0xf3b9cac2fc632551));
  fiat_np256_addcarryx_u64(&x29, &x30, 0x0, x28, x25);
  fiat_np256_addcarryx_u64(&x31, &x32, x30, x26, x23);
  fiat_np256_addcarryx_u64(&x33, &x34, x32, x24, x21);
  fiat_np256_addcarryx_u64(&x35, &x36, 0x0, x11, x27);
  fiat_np256_addcarryx_u64(&x37, &x38, x36, x13, x29);
  fiat_np256_addcarryx_u64(&x39, &x40, x38, x15, x31);
  fiat_np256_addcarryx_u64(&x41, &x42, x40, x17, x33);
  fiat_np256_addcarryx_u64(&x43, &x44, x42, (x18 + x6), (x34 + x22));
  fiat_np256_mulx_u64(&x45, &x46, x1, UINT64_C(0x66e12d94f3d95620));
  fiat_np256_mulx_u64(&x47, &x48, x1, UINT64_C(0x2845b2392b6bec59));
  fiat_np256_mulx_u64(&x49, &x50, x1, UINT64_C(0x4699799c49bd6fa6));
  fiat_np256_mulx_u64(&x51, &x52, x1, UINT64_C(0x83244c95be79eea2));
  fiat_np256_addcarryx_u64(&x53, &x54, 0x0, x52, x49);
  fiat_np256_addcarryx_u64(&x55, &x56, x54, x50, x47);
  fiat_np256_addcarryx_u64(&x57, &x58, x56, x48, x45);
  fiat_np256_addcarryx_u64(&x59, &x60, 0x0, x37, x51);
  fiat_np256_addcarryx_u64(&x61, &x62, x60, x39, x53);
  fiat_np256_addcarryx_u64(&x63, &x64, x62, x41, x55);
  fiat_np256_addcarryx_u64(&x65, &x66, x64, x43, x57);
  fiat_np256_mulx_u64(&x67, &x68, x59, UINT64_C(0xccd1c8aaee00bc4f));
  fiat_np256_mulx_u64(&x69, &x70, x67, UINT64_C(0xffffffff00000000));
  fiat_np256_mulx_u64(&x71, &x72, x67, UINT64_C(0xffffffffffffffff));
  fiat_np256_mulx_u64(&x73, &x74, x67, UINT64_C(0xbce6faada7179e84));
  fiat_np256_mulx_u64(&x75, &x76, x67, UINT64_C(0xf3b9cac2fc632551));
  fiat_np256_addcarryx_u64(&x77, &x78, 0x0, x76, x73);
  fiat_np256_addcarryx_u64(&x79, &x80, x78, x74, x71);
  fiat_np256_addcarryx_u64(&x81, &x82, x80, x72, x69);
  fiat_np256_addcarryx_u64(&x83, &x84, 0x0, x59, x75);
  fiat_np256_addcarryx_u64(&x85, &x86, x84, x61, x77);
  fiat_np256_addcarryx_u64(&x87, &x88, x86, x63, x79);
  fiat_np256_addcarryx_u64(&x89, &x90, x88, x65, x81);
  fiat_np256_addcarryx_u64(&x91, &x92, x90, (((uint64_t)x66 + x44) + (x58 + x46)), (x82 + x70));
  fiat_np256_mulx_u64(&x93, &x94, x2, UINT64_C(0x66e12d94f3d95620));
  fiat_np256_mulx_u64(&x95, &x96, x2, UINT64_C(0x2845b2392b6bec59));
  fiat_np256_mulx_u64(&x97, &x98, x2, UINT64_C(0x4699799c49bd6fa6));
  fiat_np256_mulx_u64(&x99, &x100, x2, UINT64_C(0x83244c95be79eea2));
  fiat_np256_addcarryx_u64(&x101, &x102, 0x0, x100, x97);
  fiat_np256_addcarryx_u64(&x103, &x104, x102, x98, x95);
  fiat_np256_addcarryx_u64(&x105, &x106, x104, x96, x93);
  fiat_np256_addcarryx_u64(&x107, &x108, 0x0, x85, x99);
  fiat_np256_addcarryx_u64(&x109, &x110, x108, x87, x101);
  fiat_np256_addcarryx_u64(&x111, &x112, x110, x89, x103);
  fiat_np256_addcarryx_u64(&x113, &x114, x112, x91, x105);
  fiat_np256_mulx_u64(&x115, &x116, x107, UINT64_C(0xccd1c8aaee00bc4f));
  fiat_np256_mulx_u64(&x117, &x118, x115, UINT64_C(0xffffffff00000000));
  fiat_np256_mulx_u64(&x119, &x120, x115, UINT64_C(0xffffffffffffffff));
  fiat_np256_mulx_u64(&x121, &x122, x115, UINT64_C(0xbce6faada7179e84));
  fiat_np256_mulx_u64(&x123, &x124, x115, UINT64_C(0xf3b9cac2fc632551));
  fiat_np256_addcarryx_u64(&x125, &x126, 0x0, x124, x121);
  fiat_np256_addcarryx_u64(&x127, &x128, x126, x122, x119);
  fiat_np256_addcarryx_u64(&x129, &x130, x128, x120, x117);
  fiat_np256_addcarryx_u64(&x131, &x132, 0x0, x107, x123);
  fiat_np256_addcarryx_u64(&x133, &x134, x132, x109, x125);
  fiat_np256_addcarryx_u64(&x135, &x136, x134, x111, x127);
  fiat_np256_addcarryx_u64(&x137, &x138, x136, x113, x129);
