arch/x86/crypto/camellia-aesni-avx2-asm_64.S

   1 /*
   2  * x86_64/AVX2/AES-NI assembler implementation of Camellia
   3  *
   4  * Copyright © 2013 Jussi Kivilinna <jussi.kivilinna@iki.fi>
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License as published by
   8  * the Free Software Foundation; either version 2 of the License, or
   9  * (at your option) any later version.
  10  *
  11  */
  12
  13 #include <linux/linkage.h>
  14 #include <asm/frame.h>
  15
  16 #define CAMELLIA_TABLE_BYTE_LEN 272
  17
  18 /* struct camellia_ctx: */
  19 #define key_table 0
  20 #define key_length CAMELLIA_TABLE_BYTE_LEN
  21
  22 /* register macros */
  23 #define CTX %rdi
  24 #define RIO %r8
  25
  26 /**********************************************************************
  27   helper macros
  28  **********************************************************************/
  29 #define filter_8bit(x, lo_t, hi_t, mask4bit, tmp0) \
  30         vpand x, mask4bit, tmp0; \
  31         vpandn x, mask4bit, x; \
  32         vpsrld $4, x, x; \
  33         \
  34         vpshufb tmp0, lo_t, tmp0; \
  35         vpshufb x, hi_t, x; \
  36         vpxor tmp0, x, x;
  37
  38 #define ymm0_x xmm0
  39 #define ymm1_x xmm1
  40 #define ymm2_x xmm2
  41 #define ymm3_x xmm3
  42 #define ymm4_x xmm4
  43 #define ymm5_x xmm5
  44 #define ymm6_x xmm6
  45 #define ymm7_x xmm7
  46 #define ymm8_x xmm8
  47 #define ymm9_x xmm9
  48 #define ymm10_x xmm10
  49 #define ymm11_x xmm11
  50 #define ymm12_x xmm12
  51 #define ymm13_x xmm13
  52 #define ymm14_x xmm14
  53 #define ymm15_x xmm15
  54
  55 /**********************************************************************
  56   32-way camellia
  57  **********************************************************************/
  58
  59 /*
  60  * IN:
  61  *   x0..x7: byte-sliced AB state
  62  *   mem_cd: register pointer storing CD state
  63  *   key: index for key material
  64  * OUT:
  65  *   x0..x7: new byte-sliced CD state
  66  */
  67 #define roundsm32(x0, x1, x2, x3, x4, x5, x6, x7, t0, t1, t2, t3, t4, t5, t6, \
  68                   t7, mem_cd, key) \
  69         /* \
  70          * S-function with AES subbytes \
  71          */ \
  72         vbroadcasti128 .Linv_shift_row, t4; \
  73         vpbroadcastd .L0f0f0f0f, t7; \
  74         vbroadcasti128 .Lpre_tf_lo_s1, t5; \
  75         vbroadcasti128 .Lpre_tf_hi_s1, t6; \
  76         vbroadcasti128 .Lpre_tf_lo_s4, t2; \
  77         vbroadcasti128 .Lpre_tf_hi_s4, t3; \
  78         \
  79         /* AES inverse shift rows */ \
  80         vpshufb t4, x0, x0; \
  81         vpshufb t4, x7, x7; \
  82         vpshufb t4, x3, x3; \
  83         vpshufb t4, x6, x6; \
  84         vpshufb t4, x2, x2; \
  85         vpshufb t4, x5, x5; \
  86         vpshufb t4, x1, x1; \
  87         vpshufb t4, x4, x4; \
  88         \
  89         /* prefilter sboxes 1, 2 and 3 */ \
  90         /* prefilter sbox 4 */ \
  91         filter_8bit(x0, t5, t6, t7, t4); \
  92         filter_8bit(x7, t5, t6, t7, t4); \
  93         vextracti128 $1, x0, t0##_x; \
  94         vextracti128 $1, x7, t1##_x; \
  95         filter_8bit(x3, t2, t3, t7, t4); \
  96         filter_8bit(x6, t2, t3, t7, t4); \
  97         vextracti128 $1, x3, t3##_x; \
  98         vextracti128 $1, x6, t2##_x; \
  99         filter_8bit(x2, t5, t6, t7, t4); \
 100         filter_8bit(x5, t5, t6, t7, t4); \
 101         filter_8bit(x1, t5, t6, t7, t4); \
 102         filter_8bit(x4, t5, t6, t7, t4); \
 103         \
 104         vpxor t4##_x, t4##_x, t4##_x; \
 105         \
 106         /* AES subbytes + AES shift rows */ \
 107         vextracti128 $1, x2, t6##_x; \
 108         vextracti128 $1, x5, t5##_x; \
 109         vaesenclast t4##_x, x0##_x, x0##_x; \
 110         vaesenclast t4##_x, t0##_x, t0##_x; \
 111         vinserti128 $1, t0##_x, x0, x0; \
 112         vaesenclast t4##_x, x7##_x, x7##_x; \
 113         vaesenclast t4##_x, t1##_x, t1##_x; \
 114         vinserti128 $1, t1##_x, x7, x7; \
 115         vaesenclast t4##_x, x3##_x, x3##_x; \
 116         vaesenclast t4##_x, t3##_x, t3##_x; \
