org.bouncycastle.math.ec.rfc8032.Scalar448 Maven / Gradle / Ivy
Go to download
Show more of this group Show more artifacts with this name
Show all versions of bcprov-jdk15to18 Show documentation
Show all versions of bcprov-jdk15to18 Show documentation
The Bouncy Castle Crypto package is a Java implementation of cryptographic algorithms. This jar contains JCE provider and lightweight API for the Bouncy Castle Cryptography APIs for JDK 1.5 to JDK 1.8.
package org.bouncycastle.math.ec.rfc8032;
import org.bouncycastle.math.raw.Nat;
import org.bouncycastle.math.raw.Nat448;
abstract class Scalar448
{
static final int SIZE = 14;
private static final int SCALAR_BYTES = SIZE * 4 + 1;
private static final long M26L = 0x03FFFFFFL;
private static final long M28L = 0x0FFFFFFFL;
private static final long M32L = 0xFFFFFFFFL;
private static final int TARGET_LENGTH = 447;
private static final int[] L = new int[]{ 0xAB5844F3, 0x2378C292, 0x8DC58F55, 0x216CC272, 0xAED63690, 0xC44EDB49,
0x7CCA23E9, 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF, 0x3FFFFFFF };
private static final int[] LSq = new int[]{ 0x1BA1FEA9, 0xC1ADFBB8, 0x49E0A8B2, 0xB91BF537, 0xE764D815, 0x4525492B,
0xA2B8716D, 0x4AE17CF6, 0xBA3C47C4, 0xF1A9CC14, 0x7E4D070A, 0x92052BCB, 0x9F823B72, 0xC3402A93, 0x55AC2279,
0x91BC6149, 0x46E2C7AA, 0x10B66139, 0xD76B1B48, 0xE2276DA4, 0xBE6511F4, 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF,
0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF, 0x0FFFFFFF };
private static final int L_0 = 0x04A7BB0D; // L_0:26/24
private static final int L_1 = 0x0873D6D5; // L_1:27/23
private static final int L_2 = 0x0A70AADC; // L_2:27/26
private static final int L_3 = 0x03D8D723; // L_3:26/--
private static final int L_4 = 0x096FDE93; // L_4:27/25
private static final int L_5 = 0x0B65129C; // L_5:27/26
private static final int L_6 = 0x063BB124; // L_6:27/--
private static final int L_7 = 0x08335DC1; // L_7:27/22
private static final int L4_0 = 0x029EEC34; // L4_0:25/24
private static final int L4_1 = 0x01CF5B55; // L4_1:25/--
private static final int L4_2 = 0x09C2AB72; // L4_2:27/25
private static final int L4_3 = 0x0F635C8E; // L4_3:28/--
private static final int L4_4 = 0x05BF7A4C; // L4_4:26/25
private static final int L4_5 = 0x0D944A72; // L4_5:28/--
private static final int L4_6 = 0x08EEC492; // L4_6:27/24
private static final int L4_7 = 0x20CD7705; // L4_7:29/24
static boolean checkVar(byte[] s, int[] n)
{
if (s[SCALAR_BYTES - 1] != 0x00)
{
return false;
}
decode(s, n);
return !Nat.gte(SIZE, n, L);
