1/* 2 * Twofish Cipher 8-way parallel algorithm (AVX/x86_64) 3 * 4 * Copyright (C) 2012 Johannes Goetzfried 5 * <Johannes.Goetzfried@informatik.stud.uni-erlangen.de> 6 * 7 * Copyright © 2012 Jussi Kivilinna <jussi.kivilinna@mbnet.fi> 8 * 9 * This program is free software; you can redistribute it and/or modify 10 * it under the terms of the GNU General Public License as published by 11 * the Free Software Foundation; either version 2 of the License, or 12 * (at your option) any later version. 13 * 14 * This program is distributed in the hope that it will be useful, 15 * but WITHOUT ANY WARRANTY; without even the implied warranty of 16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 17 * GNU General Public License for more details. 18 * 19 * You should have received a copy of the GNU General Public License 20 * along with this program; if not, write to the Free Software 21 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 22 * USA 23 * 24 */ 25 26.file "twofish-avx-x86_64-asm_64.S" 27.text 28 29/* structure of crypto context */ 30#define s0 0 31#define s1 1024 32#define s2 2048 33#define s3 3072 34#define w 4096 35#define k 4128 36 37/********************************************************************** 38 8-way AVX twofish 39 **********************************************************************/ 40#define CTX %rdi 41 42#define RA1 %xmm0 43#define RB1 %xmm1 44#define RC1 %xmm2 45#define RD1 %xmm3 46 47#define RA2 %xmm4 48#define RB2 %xmm5 49#define RC2 %xmm6 50#define RD2 %xmm7 51 52#define RX0 %xmm8 53#define RY0 %xmm9 54 55#define RX1 %xmm10 56#define RY1 %xmm11 57 58#define RK1 %xmm12 59#define RK2 %xmm13 60 61#define RT %xmm14 62#define RR %xmm15 63 64#define RID1 %rbp 65#define RID1d %ebp 66#define RID2 %rsi 67#define RID2d %esi 68 69#define RGI1 %rdx 70#define RGI1bl %dl 71#define RGI1bh %dh 72#define RGI2 %rcx 73#define RGI2bl %cl 74#define RGI2bh %ch 75 76#define RGI3 %rax 77#define RGI3bl %al 78#define RGI3bh %ah 79#define RGI4 %rbx 80#define RGI4bl %bl 81#define RGI4bh %bh 82 83#define RGS1 %r8 84#define RGS1d %r8d 85#define RGS2 %r9 86#define RGS2d %r9d 87#define RGS3 %r10 88#define RGS3d %r10d 89 90 91#define lookup_32bit(t0, t1, t2, t3, src, dst, interleave_op, il_reg) \ 92 movzbl src ## bl, RID1d; \ 93 movzbl src ## bh, RID2d; \ 94 shrq $16, src; \ 95 movl t0(CTX, RID1, 4), dst ## d; \ 96 movl t1(CTX, RID2, 4), RID2d; \ 97 movzbl src ## bl, RID1d; \ 98 xorl RID2d, dst ## d; \ 99 movzbl src ## bh, RID2d; \ 100 interleave_op(il_reg); \ 101 xorl t2(CTX, RID1, 4), dst ## d; \ 102 xorl t3(CTX, RID2, 4), dst ## d; 103 104#define dummy(d) /* do nothing */ 105 106#define shr_next(reg) \ 107 shrq $16, reg; 108 109#define G(gi1, gi2, x, t0, t1, t2, t3) \ 110 lookup_32bit(t0, t1, t2, t3, ##gi1, RGS1, shr_next, ##gi1); \ 111 lookup_32bit(t0, t1, t2, t3, ##gi2, RGS3, shr_next, ##gi2); \ 112 \ 113 lookup_32bit(t0, t1, t2, t3, ##gi1, RGS2, dummy, none); \ 114 shlq $32, RGS2; \ 115 orq RGS1, RGS2; \ 116 lookup_32bit(t0, t1, t2, t3, ##gi2, RGS1, dummy, none); \ 117 shlq $32, RGS1; \ 118 orq RGS1, RGS3; 119 120#define round_head_2(a, b, x1, y1, x2, y2) \ 121 vmovq b ## 1, RGI3; \ 122 vpextrq $1, b ## 