 117         vinserti128 $1, t3##_x, x3, x3; \
 118         vaesenclast t4##_x, x6##_x, x6##_x; \
 119         vaesenclast t4##_x, t2##_x, t2##_x; \
 120         vinserti128 $1, t2##_x, x6, x6; \
 121         vextracti128 $1, x1, t3##_x; \
 122         vextracti128 $1, x4, t2##_x; \
 123         vbroadcasti128 .Lpost_tf_lo_s1, t0; \
 124         vbroadcasti128 .Lpost_tf_hi_s1, t1; \
 125         vaesenclast t4##_x, x2##_x, x2##_x; \
 126         vaesenclast t4##_x, t6##_x, t6##_x; \
 127         vinserti128 $1, t6##_x, x2, x2; \
 128         vaesenclast t4##_x, x5##_x, x5##_x; \
 129         vaesenclast t4##_x, t5##_x, t5##_x; \
 130         vinserti128 $1, t5##_x, x5, x5; \
 131         vaesenclast t4##_x, x1##_x, x1##_x; \
 132         vaesenclast t4##_x, t3##_x, t3##_x; \
 133         vinserti128 $1, t3##_x, x1, x1; \
 134         vaesenclast t4##_x, x4##_x, x4##_x; \
 135         vaesenclast t4##_x, t2##_x, t2##_x; \
 136         vinserti128 $1, t2##_x, x4, x4; \
 137         \
 138         /* postfilter sboxes 1 and 4 */ \
 139         vbroadcasti128 .Lpost_tf_lo_s3, t2; \
 140         vbroadcasti128 .Lpost_tf_hi_s3, t3; \
 141         filter_8bit(x0, t0, t1, t7, t6); \
 142         filter_8bit(x7, t0, t1, t7, t6); \
 143         filter_8bit(x3, t0, t1, t7, t6); \
 144         filter_8bit(x6, t0, t1, t7, t6); \
 145         \
 146         /* postfilter sbox 3 */ \
 147         vbroadcasti128 .Lpost_tf_lo_s2, t4; \
 148         vbroadcasti128 .Lpost_tf_hi_s2, t5; \
 149         filter_8bit(x2, t2, t3, t7, t6); \
 150         filter_8bit(x5, t2, t3, t7, t6); \
 151         \
 152         vpbroadcastq key, t0; /* higher 64-bit duplicate ignored */ \
 153         \
 154         /* postfilter sbox 2 */ \
 155         filter_8bit(x1, t4, t5, t7, t2); \
 156         filter_8bit(x4, t4, t5, t7, t2); \
 157         vpxor t7, t7, t7; \
 158         \
 159         vpsrldq $1, t0, t1; \
 160         vpsrldq $2, t0, t2; \
 161         vpshufb t7, t1, t1; \
 162         vpsrldq $3, t0, t3; \
 163         \
 164         /* P-function */ \
 165         vpxor x5, x0, x0; \
 166         vpxor x6, x1, x1; \
 167         vpxor x7, x2, x2; \
 168         vpxor x4, x3, x3; \
 169         \
 170         vpshufb t7, t2, t2; \
 171         vpsrldq $4, t0, t4; \
 172         vpshufb t7, t3, t3; \
 173         vpsrldq $5, t0, t5; \
 174         vpshufb t7, t4, t4; \
 175         \
 176         vpxor x2, x4, x4; \
 177         vpxor x3, x5, x5; \
 178         vpxor x0, x6, x6; \
 179         vpxor x1, x7, x7; \
 180         \
 181         vpsrldq $6, t0, t6; \
 182         vpshufb t7, t5, t5; \
 183         vpshufb t7, t6, t6; \
 184         \
 185         vpxor x7, x0, x0; \
 186         vpxor x4, x1, x1; \
 187         vpxor x5, x2, x2; \
 188         vpxor x6, x3, x3; \
 189         \
 190         vpxor x3, x4, x4; \
 191         vpxor x0, x5, x5; \
 192         vpxor x1, x6, x6; \
 193         vpxor x2, x7, x7; /* note: high and low parts swapped */ \
 194         \
 195         /* Add key material and result to CD (x becomes new CD) */ \
 196         \
 197         vpxor t6, x1, x1; \
 198         vpxor 5 * 32(mem_cd), x1, x1; \
 199         \
 200         vpsrldq $7, t0, t6; \
 201         vpshufb t7, t0, t0; \
 202         vpshufb t7, t6, t7; \
 203         \
 204         vpxor t7, x0, x0; \
 205         vpxor 4 * 32(mem_cd), x0, x0; \
 206         \
 207         vpxor t5, x2, x2; \
 208         vpxor 6 * 32(mem_cd), x2, x2; \
 209         \
 210         vpxor t4, x3, x3; \
 211         vpxor 7 * 32(mem_cd), x3, x3; \
 212         \
 213         vpxor t3, x4, x4; \
 214         vpxor 0 * 32(mem_cd), x4, x4; \
 215         \
 216         vpxor t2, x5, x5; \
 217         vpxor 1 * 32(mem_cd), x5, x5; \
 218         \
 219         vpxor t1, x6, x6; \
 220         vpxor 2 * 32(mem_cd), x6, x6; \
 221         \
 222         vpxor t0, x7, x7; \
 223         vpxor 3 * 32(mem_cd), x7, x7;
 224
 225 /*
 226  * Size optimization... with inlined roundsm32 binary would be over 5 times
 227  * larger and would only marginally faster.
