// SPDX-License-Identifier: MPL-2.0 // (c) Hare authors // Constant time aes implementation optimized for 64bit CPUs. // The code was ported from BearSSL, which contained the following notice: // // Copyright (c) 2016 Thomas Pornin // // Permission is hereby granted, free of charge, to any person obtaining // a copy of this software and associated documentation files (the // "Software"), to deal in the Software without restriction, including // without limitation the rights to use, copy, modify, merge, publish, // distribute, sublicense, and/or sell copies of the Software, and to // permit persons to whom the Software is furnished to do so, subject to // the following conditions: // // The above copyright notice and this permission notice shall be // included in all copies or substantial portions of the Software. // // THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, // EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF // MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND // NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS // BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN // ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN // CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE // SOFTWARE. use crypto::cipher; use crypto::cipher::{blocksz,nparallel}; use crypto::math; use endian; def CT64_EXPKEYSZ: size = 960; def CT64_NPARALLEL: size = 4; // Returns an AES [[crypto::cipher::block]] cipher implementation optimized for // constant time operation on 64-bit systems. // // The caller must call [[ct64_init]] to add a key to the cipher before using // the cipher, and must call [[crypto::cipher::finish]] when they are finished // using the cipher to securely erase any secret data stored in the cipher // state. fn ct64() block = block { vtable = &ct64_vtable, ... }; const ct64_vtable: cipher::blockvtable = cipher::blockvtable { blocksz = BLOCKSZ, nparallel = CT64_NPARALLEL, encrypt = &aes_ct64_encrypt, decrypt = &aes_ct64_decrypt, finish = &block_finish, }; // Initializes the ct64 AES implementation with an encryption key. fn ct64_init(cipher: *block, key: []u8) void = { let comp_skey: [30]u64 = [0...]; cipher.rounds = br_aes_ct64_keysched(comp_skey[..], key, len(key)); br_aes_ct64_skey_expand(ct64_expkey(cipher), cipher.rounds, comp_skey[..]); }; fn ct64_expkey(b: *block) []u64 = { return (b.expkey[..]: *[*]u64)[..len(b.expkey)/size(u64)]; }; // Combines up to 4 blocks and encrypts them in one run fn aes_ct64_encrypt(b: *cipher::block, dest: []u8, src: []u8) void = { let b = b: *block; assert(len(src) % blocksz(b) == 0 && (len(src) / blocksz(b)) <= nparallel(b), "invalid block size"); let nblocks = len(src) / blocksz(b); let q: [8]u64 = [0...]; let w: [16]u32 = [0...]; br_range_dec32le(w, src); for (let i = 0z; i < nblocks; i += 1) { br_aes_ct64_interleave_in(q[i..], q[(i + 4)..], w[(i << 2)..]); }; br_aes_ct64_ortho(q); br_aes_ct64_bitslice_encrypt(b.rounds, ct64_expkey(b), q); br_aes_ct64_ortho(q); for (let i = 0z; i < nblocks; i += 1) { br_aes_ct64_interleave_out(w[(i << 2)..], q[i], q[i + 4]); }; br_range_enc32le(dest, w); }; // Combines up to 4 blocks and decrypts them in one run fn aes_ct64_decrypt(b: *cipher::block, dest: []u8, src: []u8) void = { let b = b: *block; assert(len(src) % blocksz(b) == 0 && (len(src) / blocksz(b)) <= nparallel(b), "invalid block size"); const nblocks = len(src) / blocksz(b); let q: [8]u64 = [0...]; let w: [16]u32 = [0...]; br_range_dec32le(w, src); for (let i = 0z; i < nblocks; i += 1) { br_aes_ct64_interleave_in(q[i..], q[(i + 4)..], w[(i << 2)..]); }; br_aes_ct64_ortho(q); br_aes_ct64_bitslice_decrypt(b.rounds, ct64_expkey(b), q); br_aes_ct64_ortho(q); for (let i = 0z; i < nblocks; i += 1) { br_aes_ct64_interleave_out(w[(i << 2)..], q[i], q[i + 4]); }; br_range_enc32le(dest, w); }; // see br_aes_ct64_ortho in src/inner.h of BearSSL fn br_aes_ct64_ortho(q: []u64) void = { swapn(0x5555555555555555, 0xAAAAAAAAAAAAAAAA, 1, &q[0], &q[1]); swapn(0x5555555555555555, 0xAAAAAAAAAAAAAAAA, 1, &q[2], &q[3]); swapn(0x5555555555555555, 0xAAAAAAAAAAAAAAAA, 1, &q[4], &q[5]); swapn(0x5555555555555555, 0xAAAAAAAAAAAAAAAA, 1, &q[6], &q[7]); swapn(0x3333333333333333, 0xCCCCCCCCCCCCCCCC, 2, &q[0], &q[2]); swapn(0x3333333333333333, 0xCCCCCCCCCCCCCCCC, 2, &q[1], &q[3]); swapn(0x3333333333333333, 0xCCCCCCCCCCCCCCCC, 2, &q[4], &q[6]); swapn(0x3333333333333333, 0xCCCCCCCCCCCCCCCC, 2, &q[5], &q[7]); swapn(0x0F0F0F0F0F0F0F0F, 0xF0F0F0F0F0F0F0F0, 4, &q[0], &q[4]); swapn(0x0F0F0F0F0F0F0F0F, 0xF0F0F0F0F0F0F0F0, 4, &q[1], &q[5]); swapn(0x0F0F0F0F0F0F0F0F, 0xF0F0F0F0F0F0F0F0, 4, &q[2], &q[6]); swapn(0x0F0F0F0F0F0F0F0F, 0xF0F0F0F0F0F0F0F0, 4, &q[3], &q[7]); }; // This is a macro in the C version. fn swapn(cl: u64, ch: u64, s: u32, x: *u64, y: *u64) void = { let a: u64 = *x, b: u64 = *y; *x = (a & cl) | ((b & cl) << s); *y = ((a & ch) >> s) | (b & ch); }; // see br_aes_ct64_interleave_in in src/inner.h of BearSSL fn br_aes_ct64_interleave_in(q0: []u64, q1: []u64, w: const []u32) void = { let x0 = 0u64, x1 = 0u64, x2 = 0u64, x3 = 0u64; x0 = w[0]; x1 = w[1]; x2 = w[2]; x3 = w[3]; x0 |= (x0 << 16); x1 |= (x1 << 16); x2 |= (x2 << 16); x3 |= (x3 << 16); x0 &= 0x0000FFFF0000FFFF; x1 &= 0x0000FFFF0000FFFF; x2 &= 0x0000FFFF0000FFFF; x3 &= 0x0000FFFF0000FFFF; x0 |= (x0 << 8); x1 |= (x1 << 8); x2 |= (x2 << 8); x3 |= (x3 << 8); x0 &= 0x00FF00FF00FF00FF; x1 &= 0x00FF00FF00FF00FF; x2 &= 0x00FF00FF00FF00FF; x3 &= 0x00FF00FF00FF00FF; q0[0] = x0 | (x2 << 8); q1[0] = x1 | (x3 << 8); }; // see br_aes_ct64_interleave_out in src/inner.h of BearSSL fn br_aes_ct64_interleave_out(w: []u32, q0: u64, q1: u64) void = { let x0 = 0u64, x1 = 0u64, x2 = 0u64, x3 = 0u64; x0 = q0 & 0x00FF00FF00FF00FF; x1 = q1 & 0x00FF00FF00FF00FF; x2 = (q0 >> 8) & 0x00FF00FF00FF00FF; x3 = (q1 >> 8) & 0x00FF00FF00FF00FF; x0 |= (x0 >> 8); x1 |= (x1 >> 8); x2 |= (x2 >> 8); x3 |= (x3 >> 8); x0 &= 0x0000FFFF0000FFFF; x1 &= 0x0000FFFF0000FFFF; x2 &= 0x0000FFFF0000FFFF; x3 &= 0x0000FFFF0000FFFF; w[0] = (x0 | (x0 >> 16)): u32; w[1] = (x1 | (x1 >> 16)): u32; w[2] = (x2 | (x2 >> 16)): u32; w[3] = (x3 | (x3 >> 16)): u32; }; // see br_aes_ct64_bitslice_Sbox in src/inner.h of BearSSL fn br_aes_ct64_bitslice_Sbox(q: []u64) void = { // This S-box implementation is a straightforward translation of // the circuit described by Boyar and Peralta in "A new // combinational logic minimization technique with applications // to cryptology" (https://eprint.iacr.org/2009/191.pdf). // // Note that variables x* (input) and s* (output) are numbered // in "reverse" order (x0 is the high bit, x7 is the low bit). let x0 = 0u64, x1 = 0u64, x2 = 0u64, x3 = 0u64, x4 = 0u64, x5 = 0u64, x6 = 0u64, x7 = 0u64; let y1 = 0u64, y2 = 0u64, y3 = 0u64, y4 = 0u64, y5 = 0u64, y6 = 0u64, y7 = 0u64, y8 = 0u64, y9 = 0u64, y10 = 0u64, y11 = 0u64, y12 = 0u64, y13 = 0u64, y14 = 0u64, y15 = 0u64, y16 = 0u64, y17 = 0u64, y18 = 0u64, y19 = 0u64, y20 = 0u64, y21 = 0u64; let z0 = 0u64, z1 = 0u64, z2 = 0u64, z3 = 0u64, z4 = 0u64, z5 = 0u64, z6 = 0u64, z7 = 0u64, z8 = 0u64, z9 = 0u64, z10 = 0u64, z11 = 0u64, z12 = 0u64, z13 = 0u64, z14 = 0u64, z15 = 0u64, z16 = 0u64, z17 = 0u64; let t0 = 0u64, t1 = 0u64, t2 = 0u64, t3 = 0u64, t4 = 0u64, t5 = 0u64, t6 = 0u64, t7 = 0u64, t8 = 0u64, t9 = 0u64, t10 = 0u64, t11 = 0u64, t12 = 0u64, t13 = 0u64, t14 = 0u64, t15 = 0u64, t16 = 0u64, t17 = 0u64, t18 = 0u64, t19 = 0u64, t20 = 0u64, t21 = 0u64, t22 = 0u64, t23 = 0u64, t24 = 0u64, t25 = 0u64, t26 = 0u64, t27 = 0u64, t28 = 0u64, t29 = 