cachepc-linux

Fork of AMDESE/linux with modifications for CachePC side-channel attack
git clone https://git.sinitax.com/sinitax/cachepc-linux
Log | Files | Refs | README | LICENSE | sfeed.txt

curve25519-fiat32.c (30257B)


      1// SPDX-License-Identifier: GPL-2.0 OR MIT
      2/*
      3 * Copyright (C) 2015-2016 The fiat-crypto Authors.
      4 * Copyright (C) 2018-2019 Jason A. Donenfeld <Jason@zx2c4.com>. All Rights Reserved.
      5 *
      6 * This is a machine-generated formally verified implementation of Curve25519
      7 * ECDH from: <https://github.com/mit-plv/fiat-crypto>. Though originally
      8 * machine generated, it has been tweaked to be suitable for use in the kernel.
      9 * It is optimized for 32-bit machines and machines that cannot work efficiently
     10 * with 128-bit integer types.
     11 */
     12
     13#include <asm/unaligned.h>
     14#include <crypto/curve25519.h>
     15#include <linux/string.h>
     16
     17/* fe means field element. Here the field is \Z/(2^255-19). An element t,
     18 * entries t[0]...t[9], represents the integer t[0]+2^26 t[1]+2^51 t[2]+2^77
     19 * t[3]+2^102 t[4]+...+2^230 t[9].
     20 * fe limbs are bounded by 1.125*2^26,1.125*2^25,1.125*2^26,1.125*2^25,etc.
     21 * Multiplication and carrying produce fe from fe_loose.
     22 */
     23typedef struct fe { u32 v[10]; } fe;
     24
     25/* fe_loose limbs are bounded by 3.375*2^26,3.375*2^25,3.375*2^26,3.375*2^25,etc
     26 * Addition and subtraction produce fe_loose from (fe, fe).
     27 */
     28typedef struct fe_loose { u32 v[10]; } fe_loose;
     29
     30static __always_inline void fe_frombytes_impl(u32 h[10], const u8 *s)
     31{
     32	/* Ignores top bit of s. */
     33	u32 a0 = get_unaligned_le32(s);
     34	u32 a1 = get_unaligned_le32(s+4);
     35	u32 a2 = get_unaligned_le32(s+8);
     36	u32 a3 = get_unaligned_le32(s+12);
     37	u32 a4 = get_unaligned_le32(s+16);
     38	u32 a5 = get_unaligned_le32(s+20);
     39	u32 a6 = get_unaligned_le32(s+24);
     40	u32 a7 = get_unaligned_le32(s+28);
     41	h[0] = a0&((1<<26)-1);                    /* 26 used, 32-26 left.   26 */
     42	h[1] = (a0>>26) | ((a1&((1<<19)-1))<< 6); /* (32-26) + 19 =  6+19 = 25 */
     43	h[2] = (a1>>19) | ((a2&((1<<13)-1))<<13); /* (32-19) + 13 = 13+13 = 26 */
     44	h[3] = (a2>>13) | ((a3&((1<< 6)-1))<<19); /* (32-13) +  6 = 19+ 6 = 25 */
     45	h[4] = (a3>> 6);                          /* (32- 6)              = 26 */
     46	h[5] = a4&((1<<25)-1);                    /*                        25 */
     47	h[6] = (a4>>25) | ((a5&((1<<19)-1))<< 7); /* (32-25) + 19 =  7+19 = 26 */
     48	h[7] = (a5>>19) | ((a6&((1<<12)-1))<<13); /* (32-19) + 12 = 13+12 = 25 */
     49	h[8] = (a6>>12) | ((a7&((1<< 6)-1))<<20); /* (32-12) +  6 = 20+ 6 = 26 */
     50	h[9] = (a7>> 6)&((1<<25)-1); /*                                     25 */
     51}
     52
     53static __always_inline void fe_frombytes(fe *h, const u8 *s)
     54{
     55	fe_frombytes_impl(h->v, s);
     56}
     57
     58static __always_inline u8 /*bool*/
     59addcarryx_u25(u8 /*bool*/ c, u32 a, u32 b, u32 *low)
     60{
     61	/* This function extracts 25 bits of result and 1 bit of carry
     62	 * (26 total), so a 32-bit intermediate is sufficient.
     63	 */
     64	u32 x = a + b + c;
     65	*low = x & ((1 << 25) - 1);
     66	return (x >> 25) & 1;
     67}
     68
     69static __always_inline u8 /*bool*/
     70addcarryx_u26(u8 /*bool*/ c, u32 a, u32 b, u32 *low)
     71{
     72	/* This function extracts 26 bits of result and 1 bit of carry
     73	 * (27 total), so a 32-bit intermediate is sufficient.
     74	 */
     75	u32 x = a + b + c;
     76	*low = x & ((1 << 26) - 1);
     77	return (x >> 26) & 1;
     78}
     79
     80static __always_inline u8 /*bool*/
     81subborrow_u25(u8 /*bool*/ c, u32 a, u32 b, u32 *low)
     82{
     83	/* This function extracts 25 bits of result and 1 bit of borrow
     84	 * (26 total), so a 32-bit intermediate is sufficient.
     85	 */
     86	u32 x = a - b - c;
     87	*low = x & ((1 << 25) - 1);
     88	return x >> 31;
     89}
     90
     91static __always_inline u8 /*bool*/
     92subborrow_u26(u8 /*bool*/ c, u32 a, u32 b, u32 *low)
     93{
     94	/* This function extracts 26 bits of result and 1 bit of borrow
     95	 *(27 total), so a 32-bit intermediate is sufficient.