 228  */
 229 .align 8
 230 roundsm32_x0_x1_x2_x3_x4_x5_x6_x7_y0_y1_y2_y3_y4_y5_y6_y7_cd:
 231         roundsm32(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
 232                   %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14, %ymm15,
 233                   %rcx, (%r9));
 234         ret;
 235 ENDPROC(roundsm32_x0_x1_x2_x3_x4_x5_x6_x7_y0_y1_y2_y3_y4_y5_y6_y7_cd)
 236
 237 .align 8
 238 roundsm32_x4_x5_x6_x7_x0_x1_x2_x3_y4_y5_y6_y7_y0_y1_y2_y3_ab:
 239         roundsm32(%ymm4, %ymm5, %ymm6, %ymm7, %ymm0, %ymm1, %ymm2, %ymm3,
 240                   %ymm12, %ymm13, %ymm14, %ymm15, %ymm8, %ymm9, %ymm10, %ymm11,
 241                   %rax, (%r9));
 242         ret;
 243 ENDPROC(roundsm32_x4_x5_x6_x7_x0_x1_x2_x3_y4_y5_y6_y7_y0_y1_y2_y3_ab)
 244
 245 /*
 246  * IN/OUT:
 247  *  x0..x7: byte-sliced AB state preloaded
 248  *  mem_ab: byte-sliced AB state in memory
 249  *  mem_cb: byte-sliced CD state in memory
 250  */
 251 #define two_roundsm32(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
 252                       y6, y7, mem_ab, mem_cd, i, dir, store_ab) \
 253         leaq (key_table + (i) * 8)(CTX), %r9; \
 254         call roundsm32_x0_x1_x2_x3_x4_x5_x6_x7_y0_y1_y2_y3_y4_y5_y6_y7_cd; \
 255         \
 256         vmovdqu x0, 4 * 32(mem_cd); \
 257         vmovdqu x1, 5 * 32(mem_cd); \
 258         vmovdqu x2, 6 * 32(mem_cd); \
 259         vmovdqu x3, 7 * 32(mem_cd); \
 260         vmovdqu x4, 0 * 32(mem_cd); \
 261         vmovdqu x5, 1 * 32(mem_cd); \
 262         vmovdqu x6, 2 * 32(mem_cd); \
 263         vmovdqu x7, 3 * 32(mem_cd); \
 264         \
 265         leaq (key_table + ((i) + (dir)) * 8)(CTX), %r9; \
 266         call roundsm32_x4_x5_x6_x7_x0_x1_x2_x3_y4_y5_y6_y7_y0_y1_y2_y3_ab; \
 267         \
 268         store_ab(x0, x1, x2, x3, x4, x5, x6, x7, mem_ab);
 269
 270 #define dummy_store(x0, x1, x2, x3, x4, x5, x6, x7, mem_ab) /* do nothing */
 271
 272 #define store_ab_state(x0, x1, x2, x3, x4, x5, x6, x7, mem_ab) \
 273         /* Store new AB state */ \
 274         vmovdqu x4, 4 * 32(mem_ab); \
 275         vmovdqu x5, 5 * 32(mem_ab); \
 276         vmovdqu x6, 6 * 32(mem_ab); \
 277         vmovdqu x7, 7 * 32(mem_ab); \
 278         vmovdqu x0, 0 * 32(mem_ab); \
 279         vmovdqu x1, 1 * 32(mem_ab); \
 280         vmovdqu x2, 2 * 32(mem_ab); \
 281         vmovdqu x3, 3 * 32(mem_ab);
 282
 283 #define enc_rounds32(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
 284                       y6, y7, mem_ab, mem_cd, i) \
 285         two_roundsm32(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
 286                       y6, y7, mem_ab, mem_cd, (i) + 2, 1, store_ab_state); \
 287         two_roundsm32(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
 288                       y6, y7, mem_ab, mem_cd, (i) + 4, 1, store_ab_state); \
 289         two_roundsm32(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
 290                       y6, y7, mem_ab, mem_cd, (i) + 6, 1, dummy_store);
 291
 292 #define dec_rounds32(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
 293                       y6, y7, mem_ab, mem_cd, i) \
 294         two_roundsm32(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
 295                       y6, y7, mem_ab, mem_cd, (i) + 7, -1, store_ab_state); \
 296         two_roundsm32(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
 297                       y6, y7, mem_ab, mem_cd, (i) + 5, -1, store_ab_state); \
 298         two_roundsm32(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
 299                       y6, y7, mem_ab, mem_cd, (i) + 3, -1, dummy_store);
 300
 301 /*
 302  * IN:
 303  *  v0..3: byte-sliced 32-bit integers
 304  * OUT:
 305  *  v0..3: (IN <<< 1)
 306  */
 307 #define rol32_1_32(v0, v1, v2, v3, t0, t1, t2, zero) \
 308         vpcmpgtb v0, zero, t0; \
 309         vpaddb v0, v0, v0; \
 310         vpabsb t0, t0; \
 311         \
 312         vpcmpgtb v1, zero, t1; \
 313         vpaddb v1, v1, v1; \
 314         vpabsb t1, t1; \
 315         \
 316         vpcmpgtb v2, zero, t2; \
 317         vpaddb v2, v2, v2; \
 318         vpabsb t2, t2; \
 319         \
 320         vpor t0, v1, v1; \
 321         \
 322         vpcmpgtb v3, zero, t0; \
 323         vpaddb v3, v3, v3; \
 324         vpabsb t0, t0; \
 325         \
 326         vpor t1, v2, v2; \
 327         vpor t2, v3, v3; \
 328         vpor t0, v0, v0;
 329
 330 /*
 331  * IN:
 332  *   r: byte-sliced AB state in memory
 333  *   l: byte-sliced CD state in memory
 334  * OUT:
 335  *   x0..x7: new byte-sliced CD state
 336  */
 337 #define fls32(l, l0, l1, l2, l3, l4, l5, l6, l7, r, t0, t1, t2, t3, tt0, \
 338               tt1, tt2, tt3, kll, klr, krl, krr) \
 339         /* \
 340          * t0 = kll; \
 341          * t0 &= ll; \
 342          * lr ^= rol32(t0, 1); \
 343          */ \
 344         vpbroadcastd kll, t0; /* only lowest 32-bit used */ \
 345         vpxor tt0, tt0, tt0; \
 346         vpshufb tt0, t0, t3; \
 347         vpsrldq $1, t0, t0; \
 348         vpshufb tt0, t0, t2; \
 349         vpsrldq $1, t0, t0; \