  fiat_np256_addcarryx_u64(&x139, &x140, x138, (((uint64_t)x114 + x92) + (x106 + x94)), (x130 + x118));
  fiat_np256_mulx_u64(&x141, &x142, x3, UINT64_C(0x66e12d94f3d95620));
  fiat_np256_mulx_u64(&x143, &x144, x3, UINT64_C(0x2845b2392b6bec59));
  fiat_np256_mulx_u64(&x145, &x146, x3, UINT64_C(0x4699799c49bd6fa6));
  fiat_np256_mulx_u64(&x147, &x148, x3, UINT64_C(0x83244c95be79eea2));
  fiat_np256_addcarryx_u64(&x149, &x150, 0x0, x148, x145);
  fiat_np256_addcarryx_u64(&x151, &x152, x150, x146, x143);
  fiat_np256_addcarryx_u64(&x153, &x154, x152, x144, x141);
  fiat_np256_addcarryx_u64(&x155, &x156, 0x0, x133, x147);
  fiat_np256_addcarryx_u64(&x157, &x158, x156, x135, x149);
  fiat_np256_addcarryx_u64(&x159, &x160, x158, x137, x151);
  fiat_np256_addcarryx_u64(&x161, &x162, x160, x139, x153);
  fiat_np256_mulx_u64(&x163, &x164, x155, UINT64_C(0xccd1c8aaee00bc4f));
  fiat_np256_mulx_u64(&x165, &x166, x163, UINT64_C(0xffffffff00000000));
  fiat_np256_mulx_u64(&x167, &x168, x163, UINT64_C(0xffffffffffffffff));
  fiat_np256_mulx_u64(&x169, &x170, x163, UINT64_C(0xbce6faada7179e84));
  fiat_np256_mulx_u64(&x171, &x172, x163, UINT64_C(0xf3b9cac2fc632551));
  fiat_np256_addcarryx_u64(&x173, &x174, 0x0, x172, x169);
  fiat_np256_addcarryx_u64(&x175, &x176, x174, x170, x167);
  fiat_np256_addcarryx_u64(&x177, &x178, x176, x168, x165);
  fiat_np256_addcarryx_u64(&x179, &x180, 0x0, x155, x171);
  fiat_np256_addcarryx_u64(&x181, &x182, x180, x157, x173);
  fiat_np256_addcarryx_u64(&x183, &x184, x182, x159, x175);
  fiat_np256_addcarryx_u64(&x185, &x186, x184, x161, x177);
  fiat_np256_addcarryx_u64(&x187, &x188, x186, (((uint64_t)x162 + x140) + (x154 + x142)), (x178 + x166));
  fiat_np256_subborrowx_u64(&x189, &x190, 0x0, x181, UINT64_C(0xf3b9cac2fc632551));
  fiat_np256_subborrowx_u64(&x191, &x192, x190, x183, UINT64_C(0xbce6faada7179e84));
  fiat_np256_subborrowx_u64(&x193, &x194, x192, x185, UINT64_C(0xffffffffffffffff));
  fiat_np256_subborrowx_u64(&x195, &x196, x194, x187, UINT64_C(0xffffffff00000000));
  fiat_np256_subborrowx_u64(&x197, &x198, x196, x188, 0x0);
  fiat_np256_cmovznz_u64(&x199, x198, x189, x181);
  fiat_np256_cmovznz_u64(&x200, x198, x191, x183);
  fiat_np256_cmovznz_u64(&x201, x198, x193, x185);
  fiat_np256_cmovznz_u64(&x202, x198, x195, x187);
  out1[0] = x199;
  out1[1] = x200;
  out1[2] = x201;
  out1[3] = x202;
}

/*
 * The function fiat_np256_set_one returns the field element one in the Montgomery domain.
 *
 * Postconditions:
 *   eval (from_montgomery out1) mod m = 1 mod m
 *   0 ≤ eval out1 < m
 *
 */
static void fiat_np256_set_one(fiat_np256_montgomery_domain_field_element out1) {
  out1[0] = UINT64_C(0xc46353d039cdaaf);
  out1[1] = UINT64_C(0x4319055258e8617b);
  out1[2] = 0x0;
  out1[3] = UINT32_C(0xffffffff);
}

/*
 * The function fiat_np256_msat returns the saturated representation of the prime modulus.
 *
 * Postconditions:
 *   twos_complement_eval out1 = m
 *   0 ≤ eval out1 < m
 *
 * Output Bounds:
 *   out1: [[0x0 ~> 0xffffffffffffffff], [0x0 ~> 0xffffffffffffffff], [0x0 ~> 0xffffffffffffffff], [0x0 ~> 0xffffffffffffffff], [0x0 ~> 0xffffffffffffffff]]
 */
static void fiat_np256_msat(uint64_t out1[5]) {
  out1[0] = UINT64_C(0xf3b9cac2fc632551);
  out1[1] = UINT64_C(0xbce6faada7179e84);
  out1[2] = UINT64_C(0xffffffffffffffff);
  out1[3] = UINT64_C(0xffffffff00000000);
  out1[4] = 0x0;
}

/*
 * The function fiat_np256_divstep_precomp returns the precomputed value for Bernstein-Yang-inversion (in montgomery form).