1, RGI4; \ 123 \ 124 G(RGI1, RGI2, x1, s0, s1, s2, s3); \ 125 vmovq a ## 2, RGI1; \ 126 vpextrq $1, a ## 2, RGI2; \ 127 vmovq RGS2, x1; \ 128 vpinsrq $1, RGS3, x1, x1; \ 129 \ 130 G(RGI3, RGI4, y1, s1, s2, s3, s0); \ 131 vmovq b ## 2, RGI3; \ 132 vpextrq $1, b ## 2, RGI4; \ 133 vmovq RGS2, y1; \ 134 vpinsrq $1, RGS3, y1, y1; \ 135 \ 136 G(RGI1, RGI2, x2, s0, s1, s2, s3); \ 137 vmovq RGS2, x2; \ 138 vpinsrq $1, RGS3, x2, x2; \ 139 \ 140 G(RGI3, RGI4, y2, s1, s2, s3, s0); \ 141 vmovq RGS2, y2; \ 142 vpinsrq $1, RGS3, y2, y2; 143 144#define encround_tail(a, b, c, d, x, y, prerotate) \ 145 vpaddd x, y, x; \ 146 vpaddd x, RK1, RT;\ 147 prerotate(b); \ 148 vpxor RT, c, c; \ 149 vpaddd y, x, y; \ 150 vpaddd y, RK2, y; \ 151 vpsrld $1, c, RT; \ 152 vpslld $(32 - 1), c, c; \ 153 vpor c, RT, c; \ 154 vpxor d, y, d; \ 155 156#define decround_tail(a, b, c, d, x, y, prerotate) \ 157 vpaddd x, y, x; \ 158 vpaddd x, RK1, RT;\ 159 prerotate(a); \ 160 vpxor RT, c, c; \ 161 vpaddd y, x, y; \ 162 vpaddd y, RK2, y; \ 163 vpxor d, y, d; \ 164 vpsrld $1, d, y; \ 165 vpslld $(32 - 1), d, d; \ 166 vpor d, y, d; \ 167 168#define rotate_1l(x) \ 169 vpslld $1, x, RR; \ 170 vpsrld $(32 - 1), x, x; \ 171 vpor x, RR, x; 172 173#define preload_rgi(c) \ 174 vmovq c, RGI1; \ 175 vpextrq $1, c, RGI2; 176 177#define encrypt_round(n, a, b, c, d, preload, prerotate) \ 178 vbroadcastss (k+4*(2*(n)))(CTX), RK1; \ 179 vbroadcastss (k+4*(2*(n)+1))(CTX), RK2; \ 180 round_head_2(a, b, RX0, RY0, RX1, RY1); \ 181 encround_tail(a ## 1, b ## 1, c ## 1, d ## 1, RX0, RY0, prerotate); \ 182 preload(c ## 1); \ 183 encround_tail(a ## 2, b ## 2, c ## 2, d ## 2, RX1, RY1, prerotate); 184 185#define decrypt_round(n, a, b, c, d, preload, prerotate) \ 186 vbroadcastss (k+4*(2*(n)))(CTX), RK1; \ 187 vbroadcastss (k+4*(2*(n)+1))(CTX), RK2; \ 188 round_head_2(a, b, RX0, RY0, RX1, RY1); \ 189 decround_tail(a ## 1, b ## 1, c ## 1, d ## 1, RX0, RY0, prerotate); \ 190 preload(c ## 1); \ 191 decround_tail(a ## 2, b ## 2, c ## 2, d ## 2, RX1, RY1, prerotate); 192 193#define encrypt_cycle(n) \ 194 encrypt_round((2*n), RA, RB, RC, RD, preload_rgi, rotate_1l); \ 195 encrypt_round(((2*n) + 1), RC, RD, RA, RB, preload_rgi, rotate_1l); 196 197#define encrypt_cycle_last(n) \ 198 encrypt_round((2*n), RA, RB, RC, RD, preload_rgi, rotate_1l); \ 199 encrypt_round(((2*n) + 1), RC, RD, RA, RB, dummy, dummy); 200 201#define decrypt_cycle(n) \ 202 decrypt_round(((2*n) + 1), RC, RD, RA, RB, preload_rgi, rotate_1l); \ 203 decrypt_round((2*n), RA, RB, RC, RD, preload_rgi, rotate_1l); 204 205#define decrypt_cycle_last(n) \ 206 decrypt_round(((2*n) + 1), RC, RD, RA, RB, preload_rgi, rotate_1l); \ 207 decrypt_round((2*n), RA, RB, RC, RD, dummy, dummy); 208 209#define transpose_4x4(x0, x1, x2, x3, t0, t1, t2) \ 210 vpunpckldq x1, x0, t0; \ 211 vpunpckhdq x1, x0, t2; \ 212 vpunpckldq x3, x2, t1; \ 213 vpunpckhdq x3, x2, x3; \ 214 \ 215 vpunpcklqdq t1, t0, x0; \ 216 vpunpckhqdq t1, t0, x1; \ 217 vpunpcklqdq x3, t2, x2; \ 218 vpunpckhqdq x3, t2, x3; 219 220#define inpack_blocks(in, x0, x1, x2, x3, wkey, t0, t1, t2) \ 221 vpxor (0*4*4)(in), wkey, x0; \ 222 vpxor (1*4*4)(in), wkey, x1; \ 223 vpxor (2*4*4)(in), wkey, x2; \ 224 vpxor (3*4*4)(in), wkey, x3; \ 225 \ 226 transpose_4x4(x0, x1, x2, x3, t0, t1, t2) 227 228#define outunpack_blocks(out, x0, x1, x2, x3, wkey, t0, t1, t2) \ 229 transpose_4x4(x0, x1, x2, x3, t0, t1, t2) \ 230 \ 231 vpxor x0, wkey, x0; \ 232 vmovdqu x0, (0*4*4)(out); \ 233 vpxor x1, wkey, x1; \ 234 vmovdqu x1, (1*4*4)(out); \ 235 vpxor x2, wkey, x2; \ 236 vmovdqu x2, (2*4*4)(out); \ 237 vpxor x3, wkey, x3; \ 238 vmovdqu x3, (3*4*4)(out); 239 240#define outunpack_xor_blocks(out, x0, x1, x2, x3, wkey, t0, t1, t2) \ 241 transpose_4x4(x0, x1, x2, x3, t0, t1, t2) \ 242 \ 243 vpxor x0, wkey, x0; \ 244 vpxor (0*4*4)(out), x0, x0; \ 245 vmovdqu x0, (0*4*4)(out); \ 246 vpxor x1, wkey, x1; \ 247 vpxor (1*4*4)(out), x1, x1; \ 248 vmovdqu x1, (1*4*4)(out); \ 249 vpxor x2, wkey, x2; \ 250 vpxor (2*4*4)(out), x2, x2; \ 251 vmovdqu x2, (2*4*4)(out); \ 252 vpxor x3, wkey, x3; \ 253 vpxor (3*4*4)(out), x3, x3; \ 254 vmovdqu x3, (3*4*4)(out); 255 256.align 8 257.global __twofish_enc_blk_8way 258.type __twofish_enc_blk_8way,@function; 259 260__twofish_enc_blk_8way: 261 /* input: 262 * %rdi: ctx, CTX 263 * %rsi: dst 264 * %rdx: src 265 * %rcx: bool, if true: xor output 266 */ 267 268 pushq %rbp; 269 pushq %rbx; 270 pushq %rcx; 271 272 vmovdqu w(CTX), RK1; 273 274 leaq (4*4*4)(%rdx), %rax; 275 inpack_blocks(%rdx, RA1, RB1, RC1, RD1, RK1, RX0, RY0, RK2); 276 preload_rgi(RA1); 277 rotate_1l(RD1); 278 inpack_blocks(%rax, RA2, RB2, RC2, RD2, RK1, RX0, RY0, RK2); 279 rotate_1l(RD2); 280 281 movq %rsi, %r11; 282 283 encrypt_cycle(0); 284 encrypt_cycle(1); 285 encrypt_cycle(2); 286 encrypt_cycle(3); 287 encrypt_cycle(4); 288 encrypt_cycle(5); 289 encrypt_cycle(6); 290 encrypt_cycle_last(7); 291 292 vmovdqu (w+4*4)(CTX), RK1; 293 294 popq %rcx; 295 popq %rbx; 296 popq %rbp; 297 298 leaq (4*4*4)(%r11), %rax; 299 300 testb %cl, %cl; 301 jnz __enc_xor8; 302 303 outunpack_blocks(%r11, RC1, RD1, RA1, RB1, RK1, RX0, RY0, RK2); 304 outunpack_blocks(%rax, RC2, RD2, RA2, RB2, RK1, RX0, RY0, RK2); 305 306 ret; 307 308__enc_xor8: 309 outunpack_xor_blocks(%r11, RC1, RD1, RA1, RB1, RK1, RX0, RY0, RK2); 310 outunpack_xor_blocks(%rax, RC2, RD2, RA2, RB2, RK1, RX0, RY0, RK2); 311 312 ret; 313 314.align 8 315.global twofish_dec_blk_8way 316.type twofish_dec_blk_8way,@function; 317 318twofish_dec_blk_8way: 319 /* input: 320 * %rdi: ctx, CTX 321 * %rsi: dst 322 * %rdx: src 323 */ 324 325 pushq %rbp; 326 pushq %rbx; 327 328 vmovdqu (w+4*4)(CTX), RK1; 329 330 leaq (4*4*4)(%rdx), %rax; 331 inpack_blocks(%rdx, RC1, RD1, RA1, RB1, RK1, RX0, RY0, RK2); 332 preload_rgi(RC1); 333 rotate_1l(RA1); 334 inpack_blocks(%rax, RC2, RD2, RA2, RB2, RK1, RX0, RY0, RK2); 335 rotate_1l(RA2); 336 337 movq %rsi, %r11; 338 339 decrypt_cycle(7); 340 decrypt_cycle(6); 341 decrypt_cycle(5); 342 decrypt_cycle(4); 343 decrypt_cycle(3); 344 decrypt_cycle(2); 345 decrypt_cycle(1); 346 decrypt_cycle_last(0); 347 348 vmovdqu (w)(CTX), RK1; 349 350 popq %rbx; 351 popq %rbp; 352 353 leaq (4*4*4)(%r11), %rax; 354 outunpack_blocks(%r11, RA1, RB1, RC1, RD1, RK1, RX0, RY0, RK2); 355 outunpack_blocks(%rax, RA2, RB2, RC2, RD2, RK1, RX0, RY0, RK2); 356 357 ret; 358