}
static void decode(byte[] k, int[] n)
{
// assert k[SCALAR_BYTES - 1] == 0x00;
Codec.decode32(k, 0, n, 0, SIZE);
}
static void getOrderWnafVar(int width, byte[] ws)
{
Wnaf.getSignedVar(L, width, ws);
}
static void multiply225Var(int[] x, int[] y225, int[] z)
{
// assert y225[7] >> 31 == y225[7];
int[] tt = new int[22];
Nat.mul(y225, 0, 8, x, 0, SIZE, tt, 0);
if (y225[7] < 0)
{
Nat.addTo(SIZE, L, 0, tt, 8);
Nat.subFrom(SIZE, x, 0, tt, 8);
}
byte[] bytes = new byte[88];
Codec.encode32(tt, 0, 22, bytes, 0);
byte[] r = reduce704(bytes);
decode(r, z);
}
static byte[] reduce704(byte[] n)
{
long x00 = Codec.decode32(n, 0) & M32L; // x00:32/--
long x01 = (Codec.decode24(n, 4) << 4) & M32L; // x01:28/--
long x02 = Codec.decode32(n, 7) & M32L; // x02:32/--
long x03 = (Codec.decode24(n, 11) << 4) & M32L; // x03:28/--
long x04 = Codec.decode32(n, 14) & M32L; // x04:32/--
long x05 = (Codec.decode24(n, 18) << 4) & M32L; // x05:28/--
long x06 = Codec.decode32(n, 21) & M32L; // x06:32/--
long x07 = (Codec.decode24(n, 25) << 4) & M32L; // x07:28/--
long x08 = Codec.decode32(n, 28) & M32L; // x08:32/--
long x09 = (Codec.decode24(n, 32) << 4) & M32L; // x09:28/--
long x10 = Codec.decode32(n, 35) & M32L; // x10:32/--
long x11 = (Codec.decode24(n, 39) << 4) & M32L; // x11:28/--
long x12 = Codec.decode32(n, 42) & M32L; // x12:32/--
long x13 = (Codec.decode24(n, 46) << 4) & M32L; // x13:28/--
long x14 = Codec.decode32(n, 49) & M32L; // x14:32/--
long x15 = (Codec.decode24(n, 53) << 4) & M32L; // x15:28/--
long x16 = Codec.decode32(n, 56) & M32L; // x16:32/--
long x17 = (Codec.decode24(n, 60) << 4) & M32L; // x17:28/--
long x18 = Codec.decode32(n, 63) & M32L; // x18:32/--
long x19 = (Codec.decode24(n, 67) << 4) & M32L; // x19:28/--
long x20 = Codec.decode32(n, 70) & M32L; // x20:32/--
long x21 = (Codec.decode24(n, 74) << 4) & M32L; // x21:28/--
long x22 = Codec.decode32(n, 77) & M32L; // x22:32/--
long x23 = (Codec.decode24(n, 81) << 4) & M32L; // x23:28/--
long x24 = Codec.decode32(n, 84) & M32L; // x24:32/--
long x25 = 0L & M32L;
x25 += (x24 >>> 28); x24 &= M28L; // x25:28/--, x24:28/--
x09 += x25 * L4_0; // x09:54/--
x10 += x25 * L4_1; // x10:54/53
x11 += x25 * L4_2; // x11:56/--
x12 += x25 * L4_3; // x12:57/--
x13 += x25 * L4_4; // x13:57/55
x14 += x25 * L4_5; // x14:58/--
x15 += x25 * L4_6; // x15:58/56
x16 += x25 * L4_7; // x16:59/--
x21 += (x20 >>> 28); x20 &= M28L; // x21:58/--, x20:28/--
x22 += (x21 >>> 28); x21 &= M28L; // x22:57/54, x21:28/--
x23 += (x22 >>> 28); x22 &= M28L; // x23:45/42, x22:28/--