 *
 * Postconditions:
 *   eval (from_montgomery out1) = ⌊(m - 1) / 2⌋^(if ⌊log2 m⌋ + 1 < 46 then ⌊(49 * (⌊log2 m⌋ + 1) + 80) / 17⌋ else ⌊(49 * (⌊log2 m⌋ + 1) + 57) / 17⌋)
 *   0 ≤ eval out1 < m
 *
 * Output Bounds:
 *   out1: [[0x0 ~> 0xffffffffffffffff], [0x0 ~> 0xffffffffffffffff], [0x0 ~> 0xffffffffffffffff], [0x0 ~> 0xffffffffffffffff]]
 */
static void fiat_np256_divstep_precomp(uint64_t out1[4]) {
  out1[0] = UINT64_C(0xd739262fb7fcfbb5);
  out1[1] = UINT64_C(0x8ac6f75d20074414);
  out1[2] = UINT64_C(0xc67428bfb5e3c256);
  out1[3] = UINT64_C(0x444962f2eda7aedf);
}

/*
 * The function fiat_np256_divstep computes a divstep.
 *
 * Preconditions:
 *   0 ≤ eval arg4 < m
 *   0 ≤ eval arg5 < m
 * Postconditions:
 *   out1 = (if 0 < arg1 ∧ (twos_complement_eval arg3) is odd then 1 - arg1 else 1 + arg1)
 *   twos_complement_eval out2 = (if 0 < arg1 ∧ (twos_complement_eval arg3) is odd then twos_complement_eval arg3 else twos_complement_eval arg2)
 *   twos_complement_eval out3 = (if 0 < arg1 ∧ (twos_complement_eval arg3) is odd then ⌊(twos_complement_eval arg3 - twos_complement_eval arg2) / 2⌋ else ⌊(twos_complement_eval arg3 + (twos_complement_eval arg3 mod 2) * twos_complement_eval arg2) / 2⌋)
 *   eval (from_montgomery out4) mod m = (if 0 < arg1 ∧ (twos_complement_eval arg3) is odd then (2 * eval (from_montgomery arg5)) mod m else (2 * eval (from_montgomery arg4)) mod m)
 *   eval (from_montgomery out5) mod m = (if 0 < arg1 ∧ (twos_complement_eval arg3) is odd then (eval (from_montgomery arg4) - eval (from_montgomery arg4)) mod m else (eval (from_montgomery arg5) + (twos_complement_eval arg3 mod 2) * eval (from_montgomery arg4)) mod m)
 *   0 ≤ eval out5 < m
 *   0 ≤ eval out5 < m
 *   0 ≤ eval out2 < m
 *   0 ≤ eval out3 < m
 *
 * Input Bounds:
 *   arg1: [0x0 ~> 0xffffffffffffffff]
 *   arg2: [[0x0 ~> 0xffffffffffffffff], [0x0 ~> 0xffffffffffffffff], [0x0 ~> 0xffffffffffffffff], [0x0 ~> 0xffffffffffffffff], [0x0 ~> 0xffffffffffffffff]]
 *   arg3: [[0x0 ~> 0xffffffffffffffff], [0x0 ~> 0xffffffffffffffff], [0x0 ~> 0xffffffffffffffff], [0x0 ~> 0xffffffffffffffff], [0x0 ~> 0xffffffffffffffff]]
 *   arg4: [[0x0 ~> 0xffffffffffffffff], [0x0 ~> 0xffffffffffffffff], [0x0 ~> 0xffffffffffffffff], [0x0 ~> 0xffffffffffffffff]]
 *   arg5: [[0x0 ~> 0xffffffffffffffff], [0x0 ~> 0xffffffffffffffff], [0x0 ~> 0xffffffffffffffff], [0x0 ~> 0xffffffffffffffff]]
 * Output Bounds:
 *   out1: [0x0 ~> 0xffffffffffffffff]
 *   out2: [[0x0 ~> 0xffffffffffffffff], [0x0 ~> 0xffffffffffffffff], [0x0 ~> 0xffffffffffffffff], [0x0 ~> 0xffffffffffffffff], [0x0 ~> 0xffffffffffffffff]]
 *   out3: [[0x0 ~> 0xffffffffffffffff], [0x0 ~> 0xffffffffffffffff], [0x0 ~> 0xffffffffffffffff], [0x0 ~> 0xffffffffffffffff], [0x0 ~> 0xffffffffffffffff]]
 *   out4: [[0x0 ~> 0xffffffffffffffff], [0x0 ~> 0xffffffffffffffff], [0x0 ~> 0xffffffffffffffff], [0x0 ~> 0xffffffffffffffff]]
 *   out5: [[0x0 ~> 0xffffffffffffffff], [0x0 ~> 0xffffffffffffffff], [0x0 ~> 0xffffffffffffffff], [0x0 ~> 0xffffffffffffffff]]
 */
static void fiat_np256_divstep(uint64_t* out1, uint64_t out2[5], uint64_t out3[5], uint64_t out4[4], uint64_t out5[4], uint64_t arg1, const uint64_t arg2[5], const uint64_t arg3[5], const uint64_t arg4[4], const uint64_t arg5[4]) {
  uint64_t x1;
  fiat_np256_uint1 x2;
  fiat_np256_uint1 x3;
  uint64_t x4;
  fiat_np256_uint1 x5;
  uint64_t x6;
  uint64_t x7;
  uint64_t x8;
  uint64_t x9;
  uint64_t x10;
  uint64_t x11;
  uint64_t x12;
  fiat_np256_uint1 x13;
  uint64_t x14;
  fiat_np256_uint1 x15;
  uint64_t x16;
  fiat_np256_uint1 x17;
  uint64_t x18;
  fiat_np256_uint1 x19;
  uint64_t x20;
  fiat_np256_uint1 x21;
  uint64_t x22;
  uint64_t x23;
  uint64_t x24;