 350         vpshufb tt0, t0, t1; \
 351         vpsrldq $1, t0, t0; \
 352         vpshufb tt0, t0, t0; \
 353         \
 354         vpand l0, t0, t0; \
 355         vpand l1, t1, t1; \
 356         vpand l2, t2, t2; \
 357         vpand l3, t3, t3; \
 358         \
 359         rol32_1_32(t3, t2, t1, t0, tt1, tt2, tt3, tt0); \
 360         \
 361         vpxor l4, t0, l4; \
 362         vpbroadcastd krr, t0; /* only lowest 32-bit used */ \
 363         vmovdqu l4, 4 * 32(l); \
 364         vpxor l5, t1, l5; \
 365         vmovdqu l5, 5 * 32(l); \
 366         vpxor l6, t2, l6; \
 367         vmovdqu l6, 6 * 32(l); \
 368         vpxor l7, t3, l7; \
 369         vmovdqu l7, 7 * 32(l); \
 370         \
 371         /* \
 372          * t2 = krr; \
 373          * t2 |= rr; \
 374          * rl ^= t2; \
 375          */ \
 376         \
 377         vpshufb tt0, t0, t3; \
 378         vpsrldq $1, t0, t0; \
 379         vpshufb tt0, t0, t2; \
 380         vpsrldq $1, t0, t0; \
 381         vpshufb tt0, t0, t1; \
 382         vpsrldq $1, t0, t0; \
 383         vpshufb tt0, t0, t0; \
 384         \
 385         vpor 4 * 32(r), t0, t0; \
 386         vpor 5 * 32(r), t1, t1; \
 387         vpor 6 * 32(r), t2, t2; \
 388         vpor 7 * 32(r), t3, t3; \
 389         \
 390         vpxor 0 * 32(r), t0, t0; \
 391         vpxor 1 * 32(r), t1, t1; \
 392         vpxor 2 * 32(r), t2, t2; \
 393         vpxor 3 * 32(r), t3, t3; \
 394         vmovdqu t0, 0 * 32(r); \
 395         vpbroadcastd krl, t0; /* only lowest 32-bit used */ \
 396         vmovdqu t1, 1 * 32(r); \
 397         vmovdqu t2, 2 * 32(r); \
 398         vmovdqu t3, 3 * 32(r); \
 399         \
 400         /* \
 401          * t2 = krl; \
 402          * t2 &= rl; \
 403          * rr ^= rol32(t2, 1); \
 404          */ \
 405         vpshufb tt0, t0, t3; \
 406         vpsrldq $1, t0, t0; \
 407         vpshufb tt0, t0, t2; \
 408         vpsrldq $1, t0, t0; \
 409         vpshufb tt0, t0, t1; \
 410         vpsrldq $1, t0, t0; \
 411         vpshufb tt0, t0, t0; \
 412         \
 413         vpand 0 * 32(r), t0, t0; \
 414         vpand 1 * 32(r), t1, t1; \
 415         vpand 2 * 32(r), t2, t2; \
 416         vpand 3 * 32(r), t3, t3; \
 417         \
 418         rol32_1_32(t3, t2, t1, t0, tt1, tt2, tt3, tt0); \
 419         \
 420         vpxor 4 * 32(r), t0, t0; \
 421         vpxor 5 * 32(r), t1, t1; \
 422         vpxor 6 * 32(r), t2, t2; \
 423         vpxor 7 * 32(r), t3, t3; \
 424         vmovdqu t0, 4 * 32(r); \
 425         vpbroadcastd klr, t0; /* only lowest 32-bit used */ \
 426         vmovdqu t1, 5 * 32(r); \
 427         vmovdqu t2, 6 * 32(r); \
 428         vmovdqu t3, 7 * 32(r); \
 429         \
 430         /* \
 431          * t0 = klr; \
 432          * t0 |= lr; \
 433          * ll ^= t0; \
 434          */ \
 435         \
 436         vpshufb tt0, t0, t3; \
 437         vpsrldq $1, t0, t0; \
 438         vpshufb tt0, t0, t2; \
 439         vpsrldq $1, t0, t0; \
 440         vpshufb tt0, t0, t1; \
 441         vpsrldq $1, t0, t0; \
 442         vpshufb tt0, t0, t0; \
 443         \
 444         vpor l4, t0, t0; \
 445         vpor l5, t1, t1; \
 446         vpor l6, t2, t2; \
 447         vpor l7, t3, t3; \
 448         \
 449         vpxor l0, t0, l0; \
 450         vmovdqu l0, 0 * 32(l); \
 451         vpxor l1, t1, l1; \
 452         vmovdqu l1, 1 * 32(l); \
 453         vpxor l2, t2, l2; \
 454         vmovdqu l2, 2 * 32(l); \
 455         vpxor l3, t3, l3; \
 456         vmovdqu l3, 3 * 32(l);
 457
 458 #define transpose_4x4(x0, x1, x2, x3, t1, t2) \
 459         vpunpckhdq x1, x0, t2; \
 460         vpunpckldq x1, x0, x0; \
 461         \
 462         vpunpckldq x3, x2, t1; \
 463         vpunpckhdq x3, x2, x2; \
 464         \
 465         vpunpckhqdq t1, x0, x1; \
 466         vpunpcklqdq t1, x0, x0; \
 467         \
 468         vpunpckhqdq x2, t2, x3; \
 469         vpunpcklqdq x2, t2, x2;
 470
 471 #define byteslice_16x16b_fast(a0, b0, c0, d0, a1, b1, c1, d1, a2, b2, c2, d2, \
 472                               a3, b3, c3, d3, st0, st1) \
 473         vmovdqu d2, st0; \
 474         vmovdqu d3, st1; \
 475         transpose_4x4(a0, a1, a2, a3, d2, d3); \
 476         transpose_4x4(b0, b1, b2, b3, d2, d3); \
 477         vmovdqu st0, d2; \
 478         vmovdqu st1, d3; \
 479         \
 480         vmovdqu a0, st0; \
 481         vmovdqu a1, st1; \
 482         transpose_4x4(c0, c1, c2, c3, a0, a1); \
 483         transpose_4x4(d0, d1, d2, d3, a0, a1); \
 484         \
 485         vbroadcasti128 .Lshufb_16x16b, a0; \
 486         vmovdqu st1, a1; \
 487         vpshufb a0, a2, a2; \
 488         vpshufb a0, a3, a3; \
 489         vpshufb a0, b0, b0; \
 490         vpshufb a0, b1, b1; \
 491         vpshufb a0, b2, b2; \
 492         vpshufb a0, b3, b3; \
 493         vpshufb a0, a1, a1; \
 494         vpshufb a0, c0, c0; \
 495         vpshufb a0, c1, c1; \
 496         vpshufb a0, c2, c2; \
 497         vpshufb a0, c3, c3; \
 498         vpshufb a0, d0, d0; \
 499         vpshufb a0, d1, d1; \
 500         vpshufb a0, d2, d2; \
 501         vpshufb a0, d3, d3; \