x24 += (x23 >>> 28); x23 &= M28L; // x24:28/18, x23:28/--
x08 += x24 * L4_0; // x08:54/--
x09 += x24 * L4_1; // x09:55/--
x10 += x24 * L4_2; // x10:56/46
x11 += x24 * L4_3; // x11:57/46
x12 += x24 * L4_4; // x12:57/55
x13 += x24 * L4_5; // x13:58/--
x14 += x24 * L4_6; // x14:58/56
x15 += x24 * L4_7; // x15:59/--
x07 += x23 * L4_0; // x07:54/--
x08 += x23 * L4_1; // x08:54/53
x09 += x23 * L4_2; // x09:56/53
x10 += x23 * L4_3; // x10:57/46
x11 += x23 * L4_4; // x11:57/55
x12 += x23 * L4_5; // x12:58/--
x13 += x23 * L4_6; // x13:58/56
x14 += x23 * L4_7; // x14:59/--
x06 += x22 * L4_0; // x06:54/--
x07 += x22 * L4_1; // x07:54/53
x08 += x22 * L4_2; // x08:56/--
x09 += x22 * L4_3; // x09:57/53
x10 += x22 * L4_4; // x10:57/55
x11 += x22 * L4_5; // x11:58/--
x12 += x22 * L4_6; // x12:58/56
x13 += x22 * L4_7; // x13:59/--
x18 += (x17 >>> 28); x17 &= M28L; // x18:59/31, x17:28/--
x19 += (x18 >>> 28); x18 &= M28L; // x19:58/54, x18:28/--
x20 += (x19 >>> 28); x19 &= M28L; // x20:30/29, x19:28/--
x21 += (x20 >>> 28); x20 &= M28L; // x21:28/03, x20:28/--
x05 += x21 * L4_0; // x05:54/--
x06 += x21 * L4_1; // x06:55/--
x07 += x21 * L4_2; // x07:56/31
x08 += x21 * L4_3; // x08:57/31
x09 += x21 * L4_4; // x09:57/56
x10 += x21 * L4_5; // x10:58/--
x11 += x21 * L4_6; // x11:58/56
x12 += x21 * L4_7; // x12:59/--
x04 += x20 * L4_0; // x04:54/--
x05 += x20 * L4_1; // x05:54/53
x06 += x20 * L4_2; // x06:56/53
x07 += x20 * L4_3; // x07:57/31
x08 += x20 * L4_4; // x08:57/55
x09 += x20 * L4_5; // x09:58/--
x10 += x20 * L4_6; // x10:58/56
x11 += x20 * L4_7; // x11:59/--
x03 += x19 * L4_0; // x03:54/--
x04 += x19 * L4_1; // x04:54/53
x05 += x19 * L4_2; // x05:56/--
x06 += x19 * L4_3; // x06:57/53
x07 += x19 * L4_4; // x07:57/55
x08 += x19 * L4_5; // x08:58/--
x09 += x19 * L4_6; // x09:58/56
x10 += x19 * L4_7; // x10:59/--
x15 += (x14 >>> 28); x14 &= M28L; // x15:59/31, x14:28/--
x16 += (x15 >>> 28); x15 &= M28L; // x16:59/32, x15:28/--
x17 += (x16 >>> 28); x16 &= M28L; // x17:31/29, x16:28/--
x18 += (x17 >>> 28); x17 &= M28L; // x18:28/04, x17:28/--
x02 += x18 * L4_0; // x02:54/--
x03 += x18 * L4_1; // x03:55/--
x04 += x18 * L4_2; // x04:56/32
x05 += x18 * L4_3; // x05:57/32
x06 += x18 * L4_4; // x06:57/56
x07 += x18 * L4_5; // x07:58/--
x08 += x18 * L4_6; // x08:58/56
x09 += x18 * L4_7; // x09:59/--
x01 += x17 * L4_0; // x01:54/--
x02 += x17 * L4_1; // x02:54/53
x03 += x17 * L4_2; // x03:56/53
x04 += x17 * L4_3; // x04:57/32
x05 += x17 * L4_4; // x05:57/55
x06 += x17 * L4_5; // x06:58/--
x07 += x17 * L4_6; // x07:58/56