     96	 */
     97	u32 x = a - b - c;
     98	*low = x & ((1 << 26) - 1);
     99	return x >> 31;
    100}
    101
    102static __always_inline u32 cmovznz32(u32 t, u32 z, u32 nz)
    103{
    104	t = -!!t; /* all set if nonzero, 0 if 0 */
    105	return (t&nz) | ((~t)&z);
    106}
    107
    108static __always_inline void fe_freeze(u32 out[10], const u32 in1[10])
    109{
    110	{ const u32 x17 = in1[9];
    111	{ const u32 x18 = in1[8];
    112	{ const u32 x16 = in1[7];
    113	{ const u32 x14 = in1[6];
    114	{ const u32 x12 = in1[5];
    115	{ const u32 x10 = in1[4];
    116	{ const u32 x8 = in1[3];
    117	{ const u32 x6 = in1[2];
    118	{ const u32 x4 = in1[1];
    119	{ const u32 x2 = in1[0];
    120	{ u32 x20; u8/*bool*/ x21 = subborrow_u26(0x0, x2, 0x3ffffed, &x20);
    121	{ u32 x23; u8/*bool*/ x24 = subborrow_u25(x21, x4, 0x1ffffff, &x23);
    122	{ u32 x26; u8/*bool*/ x27 = subborrow_u26(x24, x6, 0x3ffffff, &x26);
    123	{ u32 x29; u8/*bool*/ x30 = subborrow_u25(x27, x8, 0x1ffffff, &x29);
    124	{ u32 x32; u8/*bool*/ x33 = subborrow_u26(x30, x10, 0x3ffffff, &x32);
    125	{ u32 x35; u8/*bool*/ x36 = subborrow_u25(x33, x12, 0x1ffffff, &x35);
    126	{ u32 x38; u8/*bool*/ x39 = subborrow_u26(x36, x14, 0x3ffffff, &x38);
    127	{ u32 x41; u8/*bool*/ x42 = subborrow_u25(x39, x16, 0x1ffffff, &x41);
    128	{ u32 x44; u8/*bool*/ x45 = subborrow_u26(x42, x18, 0x3ffffff, &x44);
    129	{ u32 x47; u8/*bool*/ x48 = subborrow_u25(x45, x17, 0x1ffffff, &x47);
    130	{ u32 x49 = cmovznz32(x48, 0x0, 0xffffffff);
    131	{ u32 x50 = (x49 & 0x3ffffed);
    132	{ u32 x52; u8/*bool*/ x53 = addcarryx_u26(0x0, x20, x50, &x52);
    133	{ u32 x54 = (x49 & 0x1ffffff);
    134	{ u32 x56; u8/*bool*/ x57 = addcarryx_u25(x53, x23, x54, &x56);
    135	{ u32 x58 = (x49 & 0x3ffffff);
    136	{ u32 x60; u8/*bool*/ x61 = addcarryx_u26(x57, x26, x58, &x60);
    137	{ u32 x62 = (x49 & 0x1ffffff);
    138	{ u32 x64; u8/*bool*/ x65 = addcarryx_u25(x61, x29, x62, &x64);
    139	{ u32 x66 = (x49 & 0x3ffffff);
    140	{ u32 x68; u8/*bool*/ x69 = addcarryx_u26(x65, x32, x66, &x68);
    141	{ u32 x70 = (x49 & 0x1ffffff);
    142	{ u32 x72; u8/*bool*/ x73 = addcarryx_u25(x69, x35, x70, &x72);
    143	{ u32 x74 = (x49 & 0x3ffffff);
    144	{ u32 x76; u8/*bool*/ x77 = addcarryx_u26(x73, x38, x74, &x76);
    145	{ u32 x78 = (x49 & 0x1ffffff);
    146	{ u32 x80; u8/*bool*/ x81 = addcarryx_u25(x77, x41, x78, &x80);
    147	{ u32 x82 = (x49 & 0x3ffffff);
    148	{ u32 x84; u8/*bool*/ x85 = addcarryx_u26(x81, x44, x82, &x84);
    149	{ u32 x86 = (x49 & 0x1ffffff);
    150	{ u32 x88; addcarryx_u25(x85, x47, x86, &x88);
    151	out[0] = x52;
    152	out[1] = x56;
    153	out[2] = x60;
    154	out[3] = x64;
    155	out[4] = x68;
    156	out[5] = x72;
    157	out[6] = x76;
    158	out[7] = x80;
    159	out[8] = x84;
    160	out[9] = x88;
    161	}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}
    162}
    163
    164static __always_inline void fe_tobytes(u8 s[32], const fe *f)
    165{
    166	u32 h[10];
    167	fe_freeze(h, f->v);
    168	s[0] = h[0] >> 0;
    169	s[1] = h[0] >> 8;
    170	s[2] = h[0] >> 16;
    171	s[3] = (h[0] >> 24) | (h[1] << 2);
    172	s[4] = h[1] >> 6;
    173	s[5] = h[1] >> 14;
    174	s[6] = (h[1] >> 22) | (h[2] << 3);
    175	s[7] = h[2] >> 5;
    176	s[8] = h[2] >> 13;
    177	s[9] = (h[2] >> 21) | (h[3] << 5);
    178	s[10] = h[3] >> 3;
    179	s[11] = h[3] >> 11;
    180	s[12] = (h[3] >> 19) | (h[4] << 6);
    181	s[13] = h[4] >> 2;
    182	s[14] = h[4] >> 10;
    183	s[15] = h[4] >> 18;
    184	s[16] = h[5] >> 0;
    185	s[17] = h[5] >> 8;
    186	s[18] = h[5] >> 16;
    187	s[19] = (h[5] >> 24) | (h[6] << 1);
    188	s[20] = h[6] >> 7;
    189	s[21] = h[6] >> 15;
    190	s[22] = (h[6] >> 23) | (h[7] << 3);
    191	s[23] = h[7] >> 5;
    192	s[24] = h[7] >> 13;
    193	s[25] = (h[7] >> 21) | (h[8] << 4);
    194	s[26] = h[8] >> 4;
    195	s[27] = h[8] >> 12;
    196	s[28] = (h[8] >> 20) | (h[9] << 6);
    197	s[29] = h[9] >> 2;
    198	s[30] = h[9] >> 10;
    199	s[31] = h[9] >> 18;
    200}
    201
    202/* h = f */
    203static __always_inline void fe_copy(fe *h, const fe *f)
    204{
    205	memmove(h, f, sizeof(u32) * 10);
    206}
    207
    208static __always_inline void fe_copy_lt(fe_loose *h, const fe *f)
    209{
    210	memmove(h, f, sizeof(u32) * 10);
    211}
    212
    213/* h = 0 */
    214static __always_inline void fe_0(fe *h)
    215{
    216	memset(h, 0, sizeof(u32) * 10);
    217}
    218
    219/* h = 1 */
    220static __always_inline void fe_1(fe *h)
    221{
    222	memset(h, 0, sizeof(u32) * 10);
    223	h->v[0] = 1;
    224}
    225
    226static noinline void fe_add_impl(u32 out[10], const u32 in1[10], const u32 in2[10])
    227{
    228	{ const u32 x20 = in1[9];
    229	{ const u32 x21 = in1[8];
    230	{ const u32 x19 = in1[7];
    231	{ const u32 x17 = in1[6];
    232	{ const u32 x15 = in1[5];
    233	{ const u32 x13 = in1[4];
    234	{ const u32 x11 = in1[3];
    235	{ const u32 x9 = in1[2];
    236	{ const u32 x7 = in1[1];
    237	{ const u32 x5 = in1[0];
    238	{ const u32 x38 = in2[9];
    239	{ const u32 x39 = in2[8];
    240	{ const u32 x37 = in2[7];
    241	{ const u32 x35 = in2[6];
    242	{ const u32 x33 = in2[5];
    243	{ const u32 x31 = in2[4];
    244	{ const u32 x29 = in2[3];
    245	{ const u32 x27 = in2[2];
    246	{ const u32 x25 = in2[1];
    247	{ const u32 x23 = in2[0];
    248	out[0] = (x5 + x23);
    249	out[1] = (x7 + x25);
    250	out[2] = (x9 + x27);
    251	out[3] = (x11 + x29);
    252	out[4] = (x13 + x31);
    253	out[5] = (x15 + x33);
    254	out[6] = (x17 + x35);
    255	out[7] = (x19 + x37);
    256	out[8] = (x21 + x39);
    257	out[9] = (x20 + x38);
    258	}}}}}}}}}}}}}}}}}}}}
    259}
    260
    261/* h = f + g
    262 * Can overlap h with f or g.