 502         vmovdqu d3, st1; \
 503         vmovdqu st0, d3; \
 504         vpshufb a0, d3, a0; \
 505         vmovdqu d2, st0; \
 506         \
 507         transpose_4x4(a0, b0, c0, d0, d2, d3); \
 508         transpose_4x4(a1, b1, c1, d1, d2, d3); \
 509         vmovdqu st0, d2; \
 510         vmovdqu st1, d3; \
 511         \
 512         vmovdqu b0, st0; \
 513         vmovdqu b1, st1; \
 514         transpose_4x4(a2, b2, c2, d2, b0, b1); \
 515         transpose_4x4(a3, b3, c3, d3, b0, b1); \
 516         vmovdqu st0, b0; \
 517         vmovdqu st1, b1; \
 518         /* does not adjust output bytes inside vectors */
 519
 520 /* load blocks to registers and apply pre-whitening */
 521 #define inpack32_pre(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
 522                      y6, y7, rio, key) \
 523         vpbroadcastq key, x0; \
 524         vpshufb .Lpack_bswap, x0, x0; \
 525         \
 526         vpxor 0 * 32(rio), x0, y7; \
 527         vpxor 1 * 32(rio), x0, y6; \
 528         vpxor 2 * 32(rio), x0, y5; \
 529         vpxor 3 * 32(rio), x0, y4; \
 530         vpxor 4 * 32(rio), x0, y3; \
 531         vpxor 5 * 32(rio), x0, y2; \
 532         vpxor 6 * 32(rio), x0, y1; \
 533         vpxor 7 * 32(rio), x0, y0; \
 534         vpxor 8 * 32(rio), x0, x7; \
 535         vpxor 9 * 32(rio), x0, x6; \
 536         vpxor 10 * 32(rio), x0, x5; \
 537         vpxor 11 * 32(rio), x0, x4; \
 538         vpxor 12 * 32(rio), x0, x3; \
 539         vpxor 13 * 32(rio), x0, x2; \
 540         vpxor 14 * 32(rio), x0, x1; \
 541         vpxor 15 * 32(rio), x0, x0;
 542
 543 /* byteslice pre-whitened blocks and store to temporary memory */
 544 #define inpack32_post(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
 545                       y6, y7, mem_ab, mem_cd) \
 546         byteslice_16x16b_fast(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, \
 547                               y4, y5, y6, y7, (mem_ab), (mem_cd)); \
 548         \
 549         vmovdqu x0, 0 * 32(mem_ab); \
 550         vmovdqu x1, 1 * 32(mem_ab); \
 551         vmovdqu x2, 2 * 32(mem_ab); \
 552         vmovdqu x3, 3 * 32(mem_ab); \
 553         vmovdqu x4, 4 * 32(mem_ab); \
 554         vmovdqu x5, 5 * 32(mem_ab); \
 555         vmovdqu x6, 6 * 32(mem_ab); \
 556         vmovdqu x7, 7 * 32(mem_ab); \
 557         vmovdqu y0, 0 * 32(mem_cd); \
 558         vmovdqu y1, 1 * 32(mem_cd); \
 559         vmovdqu y2, 2 * 32(mem_cd); \
 560         vmovdqu y3, 3 * 32(mem_cd); \
 561         vmovdqu y4, 4 * 32(mem_cd); \
 562         vmovdqu y5, 5 * 32(mem_cd); \
 563         vmovdqu y6, 6 * 32(mem_cd); \
 564         vmovdqu y7, 7 * 32(mem_cd);
 565
 566 /* de-byteslice, apply post-whitening and store blocks */
 567 #define outunpack32(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, \
 568                     y5, y6, y7, key, stack_tmp0, stack_tmp1) \
 569         byteslice_16x16b_fast(y0, y4, x0, x4, y1, y5, x1, x5, y2, y6, x2, x6, \
 570                               y3, y7, x3, x7, stack_tmp0, stack_tmp1); \
 571         \
 572         vmovdqu x0, stack_tmp0; \
 573         \
 574         vpbroadcastq key, x0; \
 575         vpshufb .Lpack_bswap, x0, x0; \
 576         \
 577         vpxor x0, y7, y7; \
 578         vpxor x0, y6, y6; \
 579         vpxor x0, y5, y5; \
 580         vpxor x0, y4, y4; \
 581         vpxor x0, y3, y3; \
 582         vpxor x0, y2, y2; \
 583         vpxor x0, y1, y1; \
 584         vpxor x0, y0, y0; \
 585         vpxor x0, x7, x7; \
 586         vpxor x0, x6, x6; \
 587         vpxor x0, x5, x5; \
 588         vpxor x0, x4, x4; \
 589         vpxor x0, x3, x3; \
 590         vpxor x0, x2, x2; \
 591         vpxor x0, x1, x1; \
 592         vpxor stack_tmp0, x0, x0;
 593
 594 #define write_output(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
 595                      y6, y7, rio) \
 596         vmovdqu x0, 0 * 32(rio); \
 597         vmovdqu x1, 1 * 32(rio); \
 598         vmovdqu x2, 2 * 32(rio); \
 599         vmovdqu x3, 3 * 32(rio); \
 600         vmovdqu x4, 4 * 32(rio); \
 601         vmovdqu x5, 5 * 32(rio); \
 602         vmovdqu x6, 6 * 32(rio); \
 603         vmovdqu x7, 7 * 32(rio); \
 604         vmovdqu y0, 8 * 32(rio); \
 605         vmovdqu y1, 9 * 32(rio); \
 606         vmovdqu y2, 10 * 32(rio); \
 607         vmovdqu y3, 11 * 32(rio); \
 608         vmovdqu y4, 12 * 32(rio); \
 609         vmovdqu y5, 13 * 32(rio); \
 610         vmovdqu y6, 14 * 32(rio); \
 611         vmovdqu y7, 15 * 32(rio);
 612
 613
 614 .section        .rodata.cst32.shufb_16x16b, "aM", @progbits, 32
 615 .align 32
 616 #define SHUFB_BYTES(idx) \
 617         0 + (idx), 4 + (idx), 8 + (idx), 12 + (idx)
 618 .Lshufb_16x16b:
 619         .byte SHUFB_BYTES(0), SHUFB_BYTES(1), SHUFB_BYTES(2), SHUFB_BYTES(3)
 620         .byte SHUFB_BYTES(0), SHUFB_BYTES(1), SHUFB_BYTES(2), SHUFB_BYTES(3)
 621
 622 .section        .rodata.cst32.pack_bswap, "aM", @progbits, 32
 623 .align 32
 624 .Lpack_bswap:
 625         .long 0x00010203, 0x04050607, 0x80808080, 0x80808080
 626         .long 0x00010203, 0x04050607, 0x80808080, 0x80808080
