-rw-r--r-- 8157 lib25519-20240321/crypto_pow/inv25519/donna_c64/pow.c raw
/* Copyright 2008, Google Inc.
* All rights reserved.
*
* Code released into the public domain.
*
* curve25519-donna: Curve25519 elliptic curve, public key function
* (excerpt for inversion)
*
* http://code.google.com/p/curve25519-donna/
*
* Adam Langley <agl@imperialviolet.org>
* Parts optimised by floodyberry
* Derived from public domain C code by Daniel J. Bernstein <djb@cr.yp.to>
*
* More information about curve25519 can be found here
* http://cr.yp.to/ecdh.html
*
* djb's sample implementation of curve25519 is written in a special assembly
* language called qhasm and uses the floating point registers.
*
* This is, almost, a clean room reimplementation from the curve25519 paper. It
* uses many of the tricks described therein. Only the crecip function is taken
* from the sample implementation.
*/
#include <string.h>
#include <stdint.h>
#include "crypto_pow.h"
typedef uint8_t u8;
typedef uint64_t limb;
typedef limb felem[5];
// This is a special gcc mode for 128-bit integers. It's implemented on 64-bit
// platforms only as far as I know.
typedef unsigned uint128_t __attribute__((mode(TI)));
#undef force_inline
#define force_inline __attribute__((always_inline)) inline
/* Multiply two numbers: output = in2 * in
*
* output must be distinct to both inputs. The inputs are reduced coefficient
* form, the output is not.
*
* Assumes that in[i] < 2**55 and likewise for in2.
* On return, output[i] < 2**52
*/
static void force_inline
fmul(felem output, const felem in2, const felem in) {
uint128_t t[5];
limb r0,r1,r2,r3,r4,s0,s1,s2,s3,s4,c;
r0 = in[0];
r1 = in[1];
r2 = in[2];
r3 = in[3];
r4 = in[4];
s0 = in2[0];
s1 = in2[1];
s2 = in2[2];
s3 = in2[3];
s4 = in2[4];
t[0] = ((uint128_t) r0) * s0;
t[1] = ((uint128_t) r0) * s1 + ((uint128_t) r1) * s0;
t[2] = ((uint128_t) r0) * s2 + ((uint128_t) r2) * s0 + ((uint128_t) r1) * s1;
t[3] = ((uint128_t) r0) * s3 + ((uint128_t) r3) * s0 + ((uint128_t) r1) * s2 + ((uint128_t) r2) * s1;
t[4] = ((uint128_t) r0) * s4 + ((uint128_t) r4) * s0 + ((uint128_t) r3) * s1 + ((uint128_t) r1) * s3 + ((uint128_t) r2) * s2;
r4 *= 19;
r1 *= 19;
r2 *= 19;
r3 *= 19;
t[0] += ((uint128_t) r4) * s1 + ((uint128_t) r1) * s4 + ((uint128_t) r2) * s3 + ((uint128_t) r3) * s2;
t[1] += ((uint128_t) r4) * s2 + ((uint128_t) r2) * s4 + ((uint128_t) r3) * s3;
t[2] += ((uint128_t) r4) * s3 + ((uint128_t) r3) * s4;
t[3] += ((uint128_t) r4) * s4;
r0 = (limb)t[0] & 0x7ffffffffffff; c = (limb)(t[0] >> 51);
t[1] += c; r1 = (limb)t[1] & 0x7ffffffffffff; c = (limb)(t[1] >> 51);