x08 += x17 * L4_7; // x08:59/--
x16 *= 4;
x16 += (x15 >>> 26); x15 &= M26L;
x16 += 1; // x16:30/01
x00 += x16 * L_0;
x01 += x16 * L_1;
x02 += x16 * L_2;
x03 += x16 * L_3;
x04 += x16 * L_4;
x05 += x16 * L_5;
x06 += x16 * L_6;
x07 += x16 * L_7;
x01 += (x00 >>> 28); x00 &= M28L;
x02 += (x01 >>> 28); x01 &= M28L;
x03 += (x02 >>> 28); x02 &= M28L;
x04 += (x03 >>> 28); x03 &= M28L;
x05 += (x04 >>> 28); x04 &= M28L;
x06 += (x05 >>> 28); x05 &= M28L;
x07 += (x06 >>> 28); x06 &= M28L;
x08 += (x07 >>> 28); x07 &= M28L;
x09 += (x08 >>> 28); x08 &= M28L;
x10 += (x09 >>> 28); x09 &= M28L;
x11 += (x10 >>> 28); x10 &= M28L;
x12 += (x11 >>> 28); x11 &= M28L;
x13 += (x12 >>> 28); x12 &= M28L;
x14 += (x13 >>> 28); x13 &= M28L;
x15 += (x14 >>> 28); x14 &= M28L;
x16 = (x15 >>> 26); x15 &= M26L;
x16 -= 1;
// assert x16 == 0L || x16 == -1L;
x00 -= x16 & L_0;
x01 -= x16 & L_1;
x02 -= x16 & L_2;
x03 -= x16 & L_3;
x04 -= x16 & L_4;
x05 -= x16 & L_5;
x06 -= x16 & L_6;
x07 -= x16 & L_7;
x01 += (x00 >> 28); x00 &= M28L;
x02 += (x01 >> 28); x01 &= M28L;
x03 += (x02 >> 28); x02 &= M28L;
x04 += (x03 >> 28); x03 &= M28L;
x05 += (x04 >> 28); x04 &= M28L;
x06 += (x05 >> 28); x05 &= M28L;
x07 += (x06 >> 28); x06 &= M28L;
x08 += (x07 >> 28); x07 &= M28L;
x09 += (x08 >> 28); x08 &= M28L;
x10 += (x09 >> 28); x09 &= M28L;
x11 += (x10 >> 28); x10 &= M28L;
x12 += (x11 >> 28); x11 &= M28L;
x13 += (x12 >> 28); x12 &= M28L;
x14 += (x13 >> 28); x13 &= M28L;
x15 += (x14 >> 28); x14 &= M28L;
// assert x15 >>> 26 == 0L;
byte[] r = new byte[SCALAR_BYTES];
Codec.encode56(x00 | (x01 << 28), r, 0);
Codec.encode56(x02 | (x03 << 28), r, 7);
Codec.encode56(x04 | (x05 << 28), r, 14);
Codec.encode56(x06 | (x07 << 28), r, 21);
Codec.encode56(x08 | (x09 << 28), r, 28);
Codec.encode56(x10 | (x11 << 28), r, 35);
Codec.encode56(x12 | (x13 << 28), r, 42);
Codec.encode56(x14 | (x15 << 28), r, 49);
// r[SCALAR_BYTES - 1] = 0;
return r;
}
static byte[] reduce912(byte[] n)
{
long x00 = Codec.decode32(n, 0) & M32L; // x00:32/--
long x01 = (Codec.decode24(n, 4) << 4) & M32L; // x01:28/--
long x02 = Codec.decode32(n, 7) & M32L; // x02:32/--
long x03 = (Codec.decode24(n, 11) << 4) & M32L; // x03:28/--
long x04 = Codec.decode32(n, 14) & M32L; // x04:32/--
long x05 = (Codec.decode24(n, 18) << 4) & M32L; // x05:28/--
long x06 = Codec.decode32(n, 21) & M32L; // x06:32/--
long x07 = (Codec.decode24(n, 25) << 4) & M32L; // x07:28/--
long x08 = Codec.decode32(n, 28) & M32L; // x08:32/--
long x09 = (Codec.decode24(n, 32) << 4) & M32L; // x09:28/--