 627
 628 /* NB: section is mergeable, all elements must be aligned 16-byte blocks */
 629 .section        .rodata.cst16, "aM", @progbits, 16
 630 .align 16
 631
 632 /* For CTR-mode IV byteswap */
 633 .Lbswap128_mask:
 634         .byte 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0
 635
 636 /* For XTS mode */
 637 .Lxts_gf128mul_and_shl1_mask_0:
 638         .byte 0x87, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0
 639 .Lxts_gf128mul_and_shl1_mask_1:
 640         .byte 0x0e, 1, 0, 0, 0, 0, 0, 0, 2, 0, 0, 0, 0, 0, 0, 0
 641
 642 /*
 643  * pre-SubByte transform
 644  *
 645  * pre-lookup for sbox1, sbox2, sbox3:
 646  *   swap_bitendianness(
 647  *       isom_map_camellia_to_aes(
 648  *           camellia_f(
 649  *               swap_bitendianess(in)
 650  *           )
 651  *       )
 652  *   )
 653  *
 654  * (note: '⊕ 0xc5' inside camellia_f())
 655  */
 656 .Lpre_tf_lo_s1:
 657         .byte 0x45, 0xe8, 0x40, 0xed, 0x2e, 0x83, 0x2b, 0x86
 658         .byte 0x4b, 0xe6, 0x4e, 0xe3, 0x20, 0x8d, 0x25, 0x88
 659 .Lpre_tf_hi_s1:
 660         .byte 0x00, 0x51, 0xf1, 0xa0, 0x8a, 0xdb, 0x7b, 0x2a
 661         .byte 0x09, 0x58, 0xf8, 0xa9, 0x83, 0xd2, 0x72, 0x23
 662
 663 /*
 664  * pre-SubByte transform
 665  *
 666  * pre-lookup for sbox4:
 667  *   swap_bitendianness(
 668  *       isom_map_camellia_to_aes(
 669  *           camellia_f(
 670  *               swap_bitendianess(in <<< 1)
 671  *           )
 672  *       )
 673  *   )
 674  *
 675  * (note: '⊕ 0xc5' inside camellia_f())
 676  */
 677 .Lpre_tf_lo_s4:
 678         .byte 0x45, 0x40, 0x2e, 0x2b, 0x4b, 0x4e, 0x20, 0x25
 679         .byte 0x14, 0x11, 0x7f, 0x7a, 0x1a, 0x1f, 0x71, 0x74
 680 .Lpre_tf_hi_s4:
 681         .byte 0x00, 0xf1, 0x8a, 0x7b, 0x09, 0xf8, 0x83, 0x72
 682         .byte 0xad, 0x5c, 0x27, 0xd6, 0xa4, 0x55, 0x2e, 0xdf
 683
 684 /*
 685  * post-SubByte transform
 686  *
 687  * post-lookup for sbox1, sbox4:
 688  *  swap_bitendianness(
 689  *      camellia_h(
 690  *          isom_map_aes_to_camellia(
 691  *              swap_bitendianness(
 692  *                  aes_inverse_affine_transform(in)
 693  *              )
 694  *          )
 695  *      )
 696  *  )
 697  *
 698  * (note: '⊕ 0x6e' inside camellia_h())
 699  */
 700 .Lpost_tf_lo_s1:
 701         .byte 0x3c, 0xcc, 0xcf, 0x3f, 0x32, 0xc2, 0xc1, 0x31
 702         .byte 0xdc, 0x2c, 0x2f, 0xdf, 0xd2, 0x22, 0x21, 0xd1
 703 .Lpost_tf_hi_s1:
 704         .byte 0x00, 0xf9, 0x86, 0x7f, 0xd7, 0x2e, 0x51, 0xa8
 705         .byte 0xa4, 0x5d, 0x22, 0xdb, 0x73, 0x8a, 0xf5, 0x0c
 706
 707 /*
 708  * post-SubByte transform
 709  *
 710  * post-lookup for sbox2:
 711  *  swap_bitendianness(
 712  *      camellia_h(
 713  *          isom_map_aes_to_camellia(
 714  *              swap_bitendianness(
 715  *                  aes_inverse_affine_transform(in)
 716  *              )
 717  *          )
 718  *      )
 719  *  ) <<< 1
 720  *
 721  * (note: '⊕ 0x6e' inside camellia_h())
 722  */
 723 .Lpost_tf_lo_s2:
 724         .byte 0x78, 0x99, 0x9f, 0x7e, 0x64, 0x85, 0x83, 0x62
 725         .byte 0xb9, 0x58, 0x5e, 0xbf, 0xa5, 0x44, 0x42, 0xa3
 726 .Lpost_tf_hi_s2:
 727         .byte 0x00, 0xf3, 0x0d, 0xfe, 0xaf, 0x5c, 0xa2, 0x51
 728         .byte 0x49, 0xba, 0x44, 0xb7, 0xe6, 0x15, 0xeb, 0x18
 729
 730 /*
 731  * post-SubByte transform
 732  *
 733  * post-lookup for sbox3:
 734  *  swap_bitendianness(
 735  *      camellia_h(
 736  *          isom_map_aes_to_camellia(
 737  *              swap_bitendianness(
 738  *                  aes_inverse_affine_transform(in)
 739  *              )
 740  *          )
 741  *      )
 742  *  ) >>> 1
 743  *
 744  * (note: '⊕ 0x6e' inside camellia_h())
 745  */
 746 .Lpost_tf_lo_s3:
 747         .byte 0x1e, 0x66, 0xe7, 0x9f, 0x19, 0x61, 0xe0, 0x98
 748         .byte 0x6e, 0x16, 0x97, 0xef, 0x69, 0x11, 0x90, 0xe8
 749 .Lpost_tf_hi_s3:
 750         .byte 0x00, 0xfc, 0x43, 0xbf, 0xeb, 0x17, 0xa8, 0x54
 751         .byte 0x52, 0xae, 0x11, 0xed, 0xb9, 0x45, 0xfa, 0x06
 752
 753 /* For isolating SubBytes from AESENCLAST, inverse shift row */
 754 .Linv_shift_row:
 755         .byte 0x00, 0x0d, 0x0a, 0x07, 0x04, 0x01, 0x0e, 0x0b
 756         .byte 0x08, 0x05, 0x02, 0x0f, 0x0c, 0x09, 0x06, 0x03
 757
 758 .section        .rodata.cst4.L0f0f0f0f, "aM", @progbits, 4
 759 .align 4
 760 /* 4-bit mask */
 761 .L0f0f0f0f:
 762         .long 0x0f0f0f0f
 763
 764 .text
 765
 766 .align 8
 767 __camellia_enc_blk32:
 768         /* input:
 769          *      %rdi: ctx, CTX
 770          *      %rax: temporary storage, 512 bytes
 771          *      %ymm0..%ymm15: 32 plaintext blocks
 772          * output:
 773          *      %ymm0..%ymm15: 32 encrypted blocks, order swapped:
 774          *       7, 8, 6, 5, 4, 3, 2, 1, 0, 15, 14, 13, 12, 11, 10, 9, 8
 775          */
 776         FRAME_BEGIN
 777
 778         leaq 8 * 32(%rax), %rcx;
 779
 780         inpack32_post(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
 781                       %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
 782                       %ymm15, %rax, %rcx);