    263 */
    264static __always_inline void fe_add(fe_loose *h, const fe *f, const fe *g)
    265{
    266	fe_add_impl(h->v, f->v, g->v);
    267}
    268
    269static noinline void fe_sub_impl(u32 out[10], const u32 in1[10], const u32 in2[10])
    270{
    271	{ const u32 x20 = in1[9];
    272	{ const u32 x21 = in1[8];
    273	{ const u32 x19 = in1[7];
    274	{ const u32 x17 = in1[6];
    275	{ const u32 x15 = in1[5];
    276	{ const u32 x13 = in1[4];
    277	{ const u32 x11 = in1[3];
    278	{ const u32 x9 = in1[2];
    279	{ const u32 x7 = in1[1];
    280	{ const u32 x5 = in1[0];
    281	{ const u32 x38 = in2[9];
    282	{ const u32 x39 = in2[8];
    283	{ const u32 x37 = in2[7];
    284	{ const u32 x35 = in2[6];
    285	{ const u32 x33 = in2[5];
    286	{ const u32 x31 = in2[4];
    287	{ const u32 x29 = in2[3];
    288	{ const u32 x27 = in2[2];
    289	{ const u32 x25 = in2[1];
    290	{ const u32 x23 = in2[0];
    291	out[0] = ((0x7ffffda + x5) - x23);
    292	out[1] = ((0x3fffffe + x7) - x25);
    293	out[2] = ((0x7fffffe + x9) - x27);
    294	out[3] = ((0x3fffffe + x11) - x29);
    295	out[4] = ((0x7fffffe + x13) - x31);
    296	out[5] = ((0x3fffffe + x15) - x33);
    297	out[6] = ((0x7fffffe + x17) - x35);
    298	out[7] = ((0x3fffffe + x19) - x37);
    299	out[8] = ((0x7fffffe + x21) - x39);
    300	out[9] = ((0x3fffffe + x20) - x38);
    301	}}}}}}}}}}}}}}}}}}}}
    302}
    303
    304/* h = f - g
    305 * Can overlap h with f or g.
    306 */
    307static __always_inline void fe_sub(fe_loose *h, const fe *f, const fe *g)
    308{
    309	fe_sub_impl(h->v, f->v, g->v);
    310}
    311
    312static noinline void fe_mul_impl(u32 out[10], const u32 in1[10], const u32 in2[10])
    313{
    314	{ const u32 x20 = in1[9];
    315	{ const u32 x21 = in1[8];
    316	{ const u32 x19 = in1[7];
    317	{ const u32 x17 = in1[6];
    318	{ const u32 x15 = in1[5];
    319	{ const u32 x13 = in1[4];
    320	{ const u32 x11 = in1[3];
    321	{ const u32 x9 = in1[2];
    322	{ const u32 x7 = in1[1];
    323	{ const u32 x5 = in1[0];
    324	{ const u32 x38 = in2[9];
    325	{ const u32 x39 = in2[8];
    326	{ const u32 x37 = in2[7];
    327	{ const u32 x35 = in2[6];
    328	{ const u32 x33 = in2[5];
    329	{ const u32 x31 = in2[4];
    330	{ const u32 x29 = in2[3];
    331	{ const u32 x27 = in2[2];
    332	{ const u32 x25 = in2[1];
    333	{ const u32 x23 = in2[0];
    334	{ u64 x40 = ((u64)x23 * x5);
    335	{ u64 x41 = (((u64)x23 * x7) + ((u64)x25 * x5));
    336	{ u64 x42 = ((((u64)(0x2 * x25) * x7) + ((u64)x23 * x9)) + ((u64)x27 * x5));
    337	{ u64 x43 = (((((u64)x25 * x9) + ((u64)x27 * x7)) + ((u64)x23 * x11)) + ((u64)x29 * x5));
    338	{ u64 x44 = (((((u64)x27 * x9) + (0x2 * (((u64)x25 * x11) + ((u64)x29 * x7)))) + ((u64)x23 * x13)) + ((u64)x31 * x5));
    339	{ u64 x45 = (((((((u64)x27 * x11) + ((u64)x29 * x9)) + ((u64)x25 * x13)) + ((u64)x31 * x7)) + ((u64)x23 * x15)) + ((u64)x33 * x5));
    340	{ u64 x46 = (((((0x2 * ((((u64)x29 * x11) + ((u64)x25 * x15)) + ((u64)x33 * x7))) + ((u64)x27 * x13)) + ((u64)x31 * x9)) + ((u64)x23 * x17)) + ((u64)x35 * x5));
    341	{ u64 x47 = (((((((((u64)x29 * x13) + ((u64)x31 * x11)) + ((u64)x27 * x15)) + ((u64)x33 * x9)) + ((u64)x25 * x17)) + ((u64)x35 * x7)) + ((u64)x23 * x19)) + ((u64)x37 * x5));
    342	{ u64 x48 = (((((((u64)x31 * x13) + (0x2 * (((((u64)x29 * x15) + ((u64)x33 * x11)) + ((u64)x25 * x19)) + ((u64)x37 * x7)))) + ((u64)x27 * x17)) + ((u64)x35 * x9)) + ((u64)x23 * x21)) + ((u64)x39 * x5));
    343	{ u64 x49 = (((((((((((u64)x31 * x15) + ((u64)x33 * x13)) + ((u64)x29 * x17)) + ((u64)x35 * x11)) + ((u64)x27 * x19)) + ((u64)x37 * x9)) + ((u64)x25 * x21)) + ((u64)x39 * x7)) + ((u64)x23 * x20)) + ((u64)x38 * x5));
    344	{ u64 x50 = (((((0x2 * ((((((u64)x33 * x15) + ((u64)x29 * x19)) + ((u64)x37 * x11)) + ((u64)x25 * x20)) + ((u64)x38 * x7))) + ((u64)x31 * x17)) + ((u64)x35 * x13)) + ((u64)x27 * x21)) + ((u64)x39 * x9));
    345	{ u64 x51 = (((((((((u64)x33 * x17) + ((u64)x35 * x15)) + ((u64)x31 * x19)) + ((u64)x37 * x13)) + ((u64)x29 * x21)) + ((u64)x39 * x11)) + ((u64)x27 * x20)) + ((u64)x38 * x9));
    346	{ u64 x52 = (((((u64)x35 * x17) + (0x2 * (((((u64)x33 * x19) + ((u64)x37 * x15)) + ((u64)x29 * x20)) + ((u64)x38 * x11)))) + ((u64)x31 * x21)) + ((u64)x39 * x13));
    347	{ u64 x53 = (((((((u64)x35 * x19) + ((u64)x37 * x17)) + ((u64)x33 * x21)) + ((u64)x39 * x15)) + ((u64)x31 * x20)) + ((u64)x38 * x13));