long x10 = Codec.decode32(n, 35) & M32L; // x10:32/--
long x11 = (Codec.decode24(n, 39) << 4) & M32L; // x11:28/--
long x12 = Codec.decode32(n, 42) & M32L; // x12:32/--
long x13 = (Codec.decode24(n, 46) << 4) & M32L; // x13:28/--
long x14 = Codec.decode32(n, 49) & M32L; // x14:32/--
long x15 = (Codec.decode24(n, 53) << 4) & M32L; // x15:28/--
long x16 = Codec.decode32(n, 56) & M32L; // x16:32/--
long x17 = (Codec.decode24(n, 60) << 4) & M32L; // x17:28/--
long x18 = Codec.decode32(n, 63) & M32L; // x18:32/--
long x19 = (Codec.decode24(n, 67) << 4) & M32L; // x19:28/--
long x20 = Codec.decode32(n, 70) & M32L; // x20:32/--
long x21 = (Codec.decode24(n, 74) << 4) & M32L; // x21:28/--
long x22 = Codec.decode32(n, 77) & M32L; // x22:32/--
long x23 = (Codec.decode24(n, 81) << 4) & M32L; // x23:28/--
long x24 = Codec.decode32(n, 84) & M32L; // x24:32/--
long x25 = (Codec.decode24(n, 88) << 4) & M32L; // x25:28/--
long x26 = Codec.decode32(n, 91) & M32L; // x26:32/--
long x27 = (Codec.decode24(n, 95) << 4) & M32L; // x27:28/--
long x28 = Codec.decode32(n, 98) & M32L; // x28:32/--
long x29 = (Codec.decode24(n, 102) << 4) & M32L; // x29:28/--
long x30 = Codec.decode32(n, 105) & M32L; // x30:32/--
long x31 = (Codec.decode24(n, 109) << 4) & M32L; // x31:28/--
long x32 = Codec.decode16(n, 112) & M32L; // x32:16/--
// x32 += (x31 >>> 28); x31 &= M28L;
x16 += x32 * L4_0; // x16:42/--
x17 += x32 * L4_1; // x17:41/28
x18 += x32 * L4_2; // x18:43/42
x19 += x32 * L4_3; // x19:44/28
x20 += x32 * L4_4; // x20:43/--
x21 += x32 * L4_5; // x21:44/28
x22 += x32 * L4_6; // x22:43/41
x23 += x32 * L4_7; // x23:45/41
x31 += (x30 >>> 28); x30 &= M28L; // x31:28/--, x30:28/--
x15 += x31 * L4_0; // x15:54/--
x16 += x31 * L4_1; // x16:53/42
x17 += x31 * L4_2; // x17:55/54
x18 += x31 * L4_3; // x18:56/44
x19 += x31 * L4_4; // x19:55/--
x20 += x31 * L4_5; // x20:56/43
x21 += x31 * L4_6; // x21:55/53
x22 += x31 * L4_7; // x22:57/53
// x30 += (x29 >>> 28); x29 &= M28L;
x14 += x30 * L4_0; // x14:54/--
x15 += x30 * L4_1; // x15:54/53
x16 += x30 * L4_2; // x16:56/--
x17 += x30 * L4_3; // x17:57/--
x18 += x30 * L4_4; // x18:56/55
x19 += x30 * L4_5; // x19:56/55
x20 += x30 * L4_6; // x20:57/--
x21 += x30 * L4_7; // x21:57/56
x29 += (x28 >>> 28); x28 &= M28L; // x29:28/--, x28:28/--
x13 += x29 * L4_0; // x13:54/--
x14 += x29 * L4_1; // x14:54/53
x15 += x29 * L4_2; // x15:56/--
x16 += x29 * L4_3; // x16:57/--
x17 += x29 * L4_4; // x17:57/55
x18 += x29 * L4_5; // x18:57/55
x19 += x29 * L4_6; // x19:57/52
x20 += x29 * L4_7; // x20:58/52