0u64, t30 = 0u64, t31 = 0u64, t32 = 0u64, t33 = 0u64, t34 = 0u64, t35 = 0u64, t36 = 0u64, t37 = 0u64, t38 = 0u64, t39 = 0u64, t40 = 0u64, t41 = 0u64, t42 = 0u64, t43 = 0u64, t44 = 0u64, t45 = 0u64, t46 = 0u64, t47 = 0u64, t48 = 0u64, t49 = 0u64, t50 = 0u64, t51 = 0u64, t52 = 0u64, t53 = 0u64, t54 = 0u64, t55 = 0u64, t56 = 0u64, t57 = 0u64, t58 = 0u64, t59 = 0u64, t60 = 0u64, t61 = 0u64, t62 = 0u64, t63 = 0u64, t64 = 0u64, t65 = 0u64, t66 = 0u64, t67 = 0u64; let s0 = 0u64, s1 = 0u64, s2 = 0u64, s3 = 0u64, s4 = 0u64, s5 = 0u64, s6 = 0u64, s7 = 0u64; x0 = q[7]; x1 = q[6]; x2 = q[5]; x3 = q[4]; x4 = q[3]; x5 = q[2]; x6 = q[1]; x7 = q[0]; // Top linear transformation. y14 = x3 ^ x5; y13 = x0 ^ x6; y9 = x0 ^ x3; y8 = x0 ^ x5; t0 = x1 ^ x2; y1 = t0 ^ x7; y4 = y1 ^ x3; y12 = y13 ^ y14; y2 = y1 ^ x0; y5 = y1 ^ x6; y3 = y5 ^ y8; t1 = x4 ^ y12; y15 = t1 ^ x5; y20 = t1 ^ x1; y6 = y15 ^ x7; y10 = y15 ^ t0; y11 = y20 ^ y9; y7 = x7 ^ y11; y17 = y10 ^ y11; y19 = y10 ^ y8; y16 = t0 ^ y11; y21 = y13 ^ y16; y18 = x0 ^ y16; // Non-linear section. t2 = y12 & y15; t3 = y3 & y6; t4 = t3 ^ t2; t5 = y4 & x7; t6 = t5 ^ t2; t7 = y13 & y16; t8 = y5 & y1; t9 = t8 ^ t7; t10 = y2 & y7; t11 = t10 ^ t7; t12 = y9 & y11; t13 = y14 & y17; t14 = t13 ^ t12; t15 = y8 & y10; t16 = t15 ^ t12; t17 = t4 ^ t14; t18 = t6 ^ t16; t19 = t9 ^ t14; t20 = t11 ^ t16; t21 = t17 ^ y20; t22 = t18 ^ y19; t23 = t19 ^ y21; t24 = t20 ^ y18; t25 = t21 ^ t22; t26 = t21 & t23; t27 = t24 ^ t26; t28 = t25 & t27; t29 = t28 ^ t22; t30 = t23 ^ t24; t31 = t22 ^ t26; t32 = t31 & t30; t33 = t32 ^ t24; t34 = t23 ^ t33; t35 = t27 ^ t33; t36 = t24 & t35; t37 = t36 ^ t34; t38 = t27 ^ t36; t39 = t29 & t38; t40 = t25 ^ t39; t41 = t40 ^ t37; t42 = t29 ^ t33; t43 = t29 ^ t40; t44 = t33 ^ t37; t45 = t42 ^ t41; z0 = t44 & y15; z1 = t37 & y6; z2 = t33 & x7; z3 = t43 & y16; z4 = t40 & y1; z5 = t29 & y7; z6 = t42 & y11; z7 = t45 & y17; z8 = t41 & y10; z9 = t44 & y12; z10 = t37 & y3; z11 = t33 & y4; z12 = t43 & y13; z13 = t40 & y5; z14 = t29 & y2; z15 = t42 & y9; z16 = t45 & y14; z17 = t41 & y8; // Bottom linear transformation. t46 = z15 ^ z16; t47 = z10 ^ z11; t48 = z5 ^ z13; t49 = z9 ^ z10; t50 = z2 ^ z12; t51 = z2 ^ z5; t52 = z7 ^ z8; t53 = z0 ^ z3; t54 = z6 ^ z7; t55 = z16 ^ z17; t56 = z12 ^ t48; t57 = t50 ^ t53; t58 = z4 ^ t46; t59 = z3 ^ t54; t60 = t46 ^ t57; t61 = z14 ^ t57; t62 = t52 ^ t58; t63 = t49 ^ t58; t64 = z4 ^ t59; t65 = t61 ^ t62; t66 = z1 ^ t63; s0 = t59 ^ t63; s6 = t56 ^ ~t62; s7 = t48 ^ ~t60; t67 = t64 ^ t65; s3 = t53 ^ t66; s4 = t51 ^ t66; s5 = t47 ^ t65; s1 = t64 ^ ~s3; s2 = t55 ^ ~t67; q[7] = s0; q[6] = s1; q[5] = s2; q[4] = s3; q[3] = s4; q[2] = s5; q[1] = s6; q[0] = s7; }; const rcon: []u8 = [ 0x01, 0x02, 0x04, 0x08, 0x10, 0x20, 0x40, 0x80, 0x1B, 0x36 ]; fn sub_word(x: u32) u32 = { let q: [8]u64 = [x, 0...]; br_aes_ct64_ortho(q); br_aes_ct64_bitslice_Sbox(q); br_aes_ct64_ortho(q); return q[0]: u32; }; // see br_aes_ct64_keysched in src/inner.h of BearSSL fn br_aes_ct64_keysched( comp_skey: []u64, key: const []u8, key_len: size, ) uint = { let num_rounds: uint = 0; let nk: int = 0, nkf: int = 0; let tmp: u32 = 0; let skey: [60]u32 = [0...]; switch (key_len) { case 16 => num_rounds = 10; case 24 => num_rounds = 12; case 32 => num_rounds = 14; case => assert(false, "invalid key length"); }; nk = (key_len >> 2): int; nkf = ((num_rounds + 1) << 2): int; br_range_dec32le(skey, key); tmp = skey[(key_len >> 2) - 1]; for (let i = nk, j = 0, k = 0; i < nkf; i += 1) { if (j == 0) { tmp = (tmp << 24) | (tmp >> 8); tmp = sub_word(tmp) ^ rcon[k]; } else if (nk > 6 && j == 4) { tmp = sub_word(tmp); }; tmp ^= skey[i - nk]; skey[i] = tmp; j += 1; if (j == nk) { j = 0; k += 1; }; }; for (let i = 0, j = 0; i < nkf) { let q: [8]u64 = [0...]; br_aes_ct64_interleave_in(q[0..], q[4..], skey[i..]); q[1] = q[0]; q[2] = q[0]; q[3] = q[0]; q[5] = q[4]; q[6] = q[4]; q[7] = q[4]; br_aes_ct64_ortho(q[..]); comp_skey[j + 0] = (q[0] & 0x1111111111111111) | (q[1] & 0x2222222222222222) | (q[2] & 0x4444444444444444) | (q[3] & 0x8888888888888888); comp_skey[j + 1] = (q[4] & 0x1111111111111111) | (q[5] & 0x2222222222222222) | (q[6] & 0x4444444444444444) | (q[7] & 0x8888888888888888); i += 4; j += 2; }; return num_rounds; }; fn br_range_dec32le(v: []u32, src: []u8) void = { for (let i = 0z; len(src) > 0; i += 1) { v[i] = endian::legetu32(src); src = src[4..]; }; }; fn br_range_enc32le(dest: []u8, w: []u32) void = { for (let i = 0z; len(dest) > 0; i += 1) { endian::leputu32(dest, w[i]); dest = dest[4..]; }; }; // see br_aes_ct64_skey_expand in src/inner.h of BearSSL fn br_aes_ct64_skey_expand( skey: []u64, num_rounds: uint, comp_skey: const []u64, ) void = { let n: uint = (num_rounds + 1) << 1; for (let u = 0u, v = 0u; u < n) { let x0 = 0u64, x1 = 0u64, x2 = 0u64, x3 = 0u64; x0 = comp_skey[u]; x1 = comp_skey[u]; x2 = comp_skey[u]; x3 = comp_skey[u]; x0 &= 0x1111111111111111; x1 &= 0x2222222222222222; x2 &= 0x4444444444444444; x3 &= 0x8888888888888888; x1 >>= 1; x2 >>= 2; x3 >>= 3; skey[v + 0] = (x0 << 