  uint64_t x25;
  uint64_t x26;
  uint64_t x27;
  uint64_t x28;
  uint64_t x29;
  uint64_t x30;
  uint64_t x31;
  fiat_np256_uint1 x32;
  uint64_t x33;
  fiat_np256_uint1 x34;
  uint64_t x35;
  fiat_np256_uint1 x36;
  uint64_t x37;
  fiat_np256_uint1 x38;
  uint64_t x39;
  fiat_np256_uint1 x40;
  uint64_t x41;
  fiat_np256_uint1 x42;
  uint64_t x43;
  fiat_np256_uint1 x44;
  uint64_t x45;
  fiat_np256_uint1 x46;
  uint64_t x47;
  fiat_np256_uint1 x48;
  uint64_t x49;
  uint64_t x50;
  uint64_t x51;
  uint64_t x52;
  uint64_t x53;
  fiat_np256_uint1 x54;
  uint64_t x55;
  fiat_np256_uint1 x56;
  uint64_t x57;
  fiat_np256_uint1 x58;
  uint64_t x59;
  fiat_np256_uint1 x60;
  uint64_t x61;
  uint64_t x62;
  fiat_np256_uint1 x63;
  uint64_t x64;
  fiat_np256_uint1 x65;
  uint64_t x66;
  fiat_np256_uint1 x67;
  uint64_t x68;
  fiat_np256_uint1 x69;
  uint64_t x70;
  uint64_t x71;
  uint64_t x72;
  uint64_t x73;
  fiat_np256_uint1 x74;
  uint64_t x75;
  uint64_t x76;
  uint64_t x77;
  uint64_t x78;
  uint64_t x79;
  uint64_t x80;
  fiat_np256_uint1 x81;
  uint64_t x82;
  fiat_np256_uint1 x83;
  uint64_t x84;
  fiat_np256_uint1 x85;
  uint64_t x86;
  fiat_np256_uint1 x87;
  uint64_t x88;
  fiat_np256_uint1 x89;
  uint64_t x90;
  uint64_t x91;
  uint64_t x92;
  uint64_t x93;
  uint64_t x94;
  fiat_np256_uint1 x95;
  uint64_t x96;
  fiat_np256_uint1 x97;
  uint64_t x98;
  fiat_np256_uint1 x99;
  uint64_t x100;
  fiat_np256_uint1 x101;
  uint64_t x102;
  fiat_np256_uint1 x103;
  uint64_t x104;
  fiat_np256_uint1 x105;
  uint64_t x106;
  fiat_np256_uint1 x107;
  uint64_t x108;
  fiat_np256_uint1 x109;
  uint64_t x110;
  fiat_np256_uint1 x111;
  uint64_t x112;
  fiat_np256_uint1 x113;
  uint64_t x114;
  uint64_t x115;
  uint64_t x116;
  uint64_t x117;
  uint64_t x118;
  uint64_t x119;
  uint64_t x120;
  uint64_t x121;
  uint64_t x122;
  uint64_t x123;
  uint64_t x124;
  uint64_t x125;
  uint64_t x126;
  fiat_np256_addcarryx_u64(&x1, &x2, 0x0, (~arg1), 0x1);
  x3 = (fiat_np256_uint1)((fiat_np256_uint1)(x1 >> 63) & (fiat_np256_uint1)((arg3[0]) & 0x1));
  fiat_np256_addcarryx_u64(&x4, &x5, 0x0, (~arg1), 0x1);
  fiat_np256_cmovznz_u64(&x6, x3, arg1, x4);
  fiat_np256_cmovznz_u64(&x7, x3, (arg2[0]), (arg3[0]));
  fiat_np256_cmovznz_u64(&x8, x3, (arg2[1]), (arg3[1]));
  fiat_np256_cmovznz_u64(&x9, x3, (arg2[2]), (arg3[2]));
  fiat_np256_cmovznz_u64(&x10, x3, (arg2[3]), (arg3[3]));
  fiat_np256_cmovznz_u64(&x11, x3, (arg2[4]), (arg3[4]));
  fiat_np256_addcarryx_u64(&x12, &x13, 0x0, 0x1, (~(arg2[0])));
  fiat_np256_addcarryx_u64(&x14, &x15, x13, 0x0, (~(arg2[1])));
  fiat_np256_addcarryx_u64(&x16, &x17, x15, 0x0, (~(arg2[2])));
  fiat_np256_addcarryx_u64(&x18, &x19, x17, 0x0, (~(arg2[3])));
  fiat_np256_addcarryx_u64(&x20, &x21, x19, 0x0, (~(arg2[4])));
  fiat_np256_cmovznz_u64(&x22, x3, (arg3[0]), x12);
  fiat_np256_cmovznz_u64(&x23, x3, (arg3[1]), x14);
  fiat_np256_cmovznz_u64(&x24, x3, (arg3[2]), x16);
  fiat_np256_cmovznz_u64(&x25, x3, (arg3[3]), x18);
  fiat_np256_cmovznz_u64(&x26, x3, (arg3[4]), x20);
  fiat_np256_cmovznz_u64(&x27, x3, (arg4[0]), (arg5[0]));
  fiat_np256_cmovznz_u64(&x28, x3, (arg4[1]), (arg5[1]));
  fiat_np256_cmovznz_u64(&x29, x3, (arg4[2]), (arg5[2]));
  fiat_np256_cmovznz_u64(&x30, x3, (arg4[3]), (arg5[3]));
  fiat_np256_addcarryx_u64(&x31, &x32, 0x0, x27, x27);
  fiat_np256_addcarryx_u64(&x33, &x34, x32, x28, x28);
  fiat_np256_addcarryx_u64(&x35, &x36, x34, x29, x29);
  fiat_np256_addcarryx_u64(&x37, &x38, x36, x30, x30);
  fiat_np256_subborrowx_u64(&x39, &x40, 0x0, x31, UINT64_C(0xf3b9cac2fc632551));
  fiat_np256_subborrowx_u64(&x41, &x42, x40, x33, UINT64_C(0xbce6faada7179e84));