 783
 784         enc_rounds32(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
 785                      %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
 786                      %ymm15, %rax, %rcx, 0);
 787
 788         fls32(%rax, %ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
 789               %rcx, %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
 790               %ymm15,
 791               ((key_table + (8) * 8) + 0)(CTX),
 792               ((key_table + (8) * 8) + 4)(CTX),
 793               ((key_table + (8) * 8) + 8)(CTX),
 794               ((key_table + (8) * 8) + 12)(CTX));
 795
 796         enc_rounds32(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
 797                      %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
 798                      %ymm15, %rax, %rcx, 8);
 799
 800         fls32(%rax, %ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
 801               %rcx, %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
 802               %ymm15,
 803               ((key_table + (16) * 8) + 0)(CTX),
 804               ((key_table + (16) * 8) + 4)(CTX),
 805               ((key_table + (16) * 8) + 8)(CTX),
 806               ((key_table + (16) * 8) + 12)(CTX));
 807
 808         enc_rounds32(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
 809                      %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
 810                      %ymm15, %rax, %rcx, 16);
 811
 812         movl $24, %r8d;
 813         cmpl $16, key_length(CTX);
 814         jne .Lenc_max32;
 815
 816 .Lenc_done:
 817         /* load CD for output */
 818         vmovdqu 0 * 32(%rcx), %ymm8;
 819         vmovdqu 1 * 32(%rcx), %ymm9;
 820         vmovdqu 2 * 32(%rcx), %ymm10;
 821         vmovdqu 3 * 32(%rcx), %ymm11;
 822         vmovdqu 4 * 32(%rcx), %ymm12;
 823         vmovdqu 5 * 32(%rcx), %ymm13;
 824         vmovdqu 6 * 32(%rcx), %ymm14;
 825         vmovdqu 7 * 32(%rcx), %ymm15;
 826
 827         outunpack32(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
 828                     %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
 829                     %ymm15, (key_table)(CTX, %r8, 8), (%rax), 1 * 32(%rax));
 830
 831         FRAME_END
 832         ret;
 833
 834 .align 8
 835 .Lenc_max32:
 836         movl $32, %r8d;
 837
 838         fls32(%rax, %ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
 839               %rcx, %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
 840               %ymm15,
 841               ((key_table + (24) * 8) + 0)(CTX),
 842               ((key_table + (24) * 8) + 4)(CTX),
 843               ((key_table + (24) * 8) + 8)(CTX),
 844               ((key_table + (24) * 8) + 12)(CTX));
 845
 846         enc_rounds32(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
 847                      %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
 848                      %ymm15, %rax, %rcx, 24);
 849
 850         jmp .Lenc_done;
 851 ENDPROC(__camellia_enc_blk32)
 852
 853 .align 8
 854 __camellia_dec_blk32:
 855         /* input:
 856          *      %rdi: ctx, CTX
 857          *      %rax: temporary storage, 512 bytes
 858          *      %r8d: 24 for 16 byte key, 32 for larger
 859          *      %ymm0..%ymm15: 16 encrypted blocks
 860          * output:
 861          *      %ymm0..%ymm15: 16 plaintext blocks, order swapped:
 862          *       7, 8, 6, 5, 4, 3, 2, 1, 0, 15, 14, 13, 12, 11, 10, 9, 8
 863          */
 864         FRAME_BEGIN
 865
 866         leaq 8 * 32(%rax), %rcx;
 867
 868         inpack32_post(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
 869                       %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
 870                       %ymm15, %rax, %rcx);
 871
 872         cmpl $32, %r8d;
 873         je .Ldec_max32;
 874
 875 .Ldec_max24:
 876         dec_rounds32(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
 877                      %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
 878                      %ymm15, %rax, %rcx, 16);
 879
 880         fls32(%rax, %ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
 881               %rcx, %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
 882               %ymm15,
 883               ((key_table + (16) * 8) + 8)(CTX),
 884               ((key_table + (16) * 8) + 12)(CTX),
 885               ((key_table + (16) * 8) + 0)(CTX),
 886               ((key_table + (16) * 8) + 4)(CTX));
 887
 888         dec_rounds32(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
 889                      %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
 890                      %ymm15, %rax, %rcx, 8);
 891
 892         fls32(%rax, %ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
 893               %rcx, %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
 894               %ymm15,
 895               ((key_table + (8) * 8) + 8)(CTX),
 896               ((key_table + (8) * 8) + 12)(CTX),
 897               ((key_table + (8) * 8) + 0)(CTX),
 898               ((key_table + (8) * 8) + 4)(CTX));
 899