    348	{ u64 x54 = (((0x2 * ((((u64)x37 * x19) + ((u64)x33 * x20)) + ((u64)x38 * x15))) + ((u64)x35 * x21)) + ((u64)x39 * x17));
    349	{ u64 x55 = (((((u64)x37 * x21) + ((u64)x39 * x19)) + ((u64)x35 * x20)) + ((u64)x38 * x17));
    350	{ u64 x56 = (((u64)x39 * x21) + (0x2 * (((u64)x37 * x20) + ((u64)x38 * x19))));
    351	{ u64 x57 = (((u64)x39 * x20) + ((u64)x38 * x21));
    352	{ u64 x58 = ((u64)(0x2 * x38) * x20);
    353	{ u64 x59 = (x48 + (x58 << 0x4));
    354	{ u64 x60 = (x59 + (x58 << 0x1));
    355	{ u64 x61 = (x60 + x58);
    356	{ u64 x62 = (x47 + (x57 << 0x4));
    357	{ u64 x63 = (x62 + (x57 << 0x1));
    358	{ u64 x64 = (x63 + x57);
    359	{ u64 x65 = (x46 + (x56 << 0x4));
    360	{ u64 x66 = (x65 + (x56 << 0x1));
    361	{ u64 x67 = (x66 + x56);
    362	{ u64 x68 = (x45 + (x55 << 0x4));
    363	{ u64 x69 = (x68 + (x55 << 0x1));
    364	{ u64 x70 = (x69 + x55);
    365	{ u64 x71 = (x44 + (x54 << 0x4));
    366	{ u64 x72 = (x71 + (x54 << 0x1));
    367	{ u64 x73 = (x72 + x54);
    368	{ u64 x74 = (x43 + (x53 << 0x4));
    369	{ u64 x75 = (x74 + (x53 << 0x1));
    370	{ u64 x76 = (x75 + x53);
    371	{ u64 x77 = (x42 + (x52 << 0x4));
    372	{ u64 x78 = (x77 + (x52 << 0x1));
    373	{ u64 x79 = (x78 + x52);
    374	{ u64 x80 = (x41 + (x51 << 0x4));
    375	{ u64 x81 = (x80 + (x51 << 0x1));
    376	{ u64 x82 = (x81 + x51);
    377	{ u64 x83 = (x40 + (x50 << 0x4));
    378	{ u64 x84 = (x83 + (x50 << 0x1));
    379	{ u64 x85 = (x84 + x50);
    380	{ u64 x86 = (x85 >> 0x1a);
    381	{ u32 x87 = ((u32)x85 & 0x3ffffff);
    382	{ u64 x88 = (x86 + x82);
    383	{ u64 x89 = (x88 >> 0x19);
    384	{ u32 x90 = ((u32)x88 & 0x1ffffff);
    385	{ u64 x91 = (x89 + x79);
    386	{ u64 x92 = (x91 >> 0x1a);
    387	{ u32 x93 = ((u32)x91 & 0x3ffffff);
    388	{ u64 x94 = (x92 + x76);
    389	{ u64 x95 = (x94 >> 0x19);
    390	{ u32 x96 = ((u32)x94 & 0x1ffffff);
    391	{ u64 x97 = (x95 + x73);
    392	{ u64 x98 = (x97 >> 0x1a);
    393	{ u32 x99 = ((u32)x97 & 0x3ffffff);
    394	{ u64 x100 = (x98 + x70);
    395	{ u64 x101 = (x100 >> 0x19);
    396	{ u32 x102 = ((u32)x100 & 0x1ffffff);
    397	{ u64 x103 = (x101 + x67);
    398	{ u64 x104 = (x103 >> 0x1a);
    399	{ u32 x105 = ((u32)x103 & 0x3ffffff);
    400	{ u64 x106 = (x104 + x64);
    401	{ u64 x107 = (x106 >> 0x19);
    402	{ u32 x108 = ((u32)x106 & 0x1ffffff);
    403	{ u64 x109 = (x107 + x61);
    404	{ u64 x110 = (x109 >> 0x1a);
    405	{ u32 x111 = ((u32)x109 & 0x3ffffff);
    406	{ u64 x112 = (x110 + x49);
    407	{ u64 x113 = (x112 >> 0x19);
    408	{ u32 x114 = ((u32)x112 & 0x1ffffff);
    409	{ u64 x115 = (x87 + (0x13 * x113));
    410	{ u32 x116 = (u32) (x115 >> 0x1a);
    411	{ u32 x117 = ((u32)x115 & 0x3ffffff);
    412	{ u32 x118 = (x116 + x90);
    413	{ u32 x119 = (x118 >> 0x19);
    414	{ u32 x120 = (x118 & 0x1ffffff);
    415	out[0] = x117;
    416	out[1] = x120;
    417	out[2] = (x119 + x93);
    418	out[3] = x96;
    419	out[4] = x99;
    420	out[5] = x102;
    421	out[6] = x105;
    422	out[7] = x108;
    423	out[8] = x111;
    424	out[9] = x114;
    425	}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}
    426}
    427
    428static __always_inline void fe_mul_ttt(fe *h, const fe *f, const fe *g)
    429{
    430	fe_mul_impl(h->v, f->v, g->v);
    431}
    432
    433static __always_inline void fe_mul_tlt(fe *h, const fe_loose *f, const fe *g)
    434{
    435	fe_mul_impl(h->v, f->v, g->v);
    436}
    437
    438static __always_inline void
    439fe_mul_tll(fe *h, const fe_loose *f, const fe_loose *g)
    440{
    441	fe_mul_impl(h->v, f->v, g->v);
    442}
    443
    444static noinline void fe_sqr_impl(u32 out[10], const u32 in1[10])
    445{
    446	{ const u32 x17 = in1[9];
    447	{ const u32 x18 = in1[8];
    448	{ const u32 x16 = in1[7];
    449	{ const u32 x14 = in1[6];
    450	{ const u32 x12 = in1[5];
    451	{ const u32 x10 = in1[4];
    452	{ const u32 x8 = in1[3];
    453	{ const u32 x6 = in1[2];
    454	{ const u32 x4 = in1[1];
    455	{ const u32 x2 = in1[0];
    456	{ u64 x19 = ((u64)x2 * x2);
    457	{ u64 x20 = ((u64)(0x2 * x2) * x4);
    458	{ u64 x21 = (0x2 * (((u64)x4 * x4) + ((u64)x2 * x6)));
    459	{ u64 x22 = (0x2 * (((u64)x4 * x6) + ((u64)x2 * x8)));
    460	{ u64 x23 = ((((u64)x6 * x6) + ((u64)(0x4 * x4) * x8)) + ((u64)(0x2 * x2) * x10));
    461	{ u64 x24 = (0x2 * ((((u64)x6 * x8) + ((u64)x4 * x10)) + ((u64)x2 * x12)));
    462	{ u64 x25 = (0x2 * (((((u64)x8 * x8) + ((u64)x6 * x10)) + ((u64)x2 * x14)) + ((u64)(0x2 * x4) * x12)));