  fiat_np256_subborrowx_u64(&x43, &x44, x42, x35, UINT64_C(0xffffffffffffffff));
  fiat_np256_subborrowx_u64(&x45, &x46, x44, x37, UINT64_C(0xffffffff00000000));
  fiat_np256_subborrowx_u64(&x47, &x48, x46, x38, 0x0);
  x49 = (arg4[3]);
  x50 = (arg4[2]);
  x51 = (arg4[1]);
  x52 = (arg4[0]);
  fiat_np256_subborrowx_u64(&x53, &x54, 0x0, 0x0, x52);
  fiat_np256_subborrowx_u64(&x55, &x56, x54, 0x0, x51);
  fiat_np256_subborrowx_u64(&x57, &x58, x56, 0x0, x50);
  fiat_np256_subborrowx_u64(&x59, &x60, x58, 0x0, x49);
  fiat_np256_cmovznz_u64(&x61, x60, 0x0, UINT64_C(0xffffffffffffffff));
  fiat_np256_addcarryx_u64(&x62, &x63, 0x0, x53, (x61 & UINT64_C(0xf3b9cac2fc632551)));
  fiat_np256_addcarryx_u64(&x64, &x65, x63, x55, (x61 & UINT64_C(0xbce6faada7179e84)));
  fiat_np256_addcarryx_u64(&x66, &x67, x65, x57, x61);
  fiat_np256_addcarryx_u64(&x68, &x69, x67, x59, (x61 & UINT64_C(0xffffffff00000000)));
  fiat_np256_cmovznz_u64(&x70, x3, (arg5[0]), x62);
  fiat_np256_cmovznz_u64(&x71, x3, (arg5[1]), x64);
  fiat_np256_cmovznz_u64(&x72, x3, (arg5[2]), x66);
  fiat_np256_cmovznz_u64(&x73, x3, (arg5[3]), x68);
  x74 = (fiat_np256_uint1)(x22 & 0x1);
  fiat_np256_cmovznz_u64(&x75, x74, 0x0, x7);
  fiat_np256_cmovznz_u64(&x76, x74, 0x0, x8);
  fiat_np256_cmovznz_u64(&x77, x74, 0x0, x9);
  fiat_np256_cmovznz_u64(&x78, x74, 0x0, x10);
  fiat_np256_cmovznz_u64(&x79, x74, 0x0, x11);
  fiat_np256_addcarryx_u64(&x80, &x81, 0x0, x22, x75);
  fiat_np256_addcarryx_u64(&x82, &x83, x81, x23, x76);
  fiat_np256_addcarryx_u64(&x84, &x85, x83, x24, x77);
  fiat_np256_addcarryx_u64(&x86, &x87, x85, x25, x78);
  fiat_np256_addcarryx_u64(&x88, &x89, x87, x26, x79);
  fiat_np256_cmovznz_u64(&x90, x74, 0x0, x27);
  fiat_np256_cmovznz_u64(&x91, x74, 0x0, x28);
  fiat_np256_cmovznz_u64(&x92, x74, 0x0, x29);
  fiat_np256_cmovznz_u64(&x93, x74, 0x0, x30);
  fiat_np256_addcarryx_u64(&x94, &x95, 0x0, x70, x90);
  fiat_np256_addcarryx_u64(&x96, &x97, x95, x71, x91);
  fiat_np256_addcarryx_u64(&x98, &x99, x97, x72, x92);
  fiat_np256_addcarryx_u64(&x100, &x101, x99, x73, x93);
  fiat_np256_subborrowx_u64(&x102, &x103, 0x0, x94, UINT64_C(0xf3b9cac2fc632551));
  fiat_np256_subborrowx_u64(&x104, &x105, x103, x96, UINT64_C(0xbce6faada7179e84));
  fiat_np256_subborrowx_u64(&x106, &x107, x105, x98, UINT64_C(0xffffffffffffffff));
  fiat_np256_subborrowx_u64(&x108, &x109, x107, x100, UINT64_C(0xffffffff00000000));
  fiat_np256_subborrowx_u64(&x110, &x111, x109, x101, 0x0);
  fiat_np256_addcarryx_u64(&x112, &x113, 0x0, x6, 0x1);
  x114 = ((x80 >> 1) | ((x82 << 63) & UINT64_C(0xffffffffffffffff)));
  x115 = ((x82 >> 1) | ((x84 << 63) & UINT64_C(0xffffffffffffffff)));
  x116 = ((x84 >> 1) | ((x86 << 63) & UINT64_C(0xffffffffffffffff)));
  x117 = ((x86 >> 1) | ((x88 << 63) & UINT64_C(0xffffffffffffffff)));
  x118 = ((x88 & UINT64_C(0x8000000000000000)) | (x88 >> 1));
  fiat_np256_cmovznz_u64(&x119, x48, x39, x31);
  fiat_np256_cmovznz_u64(&x120, x48, x41, x33);
  fiat_np256_cmovznz_u64(&x121, x48, x43, x35);
  fiat_np256_cmovznz_u64(&x122, x48, x45, x37);
  fiat_np256_cmovznz_u64(&x123, x111, x102, x94);
  fiat_np256_cmovznz_u64(&x124, x111, x104, x96);
  fiat_np256_cmovznz_u64(&x125, x111, x106, x98);
  fiat_np256_cmovznz_u64(&x126, x111, x108, x100);
  *out1 = x112;
  out2[0] = x7;
  out2[1] = x8;
  out2[2] = x9;
  out2[3] = x10;
  out2[4] = x11;
  out3[0] = x114;
  out3[1] = x115;
  out3[2] = x116;
  out3[3] = x117;
  out3[4] = x118;
  out4[0] = x119;
  out4[1] = x120;
  out4[2] = x121;
  out4[3] = x122;
  out5[0] = x123;
  out5[1] = x124;
  out5[2] = x125;
  out5[3] = x126;
}

/*
 * The function fiat_np256_to_bytes serializes a field element NOT in the Montgomery domain to bytes in little-endian order.