t[2] += c; r2 = (limb)t[2] & 0x7ffffffffffff; c = (limb)(t[2] >> 51);
t[3] += c; r3 = (limb)t[3] & 0x7ffffffffffff; c = (limb)(t[3] >> 51);
t[4] += c; r4 = (limb)t[4] & 0x7ffffffffffff; c = (limb)(t[4] >> 51);
r0 += c * 19; c = r0 >> 51; r0 = r0 & 0x7ffffffffffff;
r1 += c; c = r1 >> 51; r1 = r1 & 0x7ffffffffffff;
r2 += c;
output[0] = r0;
output[1] = r1;
output[2] = r2;
output[3] = r3;
output[4] = r4;
}
static void force_inline
fsquare_times(felem output, const felem in, limb count) {
uint128_t t[5];
limb r0,r1,r2,r3,r4,c;
limb d0,d1,d2,d4,d419;
r0 = in[0];
r1 = in[1];
r2 = in[2];
r3 = in[3];
r4 = in[4];
do {
d0 = r0 * 2;
d1 = r1 * 2;
d2 = r2 * 2 * 19;
d419 = r4 * 19;
d4 = d419 * 2;
t[0] = ((uint128_t) r0) * r0 + ((uint128_t) d4) * r1 + (((uint128_t) d2) * (r3 ));
t[1] = ((uint128_t) d0) * r1 + ((uint128_t) d4) * r2 + (((uint128_t) r3) * (r3 * 19));
t[2] = ((uint128_t) d0) * r2 + ((uint128_t) r1) * r1 + (((uint128_t) d4) * (r3 ));
t[3] = ((uint128_t) d0) * r3 + ((uint128_t) d1) * r2 + (((uint128_t) r4) * (d419 ));
t[4] = ((uint128_t) d0) * r4 + ((uint128_t) d1) * r3 + (((uint128_t) r2) * (r2 ));
r0 = (limb)t[0] & 0x7ffffffffffff; c = (limb)(t[0] >> 51);
t[1] += c; r1 = (limb)t[1] & 0x7ffffffffffff; c = (limb)(t[1] >> 51);
t[2] += c; r2 = (limb)t[2] & 0x7ffffffffffff; c = (limb)(t[2] >> 51);
t[3] += c; r3 = (limb)t[3] & 0x7ffffffffffff; c = (limb)(t[3] >> 51);
t[4] += c; r4 = (limb)t[4] & 0x7ffffffffffff; c = (limb)(t[4] >> 51);
r0 += c * 19; c = r0 >> 51; r0 = r0 & 0x7ffffffffffff;
r1 += c; c = r1 >> 51; r1 = r1 & 0x7ffffffffffff;
r2 += c;
} while(--count);
output[0] = r0;
output[1] = r1;
output[2] = r2;
output[3] = r3;
output[4] = r4;
}
/* Take a little-endian, 32-byte number and expand it into polynomial form */
static void
fexpand(limb *output, const u8 *in) {
output[0] = *((const uint64_t *)(in)) & 0x7ffffffffffff;
output[1] = (*((const uint64_t *)(in+6)) >> 3) & 0x7ffffffffffff;
output[2] = (*((const uint64_t *)(in+12)) >> 6) & 0x7ffffffffffff;
output[3] = (*((const uint64_t *)(in+19)) >> 1) & 0x7ffffffffffff;
output[4] = (*((const uint64_t *)(in+25)) >> 4) & 0x7ffffffffffff;
}
/* Take a fully reduced polynomial form number and contract it into a
* little-endian, 32-byte array
*/
static void
fcontract(u8 *output, const felem input) {
uint128_t t[5];
t[0] = input[0];
t[1] = input[1];
t[2] = input[2];
t[3] = input[3];
t[4] = input[4];
t[1] += t[0] >> 51; t[0] &= 0x7ffffffffffff;
t[2] += t[1] >> 51; t[1] &= 0x7ffffffffffff;
t[3] += t[2] >> 51; t[2] &= 0x7ffffffffffff;
t[4] += t[3] >> 51; t[3] &= 0x7ffffffffffff;