4) - x0; skey[v + 1] = (x1 << 4) - x1; skey[v + 2] = (x2 << 4) - x2; skey[v + 3] = (x3 << 4) - x3; u += 1; v += 4; }; }; // aes_ct64_enc.c fn add_round_key(q: []u64, sk: const []u64) void = { q[0] ^= sk[0]; q[1] ^= sk[1]; q[2] ^= sk[2]; q[3] ^= sk[3]; q[4] ^= sk[4]; q[5] ^= sk[5]; q[6] ^= sk[6]; q[7] ^= sk[7]; }; fn shift_rows(q: []u64) void = { for (let i: int = 0; i < 8; i += 1) { let x: u64 = q[i]; q[i] = (x & 0x000000000000FFFF) | ((x & 0x00000000FFF00000) >> 4) | ((x & 0x00000000000F0000) << 12) | ((x & 0x0000FF0000000000) >> 8) | ((x & 0x000000FF00000000) << 8) | ((x & 0xF000000000000000) >> 12) | ((x & 0x0FFF000000000000) << 4); }; }; fn mix_columns(q: []u64) void = { let q0 = 0u64, q1 = 0u64, q2 = 0u64, q3 = 0u64, q4 = 0u64, q5 = 0u64, q6 = 0u64, q7 = 0u64; let r0 = 0u64, r1 = 0u64, r2 = 0u64, r3 = 0u64, r4 = 0u64, r5 = 0u64, r6 = 0u64, r7 = 0u64; q0 = q[0]; q1 = q[1]; q2 = q[2]; q3 = q[3]; q4 = q[4]; q5 = q[5]; q6 = q[6]; q7 = q[7]; r0 = (q0 >> 16) | (q0 << 48); r1 = (q1 >> 16) | (q1 << 48); r2 = (q2 >> 16) | (q2 << 48); r3 = (q3 >> 16) | (q3 << 48); r4 = (q4 >> 16) | (q4 << 48); r5 = (q5 >> 16) | (q5 << 48); r6 = (q6 >> 16) | (q6 << 48); r7 = (q7 >> 16) | (q7 << 48); q[0] = q7 ^ r7 ^ r0 ^ math::rotr64(q0 ^ r0, 32); q[1] = q0 ^ r0 ^ q7 ^ r7 ^ r1 ^ math::rotr64(q1 ^ r1, 32); q[2] = q1 ^ r1 ^ r2 ^ math::rotr64(q2 ^ r2, 32); q[3] = q2 ^ r2 ^ q7 ^ r7 ^ r3 ^ math::rotr64(q3 ^ r3, 32); q[4] = q3 ^ r3 ^ q7 ^ r7 ^ r4 ^ math::rotr64(q4 ^ r4, 32); q[5] = q4 ^ r4 ^ r5 ^ math::rotr64(q5 ^ r5, 32); q[6] = q5 ^ r5 ^ r6 ^ math::rotr64(q6 ^ r6, 32); q[7] = q6 ^ r6 ^ r7 ^ math::rotr64(q7 ^ r7, 32); }; // see br_aes_ct64_bitslice_encrypt in src/inner.h of BearSSL fn br_aes_ct64_bitslice_encrypt( num_rounds: uint, skey: const []u64, q: []u64, ) void = { add_round_key(q, skey); for (let u: uint = 1; u < num_rounds; u += 1) { br_aes_ct64_bitslice_Sbox(q); shift_rows(q); mix_columns(q); add_round_key(q, skey[(u << 3)..]); }; br_aes_ct64_bitslice_Sbox(q); shift_rows(q); add_round_key(q, skey[(num_rounds << 3)..]); }; // see br_aes_ct64_bitslice_invSbox in src/inner.h of BearSSL fn br_aes_ct64_bitslice_invSbox(q: []u64) void = { // See br_aes_ct_bitslice_invSbox(). This is the natural extension // to 64-bit registers. let q0 = 0u64, q1 = 0u64, q2 = 0u64, q3 = 0u64, q4 = 0u64, q5 = 0u64, q6 = 0u64, q7 = 0u64; q0 = ~q[0]; q1 = ~q[1]; q2 = q[2]; q3 = q[3]; q4 = q[4]; q5 = ~q[5]; q6 = ~q[6]; q7 = q[7]; q[7] = q1 ^ q4 ^ q6; q[6] = q0 ^ q3 ^ q5; q[5] = q7 ^ q2 ^ q4; q[4] = q6 ^ q1 ^ q3; q[3] = q5 ^ q0 ^ q2; q[2] = q4 ^ q7 ^ q1; q[1] = q3 ^ q6 ^ q0; q[0] = q2 ^ q5 ^ q7; br_aes_ct64_bitslice_Sbox(q); q0 = ~q[0]; q1 = ~q[1]; q2 = q[2]; q3 = q[3]; q4 = q[4]; q5 = ~q[5]; q6 = ~q[6]; q7 = q[7]; q[7] = q1 ^ q4 ^ q6; q[6] = q0 ^ q3 ^ q5; q[5] = q7 ^ q2 ^ q4; q[4] = q6 ^ q1 ^ q3; q[3] = q5 ^ q0 ^ q2; q[2] = q4 ^ q7 ^ q1; q[1] = q3 ^ q6 ^ q0; q[0] = q2 ^ q5 ^ q7; }; fn inv_shift_rows(q: []u64) void = { for (let i: int = 0; i < 8; i += 1) { let x: u64 = q[i]; q[i] = (x & 0x000000000000FFFF) | ((x & 0x000000000FFF0000) << 4) | ((x & 0x00000000F0000000) >> 12) | ((x & 0x000000FF00000000) << 8) | ((x & 0x0000FF0000000000) >> 8) | ((x & 0x000F000000000000) << 12) | ((x & 0xFFF0000000000000) >> 4); }; }; fn inv_mix_columns(q: []u64) void = { let q0 = 0u64, q1 = 0u64, q2 = 0u64, q3 = 0u64, q4 = 0u64, q5 = 0u64, q6 = 0u64, q7 = 0u64; let r0 = 0u64, r1 = 0u64, r2 = 0u64, r3 = 0u64, r4 = 0u64, r5 = 0u64, r6 = 0u64, r7 = 0u64; q0 = q[0]; q1 = q[1]; q2 = q[2]; q3 = q[3]; q4 = q[4]; q5 = q[5]; q6 = q[6]; q7 = q[7]; r0 = (q0 >> 16) | (q0 << 48); r1 = (q1 >> 16) | (q1 << 48); r2 = (q2 >> 16) | (q2 << 48); r3 = (q3 >> 16) | (q3 << 48); r4 = (q4 >> 16) | (q4 << 48); r5 = (q5 >> 16) | (q5 << 48); r6 = (q6 >> 16) | (q6 << 48); r7 = (q7 >> 16) | (q7 << 48); q[0] = q5 ^ q6 ^ q7 ^ r0 ^ r5 ^ r7 ^ math::rotr64(q0 ^ q5 ^ q6 ^ r0 ^ r5, 32); q[1] = q0 ^ q5 ^ r0 ^ r1 ^ r5 ^ r6 ^ r7 ^ math::rotr64(q1 ^ q5 ^ q7 ^ r1 ^ r5 ^ r6, 32); q[2] = q0 ^ q1 ^ q6 ^ r1 ^ r2 ^ r6 ^ r7 ^ math::rotr64(q0 ^ q2 ^ q6 ^ r2 ^ r6 ^ r7, 32); q[3] = q0 ^ q1 ^ q2 ^ q5 ^ q6 ^ r0 ^ r2 ^ r3 ^ r5 ^ math::rotr64(q0 ^ q1 ^ q3 ^ q5 ^ q6 ^ q7 ^ r0 ^ r3 ^ r5 ^ r7 , 32); q[4] = q1 ^ q2 ^ q3 ^ q5 ^ r1 ^ r3 ^ r4 ^ r5 ^ r6 ^ r7 ^ math::rotr64(q1 ^ q2 ^ q4 ^ q5 ^ q7 ^ r1 ^ r4 ^ r5 ^ r6, 32); q[5] = q2 ^ q3 ^ q4 ^ q6 ^ r2 ^ r4 ^ r5 ^ r6 ^ r7 ^ math::rotr64(q2 ^ q3 ^ q5 ^ q6 ^ r2 ^ r5 ^ r6 ^ r7, 32); q[6] = q3 ^ q4 ^ q5 ^ q7 ^ r3 ^ r5 ^ r6 ^ r7 ^ math::rotr64(q3 ^ q4 ^ q6 ^ q7 ^ r3 ^ r6 ^ r7, 32); q[7] = q4 ^ q5 ^ q6 ^ r4 ^ r6 ^ r7 ^ math::rotr64(q4 ^ q5 ^ q7 ^ r4 ^ r7, 32); }; // see br_aes_ct64_bitslice_decrypt in src/inner.h of BearSSL fn br_aes_ct64_bitslice_decrypt( num_rounds: size, skey: const []u64, q: []u64, ) void = { add_round_key(q, skey[(num_rounds << 3)..]); for (let u: size = num_rounds - 1; u > 0; u -= 1) { inv_shift_rows(q); br_aes_ct64_bitslice_invSbox(q); add_round_key(q, skey[(u << 3)..]); inv_mix_columns(q); }; inv_shift_rows(q); br_aes_ct64_bitslice_invSbox(q); add_round_key(q, skey); };