 *
 * Preconditions:
 *   0 ≤ eval arg1 < m
 * Postconditions:
 *   out1 = map (λ x, ⌊((eval arg1 mod m) mod 2^(8 * (x + 1))) / 2^(8 * x)⌋) [0..31]
 *
 * Input Bounds:
 *   arg1: [[0x0 ~> 0xffffffffffffffff], [0x0 ~> 0xffffffffffffffff], [0x0 ~> 0xffffffffffffffff], [0x0 ~> 0xffffffffffffffff]]
 * Output Bounds:
 *   out1: [[0x0 ~> 0xff], [0x0 ~> 0xff], [0x0 ~> 0xff], [0x0 ~> 0xff], [0x0 ~> 0xff], [0x0 ~> 0xff], [0x0 ~> 0xff], [0x0 ~> 0xff], [0x0 ~> 0xff], [0x0 ~> 0xff], [0x0 ~> 0xff], [0x0 ~> 0xff], [0x0 ~> 0xff], [0x0 ~> 0xff], [0x0 ~> 0xff], [0x0 ~> 0xff], [0x0 ~> 0xff], [0x0 ~> 0xff], [0x0 ~> 0xff], [0x0 ~> 0xff], [0x0 ~> 0xff], [0x0 ~> 0xff], [0x0 ~> 0xff], [0x0 ~> 0xff], [0x0 ~> 0xff], [0x0 ~> 0xff], [0x0 ~> 0xff], [0x0 ~> 0xff], [0x0 ~> 0xff], [0x0 ~> 0xff], [0x0 ~> 0xff], [0x0 ~> 0xff]]
 */
static void fiat_np256_to_bytes(uint8_t out1[32], const uint64_t arg1[4]) {
  uint64_t x1;
  uint64_t x2;
  uint64_t x3;
  uint64_t x4;
  uint8_t x5;
  uint64_t x6;
  uint8_t x7;
  uint64_t x8;
  uint8_t x9;
  uint64_t x10;
  uint8_t x11;
  uint64_t x12;
  uint8_t x13;
  uint64_t x14;
  uint8_t x15;
  uint64_t x16;
  uint8_t x17;
  uint8_t x18;
  uint8_t x19;
  uint64_t x20;
  uint8_t x21;
  uint64_t x22;
  uint8_t x23;
  uint64_t x24;
  uint8_t x25;
  uint64_t x26;
  uint8_t x27;
  uint64_t x28;
  uint8_t x29;
  uint64_t x30;
  uint8_t x31;
  uint8_t x32;
  uint8_t x33;
  uint64_t x34;
  uint8_t x35;
  uint64_t x36;
  uint8_t x37;
  uint64_t x38;
  uint8_t x39;
  uint64_t x40;
  uint8_t x41;
  uint64_t x42;
  uint8_t x43;
  uint64_t x44;
  uint8_t x45;
  uint8_t x46;
  uint8_t x47;
  uint64_t x48;
  uint8_t x49;
  uint64_t x50;
  uint8_t x51;
  uint64_t x52;
  uint8_t x53;
  uint64_t x54;
  uint8_t x55;
  uint64_t x56;
  uint8_t x57;
  uint64_t x58;
  uint8_t x59;
  uint8_t x60;
  x1 = (arg1[3]);
  x2 = (arg1[2]);
  x3 = (arg1[1]);
  x4 = (arg1[0]);
  x5 = (uint8_t)(x4 & UINT8_C(0xff));
  x6 = (x4 >> 8);
  x7 = (uint8_t)(x6 & UINT8_C(0xff));
  x8 = (x6 >> 8);
  x9 = (uint8_t)(x8 & UINT8_C(0xff));
  x10 = (x8 >> 8);
  x11 = (uint8_t)(x10 & UINT8_C(0xff));
  x12 = (x10 >> 8);
  x13 = (uint8_t)(x12 & UINT8_C(0xff));
  x14 = (x12 >> 8);
  x15 = (uint8_t)(x14 & UINT8_C(0xff));
  x16 = (x14 >> 8);
  x17 = (uint8_t)(x16 & UINT8_C(0xff));
  x18 = (uint8_t)(x16 >> 8);
  x19 = (uint8_t)(x3 & UINT8_C(0xff));
  x20 = (x3 >> 8);
  x21 = (uint8_t)(x20 & UINT8_C(0xff));
  x22 = (x20 >> 8);
  x23 = (uint8_t)(x22 & UINT8_C(0xff));
  x24 = (x22 >> 8);
  x25 = (uint8_t)(x24 & UINT8_C(0xff));
  x26 = (x24 >> 8);
  x27 = (uint8_t)(x26 & UINT8_C(0xff));
  x28 = (x26 >> 8);
  x29 = (uint8_t)(x28 & UINT8_C(0xff));
  x30 = (x28 >> 8);
  x31 = (uint8_t)(x30 & UINT8_C(0xff));
  x32 = (uint8_t)(x30 >> 8);
  x33 = (uint8_t)(x2 & UINT8_C(0xff));
  x34 = (x2 >> 8);
  x35 = (uint8_t)(x34 & UINT8_C(0xff));
  x36 = (x34 >> 8);
  x37 = (uint8_t)(x36 & UINT8_C(0xff));
  x38 = (x36 >> 8);
  x39 = (uint8_t)(x38 & UINT8_C(0xff));
  x40 = (x38 >> 8);
  x41 = (uint8_t)(x40 & UINT8_C(0xff));