t[0] += 19 * (t[4] >> 51); t[4] &= 0x7ffffffffffff;
t[1] += t[0] >> 51; t[0] &= 0x7ffffffffffff;
t[2] += t[1] >> 51; t[1] &= 0x7ffffffffffff;
t[3] += t[2] >> 51; t[2] &= 0x7ffffffffffff;
t[4] += t[3] >> 51; t[3] &= 0x7ffffffffffff;
t[0] += 19 * (t[4] >> 51); t[4] &= 0x7ffffffffffff;
/* now t is between 0 and 2^255-1, properly carried. */
/* case 1: between 0 and 2^255-20. case 2: between 2^255-19 and 2^255-1. */
t[0] += 19;
t[1] += t[0] >> 51; t[0] &= 0x7ffffffffffff;
t[2] += t[1] >> 51; t[1] &= 0x7ffffffffffff;
t[3] += t[2] >> 51; t[2] &= 0x7ffffffffffff;
t[4] += t[3] >> 51; t[3] &= 0x7ffffffffffff;
t[0] += 19 * (t[4] >> 51); t[4] &= 0x7ffffffffffff;
/* now between 19 and 2^255-1 in both cases, and offset by 19. */
t[0] += 0x8000000000000 - 19;
t[1] += 0x8000000000000 - 1;
t[2] += 0x8000000000000 - 1;
t[3] += 0x8000000000000 - 1;
t[4] += 0x8000000000000 - 1;
/* now between 2^255 and 2^256-20, and offset by 2^255. */
t[1] += t[0] >> 51; t[0] &= 0x7ffffffffffff;
t[2] += t[1] >> 51; t[1] &= 0x7ffffffffffff;
t[3] += t[2] >> 51; t[2] &= 0x7ffffffffffff;
t[4] += t[3] >> 51; t[3] &= 0x7ffffffffffff;
t[4] &= 0x7ffffffffffff;
*((uint64_t *)(output)) = t[0] | (t[1] << 51);
*((uint64_t *)(output+8)) = (t[1] >> 13) | (t[2] << 38);
*((uint64_t *)(output+16)) = (t[2] >> 26) | (t[3] << 25);
*((uint64_t *)(output+24)) = (t[3] >> 39) | (t[4] << 12);
}
// -----------------------------------------------------------------------------
// Shamelessly copied from djb's code, tightened a little
// -----------------------------------------------------------------------------
static void
crecip(felem out, const felem z) {
felem a,t0,b,c;
/* 2 */ fsquare_times(a, z, 1); // a = 2
/* 8 */ fsquare_times(t0, a, 2);
/* 9 */ fmul(b, t0, z); // b = 9
/* 11 */ fmul(a, b, a); // a = 11
/* 22 */ fsquare_times(t0, a, 1);
/* 2^5 - 2^0 = 31 */ fmul(b, t0, b);
/* 2^10 - 2^5 */ fsquare_times(t0, b, 5);
/* 2^10 - 2^0 */ fmul(b, t0, b);
/* 2^20 - 2^10 */ fsquare_times(t0, b, 10);
/* 2^20 - 2^0 */ fmul(c, t0, b);
/* 2^40 - 2^20 */ fsquare_times(t0, c, 20);
/* 2^40 - 2^0 */ fmul(t0, t0, c);
/* 2^50 - 2^10 */ fsquare_times(t0, t0, 10);
/* 2^50 - 2^0 */ fmul(b, t0, b);
/* 2^100 - 2^50 */ fsquare_times(t0, b, 50);
/* 2^100 - 2^0 */ fmul(c, t0, b);
/* 2^200 - 2^100 */ fsquare_times(t0, c, 100);
/* 2^200 - 2^0 */ fmul(t0, t0, c);
/* 2^250 - 2^50 */ fsquare_times(t0, t0, 50);
/* 2^250 - 2^0 */ fmul(t0, t0, b);
/* 2^255 - 2^5 */ fsquare_times(t0, t0, 5);
/* 2^255 - 21 */ fmul(out, t0, a);
}
void
crypto_pow(u8 *q, const u8 *p) {
limb x[5];
fexpand(x, p);
crecip(x, x);
fcontract(q, x);
}