    463	{ u64 x26 = (0x2 * (((((u64)x8 * x10) + ((u64)x6 * x12)) + ((u64)x4 * x14)) + ((u64)x2 * x16)));
    464	{ u64 x27 = (((u64)x10 * x10) + (0x2 * ((((u64)x6 * x14) + ((u64)x2 * x18)) + (0x2 * (((u64)x4 * x16) + ((u64)x8 * x12))))));
    465	{ u64 x28 = (0x2 * ((((((u64)x10 * x12) + ((u64)x8 * x14)) + ((u64)x6 * x16)) + ((u64)x4 * x18)) + ((u64)x2 * x17)));
    466	{ u64 x29 = (0x2 * (((((u64)x12 * x12) + ((u64)x10 * x14)) + ((u64)x6 * x18)) + (0x2 * (((u64)x8 * x16) + ((u64)x4 * x17)))));
    467	{ u64 x30 = (0x2 * (((((u64)x12 * x14) + ((u64)x10 * x16)) + ((u64)x8 * x18)) + ((u64)x6 * x17)));
    468	{ u64 x31 = (((u64)x14 * x14) + (0x2 * (((u64)x10 * x18) + (0x2 * (((u64)x12 * x16) + ((u64)x8 * x17))))));
    469	{ u64 x32 = (0x2 * ((((u64)x14 * x16) + ((u64)x12 * x18)) + ((u64)x10 * x17)));
    470	{ u64 x33 = (0x2 * ((((u64)x16 * x16) + ((u64)x14 * x18)) + ((u64)(0x2 * x12) * x17)));
    471	{ u64 x34 = (0x2 * (((u64)x16 * x18) + ((u64)x14 * x17)));
    472	{ u64 x35 = (((u64)x18 * x18) + ((u64)(0x4 * x16) * x17));
    473	{ u64 x36 = ((u64)(0x2 * x18) * x17);
    474	{ u64 x37 = ((u64)(0x2 * x17) * x17);
    475	{ u64 x38 = (x27 + (x37 << 0x4));
    476	{ u64 x39 = (x38 + (x37 << 0x1));
    477	{ u64 x40 = (x39 + x37);
    478	{ u64 x41 = (x26 + (x36 << 0x4));
    479	{ u64 x42 = (x41 + (x36 << 0x1));
    480	{ u64 x43 = (x42 + x36);
    481	{ u64 x44 = (x25 + (x35 << 0x4));
    482	{ u64 x45 = (x44 + (x35 << 0x1));
    483	{ u64 x46 = (x45 + x35);
    484	{ u64 x47 = (x24 + (x34 << 0x4));
    485	{ u64 x48 = (x47 + (x34 << 0x1));
    486	{ u64 x49 = (x48 + x34);
    487	{ u64 x50 = (x23 + (x33 << 0x4));
    488	{ u64 x51 = (x50 + (x33 << 0x1));
    489	{ u64 x52 = (x51 + x33);
    490	{ u64 x53 = (x22 + (x32 << 0x4));
    491	{ u64 x54 = (x53 + (x32 << 0x1));
    492	{ u64 x55 = (x54 + x32);
    493	{ u64 x56 = (x21 + (x31 << 0x4));
    494	{ u64 x57 = (x56 + (x31 << 0x1));
    495	{ u64 x58 = (x57 + x31);
    496	{ u64 x59 = (x20 + (x30 << 0x4));
    497	{ u64 x60 = (x59 + (x30 << 0x1));
    498	{ u64 x61 = (x60 + x30);
    499	{ u64 x62 = (x19 + (x29 << 0x4));
    500	{ u64 x63 = (x62 + (x29 << 0x1));
    501	{ u64 x64 = (x63 + x29);
    502	{ u64 x65 = (x64 >> 0x1a);
    503	{ u32 x66 = ((u32)x64 & 0x3ffffff);
    504	{ u64 x67 = (x65 + x61);
    505	{ u64 x68 = (x67 >> 0x19);
    506	{ u32 x69 = ((u32)x67 & 0x1ffffff);
    507	{ u64 x70 = (x68 + x58);
    508	{ u64 x71 = (x70 >> 0x1a);
    509	{ u32 x72 = ((u32)x70 & 0x3ffffff);
    510	{ u64 x73 = (x71 + x55);
    511	{ u64 x74 = (x73 >> 0x19);
    512	{ u32 x75 = ((u32)x73 & 0x1ffffff);
    513	{ u64 x76 = (x74 + x52);
    514	{ u64 x77 = (x76 >> 0x1a);
    515	{ u32 x78 = ((u32)x76 & 0x3ffffff);
    516	{ u64 x79 = (x77 + x49);
    517	{ u64 x80 = (x79 >> 0x19);
    518	{ u32 x81 = ((u32)x79 & 0x1ffffff);
    519	{ u64 x82 = (x80 + x46);
    520	{ u64 x83 = (x82 >> 0x1a);
    521	{ u32 x84 = ((u32)x82 & 0x3ffffff);
    522	{ u64 x85 = (x83 + x43);
    523	{ u64 x86 = (x85 >> 0x19);
    524	{ u32 x87 = ((u32)x85 & 0x1ffffff);
    525	{ u64 x88 = (x86 + x40);
    526	{ u64 x89 = (x88 >> 0x1a);
    527	{ u32 x90 = ((u32)x88 & 0x3ffffff);
    528	{ u64 x91 = (x89 + x28);
    529	{ u64 x92 = (x91 >> 0x19);
    530	{ u32 x93 = ((u32)x91 & 0x1ffffff);
    531	{ u64 x94 = (x66 + (0x13 * x92));
    532	{ u32 x95 = (u32) (x94 >> 0x1a);
    533	{ u32 x96 = ((u32)x94 & 0x3ffffff);
    534	{ u32 x97 = (x95 + x69);
    535	{ u32 x98 = (x97 >> 0x19);
    536	{ u32 x99 = (x97 & 0x1ffffff);
    537	out[0] = x96;
    538	out[1] = x99;
    539	out[2] = (x98 + x72);
    540	out[3] = x75;
    541	out[4] = x78;
    542	out[5] = x81;
    543	out[6] = x84;
    544	out[7] = x87;
    545	out[8] = x90;
    546	out[9] = x93;
    547	}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}
    548}
    549
    550static __always_inline void fe_sq_tl(fe *h, const fe_loose *f)
    551{
    552	fe_sqr_impl(h->v, f->v);
    553}
    554
    555static __always_inline void fe_sq_tt(fe *h, const fe *f)
    556{
    557	fe_sqr_impl(h->v, f->v);
    558}
    559
    560static __always_inline void fe_loose_invert(fe *out, const fe_loose *z)
    561{
    562	fe t0;
    563	fe t1;
    564	fe t2;
    565	fe t3;
    566	int i;
    567
    568	fe_sq_tl(&t0, z);
    569	fe_sq_tt(&t1, &t0);
    570	for (i = 1; i < 2; ++i)
    571		fe_sq_tt(&t1, &t1);
    572	fe_mul_tlt(&t1, z, &t1);
    573	fe_mul_ttt(&t0, &t0, &t1);