// x28 += (x27 >>> 28); x27 &= M28L;
x12 += x28 * L4_0; // x12:54/--
x13 += x28 * L4_1; // x13:54/53
x14 += x28 * L4_2; // x14:56/--
x15 += x28 * L4_3; // x15:57/--
x16 += x28 * L4_4; // x16:57/55
x17 += x28 * L4_5; // x17:58/--
x18 += x28 * L4_6; // x18:58/--
x19 += x28 * L4_7; // x19:58/53
x27 += (x26 >>> 28); x26 &= M28L; // x27:28/--, x26:28/--
x11 += x27 * L4_0; // x11:54/--
x12 += x27 * L4_1; // x12:54/53
x13 += x27 * L4_2; // x13:56/--
x14 += x27 * L4_3; // x14:57/--
x15 += x27 * L4_4; // x15:57/55
x16 += x27 * L4_5; // x16:58/--
x17 += x27 * L4_6; // x17:58/56
x18 += x27 * L4_7; // x18:59/--
// x26 += (x25 >>> 28); x25 &= M28L;
x10 += x26 * L4_0; // x10:54/--
x11 += x26 * L4_1; // x11:54/53
x12 += x26 * L4_2; // x12:56/--
x13 += x26 * L4_3; // x13:57/--
x14 += x26 * L4_4; // x14:57/55
x15 += x26 * L4_5; // x15:58/--
x16 += x26 * L4_6; // x16:58/56
x17 += x26 * L4_7; // x17:59/--
x25 += (x24 >>> 28); x24 &= M28L; // x25:28/--, x24:28/--
x09 += x25 * L4_0; // x09:54/--
x10 += x25 * L4_1; // x10:54/53
x11 += x25 * L4_2; // x11:56/--
x12 += x25 * L4_3; // x12:57/--
x13 += x25 * L4_4; // x13:57/55
x14 += x25 * L4_5; // x14:58/--
x15 += x25 * L4_6; // x15:58/56
x16 += x25 * L4_7; // x16:59/--
x21 += (x20 >>> 28); x20 &= M28L; // x21:58/--, x20:28/--
x22 += (x21 >>> 28); x21 &= M28L; // x22:57/54, x21:28/--
x23 += (x22 >>> 28); x22 &= M28L; // x23:45/42, x22:28/--
x24 += (x23 >>> 28); x23 &= M28L; // x24:28/18, x23:28/--
x08 += x24 * L4_0; // x08:54/--
x09 += x24 * L4_1; // x09:55/--
x10 += x24 * L4_2; // x10:56/46
x11 += x24 * L4_3; // x11:57/46
x12 += x24 * L4_4; // x12:57/55
x13 += x24 * L4_5; // x13:58/--
x14 += x24 * L4_6; // x14:58/56
x15 += x24 * L4_7; // x15:59/--
x07 += x23 * L4_0; // x07:54/--
x08 += x23 * L4_1; // x08:54/53
x09 += x23 * L4_2; // x09:56/53
x10 += x23 * L4_3; // x10:57/46
x11 += x23 * L4_4; // x11:57/55
x12 += x23 * L4_5; // x12:58/--
x13 += x23 * L4_6; // x13:58/56
x14 += x23 * L4_7; // x14:59/--
x06 += x22 * L4_0; // x06:54/--
x07 += x22 * L4_1; // x07:54/53
x08 += x22 * L4_2; // x08:56/--
x09 += x22 * L4_3; // x09:57/53
x10 += x22 * L4_4; // x10:57/55
x11 += x22 * L4_5; // x11:58/--
x12 += x22 * L4_6; // x12:58/56
x13 += x22 * L4_7; // x13:59/--
x18 += (x17 >>> 28); x17 &= M28L; // x18:59/31, x17:28/--
x19 += (x18 >>> 28); x18 &= M28L; // x19:58/54, x18:28/--
x20 += (x19 >>> 28); x19 &= M28L; // x20:30/29, x19:28/--
x21 += (x20 >>> 28); x20 &= M28L; // x21:28/03, x20:28/--
x05 += x21 * L4_0; // x05:54/--
x06 += x21 * L4_1; // x06:55/--
x07 += x21 * L4_2; // x07:56/31