  x42 = (x40 >> 8);
  x43 = (uint8_t)(x42 & UINT8_C(0xff));
  x44 = (x42 >> 8);
  x45 = (uint8_t)(x44 & UINT8_C(0xff));
  x46 = (uint8_t)(x44 >> 8);
  x47 = (uint8_t)(x1 & UINT8_C(0xff));
  x48 = (x1 >> 8);
  x49 = (uint8_t)(x48 & UINT8_C(0xff));
  x50 = (x48 >> 8);
  x51 = (uint8_t)(x50 & UINT8_C(0xff));
  x52 = (x50 >> 8);
  x53 = (uint8_t)(x52 & UINT8_C(0xff));
  x54 = (x52 >> 8);
  x55 = (uint8_t)(x54 & UINT8_C(0xff));
  x56 = (x54 >> 8);
  x57 = (uint8_t)(x56 & UINT8_C(0xff));
  x58 = (x56 >> 8);
  x59 = (uint8_t)(x58 & UINT8_C(0xff));
  x60 = (uint8_t)(x58 >> 8);
  out1[0] = x5;
  out1[1] = x7;
  out1[2] = x9;
  out1[3] = x11;
  out1[4] = x13;
  out1[5] = x15;
  out1[6] = x17;
  out1[7] = x18;
  out1[8] = x19;
  out1[9] = x21;
  out1[10] = x23;
  out1[11] = x25;
  out1[12] = x27;
  out1[13] = x29;
  out1[14] = x31;
  out1[15] = x32;
  out1[16] = x33;
  out1[17] = x35;
  out1[18] = x37;
  out1[19] = x39;
  out1[20] = x41;
  out1[21] = x43;
  out1[22] = x45;
  out1[23] = x46;
  out1[24] = x47;
  out1[25] = x49;
  out1[26] = x51;
  out1[27] = x53;
  out1[28] = x55;
  out1[29] = x57;
  out1[30] = x59;
  out1[31] = x60;
}

/*
 * The function fiat_np256_from_bytes deserializes a field element NOT in the Montgomery domain from bytes in little-endian order.
 *
 * Preconditions:
 *   0 ≤ bytes_eval arg1 < m
 * Postconditions:
 *   eval out1 mod m = bytes_eval arg1 mod m
 *   0 ≤ eval out1 < m
 *
 * Input Bounds:
 *   arg1: [[0x0 ~> 0xff], [0x0 ~> 0xff], [0x0 ~> 0xff], [0x0 ~> 0xff], [0x0 ~> 0xff], [0x0 ~> 0xff], [0x0 ~> 0xff], [0x0 ~> 0xff], [0x0 ~> 0xff], [0x0 ~> 0xff], [0x0 ~> 0xff], [0x0 ~> 0xff], [0x0 ~> 0xff], [0x0 ~> 0xff], [0x0 ~> 0xff], [0x0 ~> 0xff], [0x0 ~> 0xff], [0x0 ~> 0xff], [0x0 ~> 0xff], [0x0 ~> 0xff], [0x0 ~> 0xff], [0x0 ~> 0xff], [0x0 ~> 0xff], [0x0 ~> 0xff], [0x0 ~> 0xff], [0x0 ~> 0xff], [0x0 ~> 0xff], [0x0 ~> 0xff], [0x0 ~> 0xff], [0x0 ~> 0xff], [0x0 ~> 0xff], [0x0 ~> 0xff]]
 * Output Bounds:
 *   out1: [[0x0 ~> 0xffffffffffffffff], [0x0 ~> 0xffffffffffffffff], [0x0 ~> 0xffffffffffffffff], [0x0 ~> 0xffffffffffffffff]]
 */
static void fiat_np256_from_bytes(uint64_t out1[4], const uint8_t arg1[32]) {
  uint64_t x1;
  uint64_t x2;
  uint64_t x3;
  uint64_t x4;
  uint64_t x5;
  uint64_t x6;
  uint64_t x7;
  uint8_t x8;
  uint64_t x9;
  uint64_t x10;
  uint64_t x11;
  uint64_t x12;
  uint64_t x13;
  uint64_t x14;
  uint64_t x15;
  uint8_t x16;
  uint64_t x17;
  uint64_t x18;
  uint64_t x19;
  uint64_t x20;
  uint64_t x21;
  uint64_t x22;
  uint64_t x23;
  uint8_t x24;
  uint64_t x25;
  uint64_t x26;
  uint64_t x27;
  uint64_t x28;
  uint64_t x29;
  uint64_t x30;
  uint64_t x31;
  uint8_t x32;
  uint64_t x33;
  uint64_t x34;
  uint64_t x35;
  uint64_t x36;
  uint64_t x37;
  uint64_t x38;
  uint64_t x39;
  uint64_t x40;
  uint64_t x41;
  uint64_t x42;
  uint64_t x43;
  uint64_t x44;
  uint64_t x45;
  uint64_t x46;
  uint64_t x47;
  uint64_t x48;
  uint64_t x49;
  uint64_t x50;
  uint64_t x51;
  uint64_t x52;
  uint64_t x53;