    574	fe_sq_tt(&t2, &t0);
    575	fe_mul_ttt(&t1, &t1, &t2);
    576	fe_sq_tt(&t2, &t1);
    577	for (i = 1; i < 5; ++i)
    578		fe_sq_tt(&t2, &t2);
    579	fe_mul_ttt(&t1, &t2, &t1);
    580	fe_sq_tt(&t2, &t1);
    581	for (i = 1; i < 10; ++i)
    582		fe_sq_tt(&t2, &t2);
    583	fe_mul_ttt(&t2, &t2, &t1);
    584	fe_sq_tt(&t3, &t2);
    585	for (i = 1; i < 20; ++i)
    586		fe_sq_tt(&t3, &t3);
    587	fe_mul_ttt(&t2, &t3, &t2);
    588	fe_sq_tt(&t2, &t2);
    589	for (i = 1; i < 10; ++i)
    590		fe_sq_tt(&t2, &t2);
    591	fe_mul_ttt(&t1, &t2, &t1);
    592	fe_sq_tt(&t2, &t1);
    593	for (i = 1; i < 50; ++i)
    594		fe_sq_tt(&t2, &t2);
    595	fe_mul_ttt(&t2, &t2, &t1);
    596	fe_sq_tt(&t3, &t2);
    597	for (i = 1; i < 100; ++i)
    598		fe_sq_tt(&t3, &t3);
    599	fe_mul_ttt(&t2, &t3, &t2);
    600	fe_sq_tt(&t2, &t2);
    601	for (i = 1; i < 50; ++i)
    602		fe_sq_tt(&t2, &t2);
    603	fe_mul_ttt(&t1, &t2, &t1);
    604	fe_sq_tt(&t1, &t1);
    605	for (i = 1; i < 5; ++i)
    606		fe_sq_tt(&t1, &t1);
    607	fe_mul_ttt(out, &t1, &t0);
    608}
    609
    610static __always_inline void fe_invert(fe *out, const fe *z)
    611{
    612	fe_loose l;
    613	fe_copy_lt(&l, z);
    614	fe_loose_invert(out, &l);
    615}
    616
    617/* Replace (f,g) with (g,f) if b == 1;
    618 * replace (f,g) with (f,g) if b == 0.
    619 *
    620 * Preconditions: b in {0,1}
    621 */
    622static noinline void fe_cswap(fe *f, fe *g, unsigned int b)
    623{
    624	unsigned i;
    625	b = 0 - b;
    626	for (i = 0; i < 10; i++) {
    627		u32 x = f->v[i] ^ g->v[i];
    628		x &= b;
    629		f->v[i] ^= x;
    630		g->v[i] ^= x;
    631	}
    632}
    633
    634/* NOTE: based on fiat-crypto fe_mul, edited for in2=121666, 0, 0.*/
    635static __always_inline void fe_mul_121666_impl(u32 out[10], const u32 in1[10])
    636{
    637	{ const u32 x20 = in1[9];
    638	{ const u32 x21 = in1[8];
    639	{ const u32 x19 = in1[7];
    640	{ const u32 x17 = in1[6];
    641	{ const u32 x15 = in1[5];
    642	{ const u32 x13 = in1[4];
    643	{ const u32 x11 = in1[3];
    644	{ const u32 x9 = in1[2];
    645	{ const u32 x7 = in1[1];
    646	{ const u32 x5 = in1[0];
    647	{ const u32 x38 = 0;
    648	{ const u32 x39 = 0;
    649	{ const u32 x37 = 0;
    650	{ const u32 x35 = 0;
    651	{ const u32 x33 = 0;
    652	{ const u32 x31 = 0;
    653	{ const u32 x29 = 0;
    654	{ const u32 x27 = 0;
    655	{ const u32 x25 = 0;
    656	{ const u32 x23 = 121666;
    657	{ u64 x40 = ((u64)x23 * x5);
    658	{ u64 x41 = (((u64)x23 * x7) + ((u64)x25 * x5));
    659	{ u64 x42 = ((((u64)(0x2 * x25) * x7) + ((u64)x23 * x9)) + ((u64)x27 * x5));
    660	{ u64 x43 = (((((u64)x25 * x9) + ((u64)x27 * x7)) + ((u64)x23 * x11)) + ((u64)x29 * x5));
    661	{ u64 x44 = (((((u64)x27 * x9) + (0x2 * (((u64)x25 * x11) + ((u64)x29 * x7)))) + ((u64)x23 * x13)) + ((u64)x31 * x5));
    662	{ u64 x45 = (((((((u64)x27 * x11) + ((u64)x29 * x9)) + ((u64)x25 * x13)) + ((u64)x31 * x7)) + ((u64)x23 * x15)) + ((u64)x33 * x5));
    663	{ u64 x46 = (((((0x2 * ((((u64)x29 * x11) + ((u64)x25 * x15)) + ((u64)x33 * x7))) + ((u64)x27 * x13)) + ((u64)x31 * x9)) + ((u64)x23 * x17)) + ((u64)x35 * x5));
    664	{ u64 x47 = (((((((((u64)x29 * x13) + ((u64)x31 * x11)) + ((u64)x27 * x15)) + ((u64)x33 * x9)) + ((u64)x25 * x17)) + ((u64)x35 * x7)) + ((u64)x23 * x19)) + ((u64)x37 * x5));
    665	{ u64 x48 = (((((((u64)x31 * x13) + (0x2 * (((((u64)x29 * x15) + ((u64)x33 * x11)) + ((u64)x25 * x19)) + ((u64)x37 * x7)))) + ((u64)x27 * x17)) + ((u64)x35 * x9)) + ((u64)x23 * x21)) + ((u64)x39 * x5));
    666	{ u64 x49 = (((((((((((u64)x31 * x15) + ((u64)x33 * x13)) + ((u64)x29 * x17)) + ((u64)x35 * x11)) + ((u64)x27 * x19)) + ((u64)x37 * x9)) + ((u64)x25 * x21)) + ((u64)x39 * x7)) + ((u64)x23 * x20)) + ((u64)x38 * x5));
    667	{ u64 x50 = (((((0x2 * ((((((u64)x33 * x15) + ((u64)x29 * x19)) + ((u64)x37 * x11)) + ((u64)x25 * x20)) + ((u64)x38 * x7))) + ((u64)x31 * x17)) + ((u64)x35 * x13)) + ((u64)x27 * x21)) + ((u64)x39 * x9));
    668	{ u64 x51 = (((((((((u64)x33 * x17) + ((u64)x35 * x15)) + ((u64)x31 * x19)) + ((u64)x37 * x13)) + ((u64)x29 * x21)) + ((u64)x39 * x11)) + ((u64)x27 * x20)) + ((u64)x38 * x9));
    669	{ u64 x52 = (((((u64)x35 * x17) + (0x2 * (((((u64)x33 * x19) + ((u64)x37 * x15)) + ((u64)x29 * x20)) + ((u64)x38 * x11)))) + ((u64)x31 * x21)) + ((u64)x39 * x13));
    670	{ u64 x53 = (((((((u64)x35 * x19) + ((u64)x37 * x17)) + ((u64)x33 * x21)) + ((u64)x39 * x15)) + ((u64)x31 * x20)) + ((u64)x38 * x13));