x08 += x21 * L4_3; // x08:57/31
x09 += x21 * L4_4; // x09:57/56
x10 += x21 * L4_5; // x10:58/--
x11 += x21 * L4_6; // x11:58/56
x12 += x21 * L4_7; // x12:59/--
x04 += x20 * L4_0; // x04:54/--
x05 += x20 * L4_1; // x05:54/53
x06 += x20 * L4_2; // x06:56/53
x07 += x20 * L4_3; // x07:57/31
x08 += x20 * L4_4; // x08:57/55
x09 += x20 * L4_5; // x09:58/--
x10 += x20 * L4_6; // x10:58/56
x11 += x20 * L4_7; // x11:59/--
x03 += x19 * L4_0; // x03:54/--
x04 += x19 * L4_1; // x04:54/53
x05 += x19 * L4_2; // x05:56/--
x06 += x19 * L4_3; // x06:57/53
x07 += x19 * L4_4; // x07:57/55
x08 += x19 * L4_5; // x08:58/--
x09 += x19 * L4_6; // x09:58/56
x10 += x19 * L4_7; // x10:59/--
x15 += (x14 >>> 28); x14 &= M28L; // x15:59/31, x14:28/--
x16 += (x15 >>> 28); x15 &= M28L; // x16:59/32, x15:28/--
x17 += (x16 >>> 28); x16 &= M28L; // x17:31/29, x16:28/--
x18 += (x17 >>> 28); x17 &= M28L; // x18:28/04, x17:28/--
x02 += x18 * L4_0; // x02:54/--
x03 += x18 * L4_1; // x03:55/--
x04 += x18 * L4_2; // x04:56/32
x05 += x18 * L4_3; // x05:57/32
x06 += x18 * L4_4; // x06:57/56
x07 += x18 * L4_5; // x07:58/--
x08 += x18 * L4_6; // x08:58/56
x09 += x18 * L4_7; // x09:59/--
x01 += x17 * L4_0; // x01:54/--
x02 += x17 * L4_1; // x02:54/53
x03 += x17 * L4_2; // x03:56/53
x04 += x17 * L4_3; // x04:57/32
x05 += x17 * L4_4; // x05:57/55
x06 += x17 * L4_5; // x06:58/--
x07 += x17 * L4_6; // x07:58/56
x08 += x17 * L4_7; // x08:59/--
x16 *= 4;
x16 += (x15 >>> 26); x15 &= M26L;
x16 += 1; // x16:30/01
x00 += x16 * L_0;
x01 += x16 * L_1;
x02 += x16 * L_2;
x03 += x16 * L_3;
x04 += x16 * L_4;
x05 += x16 * L_5;
x06 += x16 * L_6;
x07 += x16 * L_7;
x01 += (x00 >>> 28); x00 &= M28L;
x02 += (x01 >>> 28); x01 &= M28L;
x03 += (x02 >>> 28); x02 &= M28L;
x04 += (x03 >>> 28); x03 &= M28L;
x05 += (x04 >>> 28); x04 &= M28L;
x06 += (x05 >>> 28); x05 &= M28L;
x07 += (x06 >>> 28); x06 &= M28L;
x08 += (x07 >>> 28); x07 &= M28L;
x09 += (x08 >>> 28); x08 &= M28L;
x10 += (x09 >>> 28); x09 &= M28L;
x11 += (x10 >>> 28); x10 &= M28L;
x12 += (x11 >>> 28); x11 &= M28L;
x13 += (x12 >>> 28); x12 &= M28L;
x14 += (x13 >>> 28); x13 &= M28L;
x15 += (x14 >>> 28); x14 &= M28L;
x16 = (x15 >>> 26); x15 &= M26L;
x16 -= 1;
// assert x16 == 0L || x16 == -1L;
x00 -= x16 & L_0;
x01 -= x16 & L_1;
x02 -= x16 & L_2;
x03 -= x16 & L_3;
x04 -= x16 & L_4;
x05 -= x16 & L_5;
x06 -= x16 & L_6;
x07 -= x16 & L_7;
x01 += (x00 >> 28); x00 &= M28L;
x02 += (x01 >> 28); x01 &= M28L;
x03 += (x02 >> 28); x02 &= M28L;
x04 += (x03 >> 28); x03 &= M28L;
x05 += (x04 >> 28); x04 &= M28L;