  uint64_t x54;
  uint64_t x55;
  uint64_t x56;
  uint64_t x57;
  uint64_t x58;
  uint64_t x59;
  uint64_t x60;
  x1 = ((uint64_t)(arg1[31]) << 56);
  x2 = ((uint64_t)(arg1[30]) << 48);
  x3 = ((uint64_t)(arg1[29]) << 40);
  x4 = ((uint64_t)(arg1[28]) << 32);
  x5 = ((uint64_t)(arg1[27]) << 24);
  x6 = ((uint64_t)(arg1[26]) << 16);
  x7 = ((uint64_t)(arg1[25]) << 8);
  x8 = (arg1[24]);
  x9 = ((uint64_t)(arg1[23]) << 56);
  x10 = ((uint64_t)(arg1[22]) << 48);
  x11 = ((uint64_t)(arg1[21]) << 40);
  x12 = ((uint64_t)(arg1[20]) << 32);
  x13 = ((uint64_t)(arg1[19]) << 24);
  x14 = ((uint64_t)(arg1[18]) << 16);
  x15 = ((uint64_t)(arg1[17]) << 8);
  x16 = (arg1[16]);
  x17 = ((uint64_t)(arg1[15]) << 56);
  x18 = ((uint64_t)(arg1[14]) << 48);
  x19 = ((uint64_t)(arg1[13]) << 40);
  x20 = ((uint64_t)(arg1[12]) << 32);
  x21 = ((uint64_t)(arg1[11]) << 24);
  x22 = ((uint64_t)(arg1[10]) << 16);
  x23 = ((uint64_t)(arg1[9]) << 8);
  x24 = (arg1[8]);
  x25 = ((uint64_t)(arg1[7]) << 56);
  x26 = ((uint64_t)(arg1[6]) << 48);
  x27 = ((uint64_t)(arg1[5]) << 40);
  x28 = ((uint64_t)(arg1[4]) << 32);
  x29 = ((uint64_t)(arg1[3]) << 24);
  x30 = ((uint64_t)(arg1[2]) << 16);
  x31 = ((uint64_t)(arg1[1]) << 8);
  x32 = (arg1[0]);
  x33 = (x31 + (uint64_t)x32);
  x34 = (x30 + x33);
  x35 = (x29 + x34);
  x36 = (x28 + x35);
  x37 = (x27 + x36);
  x38 = (x26 + x37);
  x39 = (x25 + x38);
  x40 = (x23 + (uint64_t)x24);
  x41 = (x22 + x40);
  x42 = (x21 + x41);
  x43 = (x20 + x42);
  x44 = (x19 + x43);
  x45 = (x18 + x44);
  x46 = (x17 + x45);
  x47 = (x15 + (uint64_t)x16);
  x48 = (x14 + x47);
  x49 = (x13 + x48);
  x50 = (x12 + x49);
  x51 = (x11 + x50);
  x52 = (x10 + x51);
  x53 = (x9 + x52);
  x54 = (x7 + (uint64_t)x8);
  x55 = (x6 + x54);
  x56 = (x5 + x55);
  x57 = (x4 + x56);
  x58 = (x3 + x57);
  x59 = (x2 + x58);
  x60 = (x1 + x59);
  out1[0] = x39;
  out1[1] = x46;
  out1[2] = x53;
  out1[3] = x60;
}

/*
 * The function fiat_np256_selectznz is a multi-limb conditional select.
 *
 * Postconditions:
 *   out1 = (if arg1 = 0 then arg2 else arg3)
 *
 * Input Bounds:
 *   arg1: [0x0 ~> 0x1]
 *   arg2: [[0x0 ~> 0xffffffffffffffff], [0x0 ~> 0xffffffffffffffff], [0x0 ~> 0xffffffffffffffff], [0x0 ~> 0xffffffffffffffff]]
 *   arg3: [[0x0 ~> 0xffffffffffffffff], [0x0 ~> 0xffffffffffffffff], [0x0 ~> 0xffffffffffffffff], [0x0 ~> 0xffffffffffffffff]]
 * Output Bounds:
 *   out1: [[0x0 ~> 0xffffffffffffffff], [0x0 ~> 0xffffffffffffffff], [0x0 ~> 0xffffffffffffffff], [0x0 ~> 0xffffffffffffffff]]
 */
static void fiat_np256_selectznz(uint64_t out1[4], fiat_np256_uint1 arg1, const uint64_t arg2[4], const uint64_t arg3[4]) {
  uint64_t x1;
  uint64_t x2;
  uint64_t x3;
  uint64_t x4;
  fiat_np256_cmovznz_u64(&x1, arg1, (arg2[0]), (arg3[0]));
  fiat_np256_cmovznz_u64(&x2, arg1, (arg2[1]), (arg3[1]));
  fiat_np256_cmovznz_u64(&x3, arg1, (arg2[2]), (arg3[2]));
  fiat_np256_cmovznz_u64(&x4, arg1, (arg2[3]), (arg3[3]));
  out1[0] = x1;
  out1[1] = x2;
  out1[2] = x3;
  out1[3] = x4;
}