    671	{ u64 x54 = (((0x2 * ((((u64)x37 * x19) + ((u64)x33 * x20)) + ((u64)x38 * x15))) + ((u64)x35 * x21)) + ((u64)x39 * x17));
    672	{ u64 x55 = (((((u64)x37 * x21) + ((u64)x39 * x19)) + ((u64)x35 * x20)) + ((u64)x38 * x17));
    673	{ u64 x56 = (((u64)x39 * x21) + (0x2 * (((u64)x37 * x20) + ((u64)x38 * x19))));
    674	{ u64 x57 = (((u64)x39 * x20) + ((u64)x38 * x21));
    675	{ u64 x58 = ((u64)(0x2 * x38) * x20);
    676	{ u64 x59 = (x48 + (x58 << 0x4));
    677	{ u64 x60 = (x59 + (x58 << 0x1));
    678	{ u64 x61 = (x60 + x58);
    679	{ u64 x62 = (x47 + (x57 << 0x4));
    680	{ u64 x63 = (x62 + (x57 << 0x1));
    681	{ u64 x64 = (x63 + x57);
    682	{ u64 x65 = (x46 + (x56 << 0x4));
    683	{ u64 x66 = (x65 + (x56 << 0x1));
    684	{ u64 x67 = (x66 + x56);
    685	{ u64 x68 = (x45 + (x55 << 0x4));
    686	{ u64 x69 = (x68 + (x55 << 0x1));
    687	{ u64 x70 = (x69 + x55);
    688	{ u64 x71 = (x44 + (x54 << 0x4));
    689	{ u64 x72 = (x71 + (x54 << 0x1));
    690	{ u64 x73 = (x72 + x54);
    691	{ u64 x74 = (x43 + (x53 << 0x4));
    692	{ u64 x75 = (x74 + (x53 << 0x1));
    693	{ u64 x76 = (x75 + x53);
    694	{ u64 x77 = (x42 + (x52 << 0x4));
    695	{ u64 x78 = (x77 + (x52 << 0x1));
    696	{ u64 x79 = (x78 + x52);
    697	{ u64 x80 = (x41 + (x51 << 0x4));
    698	{ u64 x81 = (x80 + (x51 << 0x1));
    699	{ u64 x82 = (x81 + x51);
    700	{ u64 x83 = (x40 + (x50 << 0x4));
    701	{ u64 x84 = (x83 + (x50 << 0x1));
    702	{ u64 x85 = (x84 + x50);
    703	{ u64 x86 = (x85 >> 0x1a);
    704	{ u32 x87 = ((u32)x85 & 0x3ffffff);
    705	{ u64 x88 = (x86 + x82);
    706	{ u64 x89 = (x88 >> 0x19);
    707	{ u32 x90 = ((u32)x88 & 0x1ffffff);
    708	{ u64 x91 = (x89 + x79);
    709	{ u64 x92 = (x91 >> 0x1a);
    710	{ u32 x93 = ((u32)x91 & 0x3ffffff);
    711	{ u64 x94 = (x92 + x76);
    712	{ u64 x95 = (x94 >> 0x19);
    713	{ u32 x96 = ((u32)x94 & 0x1ffffff);
    714	{ u64 x97 = (x95 + x73);
    715	{ u64 x98 = (x97 >> 0x1a);
    716	{ u32 x99 = ((u32)x97 & 0x3ffffff);
    717	{ u64 x100 = (x98 + x70);
    718	{ u64 x101 = (x100 >> 0x19);
    719	{ u32 x102 = ((u32)x100 & 0x1ffffff);
    720	{ u64 x103 = (x101 + x67);
    721	{ u64 x104 = (x103 >> 0x1a);
    722	{ u32 x105 = ((u32)x103 & 0x3ffffff);
    723	{ u64 x106 = (x104 + x64);
    724	{ u64 x107 = (x106 >> 0x19);
    725	{ u32 x108 = ((u32)x106 & 0x1ffffff);
    726	{ u64 x109 = (x107 + x61);
    727	{ u64 x110 = (x109 >> 0x1a);
    728	{ u32 x111 = ((u32)x109 & 0x3ffffff);
    729	{ u64 x112 = (x110 + x49);
    730	{ u64 x113 = (x112 >> 0x19);
    731	{ u32 x114 = ((u32)x112 & 0x1ffffff);
    732	{ u64 x115 = (x87 + (0x13 * x113));
    733	{ u32 x116 = (u32) (x115 >> 0x1a);
    734	{ u32 x117 = ((u32)x115 & 0x3ffffff);
    735	{ u32 x118 = (x116 + x90);
    736	{ u32 x119 = (x118 >> 0x19);
    737	{ u32 x120 = (x118 & 0x1ffffff);
    738	out[0] = x117;
    739	out[1] = x120;
    740	out[2] = (x119 + x93);
    741	out[3] = x96;
    742	out[4] = x99;
    743	out[5] = x102;
    744	out[6] = x105;
    745	out[7] = x108;
    746	out[8] = x111;
    747	out[9] = x114;
    748	}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}
    749}
    750
    751static __always_inline void fe_mul121666(fe *h, const fe_loose *f)
    752{
    753	fe_mul_121666_impl(h->v, f->v);
    754}
    755
    756void curve25519_generic(u8 out[CURVE25519_KEY_SIZE],
    757			const u8 scalar[CURVE25519_KEY_SIZE],
    758			const u8 point[CURVE25519_KEY_SIZE])
    759{
    760	fe x1, x2, z2, x3, z3;
    761	fe_loose x2l, z2l, x3l;
    762	unsigned swap = 0;
    763	int pos;
    764	u8 e[32];
    765
    766	memcpy(e, scalar, 32);
    767	curve25519_clamp_secret(e);
    768
    769	/* The following implementation was transcribed to Coq and proven to
    770	 * correspond to unary scalar multiplication in affine coordinates given
    771	 * that x1 != 0 is the x coordinate of some point on the curve. It was
    772	 * also checked in Coq that doing a ladderstep with x1 = x3 = 0 gives
    773	 * z2' = z3' = 0, and z2 = z3 = 0 gives z2' = z3' = 0. The statement was
    774	 * quantified over the underlying field, so it applies to Curve25519
    775	 * itself and the quadratic twist of Curve25519. It was not proven in
    776	 * Coq that prime-field arithmetic correctly simulates extension-field
    777	 * arithmetic on prime-field values. The decoding of the byte array
    778	 * representation of e was not considered.