x06 += (x05 >> 28); x05 &= M28L;
x07 += (x06 >> 28); x06 &= M28L;
x08 += (x07 >> 28); x07 &= M28L;
x09 += (x08 >> 28); x08 &= M28L;
x10 += (x09 >> 28); x09 &= M28L;
x11 += (x10 >> 28); x10 &= M28L;
x12 += (x11 >> 28); x11 &= M28L;
x13 += (x12 >> 28); x12 &= M28L;
x14 += (x13 >> 28); x13 &= M28L;
x15 += (x14 >> 28); x14 &= M28L;
// assert x15 >>> 26 == 0L;
byte[] r = new byte[SCALAR_BYTES];
Codec.encode56(x00 | (x01 << 28), r, 0);
Codec.encode56(x02 | (x03 << 28), r, 7);
Codec.encode56(x04 | (x05 << 28), r, 14);
Codec.encode56(x06 | (x07 << 28), r, 21);
Codec.encode56(x08 | (x09 << 28), r, 28);
Codec.encode56(x10 | (x11 << 28), r, 35);
Codec.encode56(x12 | (x13 << 28), r, 42);
Codec.encode56(x14 | (x15 << 28), r, 49);
// r[SCALAR_BYTES - 1] = 0;
return r;
}
static boolean reduceBasisVar(int[] k, int[] z0, int[] z1)
{
/*
* Split scalar k into two half-size scalars z0 and z1, such that z1 * k == z0 mod L.
*
* See https://ia.cr/2020/454 (Pornin).
*/
int[] Nu = new int[28]; System.arraycopy(LSq, 0, Nu, 0, 28);
int[] Nv = new int[28]; Nat448.square(k, Nv); ++Nv[0];
int[] p = new int[28]; Nat448.mul(L, k, p);
int[] t = new int[28]; // temp array
int[] u0 = new int[8]; System.arraycopy(L, 0, u0, 0, 8);
int[] u1 = new int[8];
int[] v0 = new int[8]; System.arraycopy(k, 0, v0, 0, 8);
int[] v1 = new int[8]; v1[0] = 1;
// Conservative upper bound on the number of loop iterations needed
int iterations = TARGET_LENGTH * 4;
int last = 27;
int len_Nv = ScalarUtil.getBitLengthPositive(last, Nv);
while (len_Nv > TARGET_LENGTH)
{
if (--iterations < 0)
{
return false;
}
int len_p = ScalarUtil.getBitLength(last, p);
int s = len_p - len_Nv;
s &= ~(s >> 31);
if (p[last] < 0)
{
ScalarUtil.addShifted_NP(last, s, Nu, Nv, p, t);
ScalarUtil.addShifted_UV(7, s, u0, u1, v0, v1);
}
else
{
ScalarUtil.subShifted_NP(last, s, Nu, Nv, p, t);
ScalarUtil.subShifted_UV(7, s, u0, u1, v0, v1);
}
if (ScalarUtil.lessThan(last, Nu, Nv))
{
int[] t0 = u0; u0 = v0; v0 = t0;
int[] t1 = u1; u1 = v1; v1 = t1;
int[] tN = Nu; Nu = Nv; Nv = tN;
last = len_Nv >>> 5;
len_Nv = ScalarUtil.getBitLengthPositive(last, Nv);
}
}
// assert v0[7] >> 31 == v0[7];
// assert v1[7] >> 31 == v1[7];
// v1 * k == v0 mod L
System.arraycopy(v0, 0, z0, 0, 8);
System.arraycopy(v1, 0, z1, 0, 8);
return true;
}
static void toSignedDigits(int bits, int[] x, int[] z)
{
// assert 448 < bits && bits < 480;
// assert z.length > SIZE;
z[SIZE] = (1 << (bits - 448))
+ Nat.cadd(SIZE, ~x[0] & 1, x, L, z);
// int c =
Nat.shiftDownBit(SIZE + 1, z, 0);
// assert c == (1 << 31);
}
}