    779	 *
    780	 * Specification of Montgomery curves in affine coordinates:
    781	 * <https://github.com/mit-plv/fiat-crypto/blob/2456d821825521f7e03e65882cc3521795b0320f/src/Spec/MontgomeryCurve.v#L27>
    782	 *
    783	 * Proof that these form a group that is isomorphic to a Weierstrass
    784	 * curve:
    785	 * <https://github.com/mit-plv/fiat-crypto/blob/2456d821825521f7e03e65882cc3521795b0320f/src/Curves/Montgomery/AffineProofs.v#L35>
    786	 *
    787	 * Coq transcription and correctness proof of the loop
    788	 * (where scalarbits=255):
    789	 * <https://github.com/mit-plv/fiat-crypto/blob/2456d821825521f7e03e65882cc3521795b0320f/src/Curves/Montgomery/XZ.v#L118>
    790	 * <https://github.com/mit-plv/fiat-crypto/blob/2456d821825521f7e03e65882cc3521795b0320f/src/Curves/Montgomery/XZProofs.v#L278>
    791	 * preconditions: 0 <= e < 2^255 (not necessarily e < order),
    792	 * fe_invert(0) = 0
    793	 */
    794	fe_frombytes(&x1, point);
    795	fe_1(&x2);
    796	fe_0(&z2);
    797	fe_copy(&x3, &x1);
    798	fe_1(&z3);
    799
    800	for (pos = 254; pos >= 0; --pos) {
    801		fe tmp0, tmp1;
    802		fe_loose tmp0l, tmp1l;
    803		/* loop invariant as of right before the test, for the case
    804		 * where x1 != 0:
    805		 *   pos >= -1; if z2 = 0 then x2 is nonzero; if z3 = 0 then x3
    806		 *   is nonzero
    807		 *   let r := e >> (pos+1) in the following equalities of
    808		 *   projective points:
    809		 *   to_xz (r*P)     === if swap then (x3, z3) else (x2, z2)
    810		 *   to_xz ((r+1)*P) === if swap then (x2, z2) else (x3, z3)
    811		 *   x1 is the nonzero x coordinate of the nonzero
    812		 *   point (r*P-(r+1)*P)
    813		 */
    814		unsigned b = 1 & (e[pos / 8] >> (pos & 7));
    815		swap ^= b;
    816		fe_cswap(&x2, &x3, swap);
    817		fe_cswap(&z2, &z3, swap);
    818		swap = b;
    819		/* Coq transcription of ladderstep formula (called from
    820		 * transcribed loop):
    821		 * <https://github.com/mit-plv/fiat-crypto/blob/2456d821825521f7e03e65882cc3521795b0320f/src/Curves/Montgomery/XZ.v#L89>
    822		 * <https://github.com/mit-plv/fiat-crypto/blob/2456d821825521f7e03e65882cc3521795b0320f/src/Curves/Montgomery/XZProofs.v#L131>
    823		 * x1 != 0 <https://github.com/mit-plv/fiat-crypto/blob/2456d821825521f7e03e65882cc3521795b0320f/src/Curves/Montgomery/XZProofs.v#L217>
    824		 * x1  = 0 <https://github.com/mit-plv/fiat-crypto/blob/2456d821825521f7e03e65882cc3521795b0320f/src/Curves/Montgomery/XZProofs.v#L147>
    825		 */
    826		fe_sub(&tmp0l, &x3, &z3);
    827		fe_sub(&tmp1l, &x2, &z2);
    828		fe_add(&x2l, &x2, &z2);
    829		fe_add(&z2l, &x3, &z3);
    830		fe_mul_tll(&z3, &tmp0l, &x2l);
    831		fe_mul_tll(&z2, &z2l, &tmp1l);
    832		fe_sq_tl(&tmp0, &tmp1l);
    833		fe_sq_tl(&tmp1, &x2l);
    834		fe_add(&x3l, &z3, &z2);
    835		fe_sub(&z2l, &z3, &z2);
    836		fe_mul_ttt(&x2, &tmp1, &tmp0);
    837		fe_sub(&tmp1l, &tmp1, &tmp0);
    838		fe_sq_tl(&z2, &z2l);
    839		fe_mul121666(&z3, &tmp1l);
    840		fe_sq_tl(&x3, &x3l);
    841		fe_add(&tmp0l, &tmp0, &z3);
    842		fe_mul_ttt(&z3, &x1, &z2);
    843		fe_mul_tll(&z2, &tmp1l, &tmp0l);
    844	}
    845	/* here pos=-1, so r=e, so to_xz (e*P) === if swap then (x3, z3)
    846	 * else (x2, z2)
    847	 */
    848	fe_cswap(&x2, &x3, swap);
    849	fe_cswap(&z2, &z3, swap);
    850
    851	fe_invert(&z2, &z2);
    852	fe_mul_ttt(&x2, &x2, &z2);
    853	fe_tobytes(out, &x2);
    854
    855	memzero_explicit(&x1, sizeof(x1));
    856	memzero_explicit(&x2, sizeof(x2));
    857	memzero_explicit(&z2, sizeof(z2));
    858	memzero_explicit(&x3, sizeof(x3));
    859	memzero_explicit(&z3, sizeof(z3));
    860	memzero_explicit(&x2l, sizeof(x2l));
    861	memzero_explicit(&z2l, sizeof(z2l));
    862	memzero_explicit(&x3l, sizeof(x3l));
    863	memzero_explicit(&e, sizeof(e));
    864}