 900         dec_rounds32(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
 901                      %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
 902                      %ymm15, %rax, %rcx, 0);
 903
 904         /* load CD for output */
 905         vmovdqu 0 * 32(%rcx), %ymm8;
 906         vmovdqu 1 * 32(%rcx), %ymm9;
 907         vmovdqu 2 * 32(%rcx), %ymm10;
 908         vmovdqu 3 * 32(%rcx), %ymm11;
 909         vmovdqu 4 * 32(%rcx), %ymm12;
 910         vmovdqu 5 * 32(%rcx), %ymm13;
 911         vmovdqu 6 * 32(%rcx), %ymm14;
 912         vmovdqu 7 * 32(%rcx), %ymm15;
 913
 914         outunpack32(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
 915                     %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
 916                     %ymm15, (key_table)(CTX), (%rax), 1 * 32(%rax));
 917
 918         FRAME_END
 919         ret;
 920
 921 .align 8
 922 .Ldec_max32:
 923         dec_rounds32(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
 924                      %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
 925                      %ymm15, %rax, %rcx, 24);
 926
 927         fls32(%rax, %ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
 928               %rcx, %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
 929               %ymm15,
 930               ((key_table + (24) * 8) + 8)(CTX),
 931               ((key_table + (24) * 8) + 12)(CTX),
 932               ((key_table + (24) * 8) + 0)(CTX),
 933               ((key_table + (24) * 8) + 4)(CTX));
 934
 935         jmp .Ldec_max24;
 936 ENDPROC(__camellia_dec_blk32)
 937
 938 ENTRY(camellia_ecb_enc_32way)
 939         /* input:
 940          *      %rdi: ctx, CTX
 941          *      %rsi: dst (32 blocks)
 942          *      %rdx: src (32 blocks)
 943          */
 944         FRAME_BEGIN
 945
 946         vzeroupper;
 947
 948         inpack32_pre(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
 949                      %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
 950                      %ymm15, %rdx, (key_table)(CTX));
 951
 952         /* now dst can be used as temporary buffer (even in src == dst case) */
 953         movq    %rsi, %rax;
 954
 955         call __camellia_enc_blk32;
 956
 957         write_output(%ymm7, %ymm6, %ymm5, %ymm4, %ymm3, %ymm2, %ymm1, %ymm0,
 958                      %ymm15, %ymm14, %ymm13, %ymm12, %ymm11, %ymm10, %ymm9,
 959                      %ymm8, %rsi);
 960
 961         vzeroupper;
 962
 963         FRAME_END
 964         ret;
 965 ENDPROC(camellia_ecb_enc_32way)
 966
 967 ENTRY(camellia_ecb_dec_32way)
 968         /* input:
 969          *      %rdi: ctx, CTX
 970          *      %rsi: dst (32 blocks)
 971          *      %rdx: src (32 blocks)
 972          */
 973         FRAME_BEGIN
 974
 975         vzeroupper;
 976
 977         cmpl $16, key_length(CTX);
 978         movl $32, %r8d;
 979         movl $24, %eax;
 980         cmovel %eax, %r8d; /* max */
 981
 982         inpack32_pre(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
 983                      %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
 984                      %ymm15, %rdx, (key_table)(CTX, %r8, 8));
 985
 986         /* now dst can be used as temporary buffer (even in src == dst case) */
 987         movq    %rsi, %rax;
 988
 989         call __camellia_dec_blk32;
 990
 991         write_output(%ymm7, %ymm6, %ymm5, %ymm4, %ymm3, %ymm2, %ymm1, %ymm0,
 992                      %ymm15, %ymm14, %ymm13, %ymm12, %ymm11, %ymm10, %ymm9,
 993                      %ymm8, %rsi);
 994
 995         vzeroupper;
 996
 997         FRAME_END
 998         ret;
 999 ENDPROC(camellia_ecb_dec_32way)
1000
1001 ENTRY(camellia_cbc_dec_32way)
1002         /* input:
1003          *      %rdi: ctx, CTX
1004          *      %rsi: dst (32 blocks)
1005          *      %rdx: src (32 blocks)
1006          */
1007         FRAME_BEGIN
1008
1009         vzeroupper;
1010
1011         cmpl $16, key_length(CTX);
1012         movl $32, %r8d;
1013         movl $24, %eax;
1014         cmovel %eax, %r8d; /* max */
1015
1016         inpack32_pre(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
1017                      %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
1018                      %ymm15, %rdx, (key_table)(CTX, %r8, 8));
1019
1020         movq %rsp, %r10;
1021         cmpq %rsi, %rdx;
1022         je .Lcbc_dec_use_stack;
1023
1024         /* dst can be used as temporary storage, src is not overwritten. */
1025         movq %rsi, %rax;
1026         jmp .Lcbc_dec_continue;
1027
1028 .Lcbc_dec_use_stack:
1029         /*
1030          * dst still in-use (because dst == src), so use stack for temporary
1031          * storage.
1032          */
1033         subq $(16 * 32), %rsp;
1034         movq %rsp, %rax;
1035
1036 .Lcbc_dec_continue:
1037         call __camellia_dec_blk32;
1038
1039         vmovdqu %ymm7, (%rax);
1040         vpxor %ymm7, %ymm7, %ymm7;
1041         vinserti128 $1, (%rdx), %ymm7, %ymm7;
1042         vpxor (%rax), %ymm7, %ymm7;
1043         movq %r10, %rsp;
1044         vpxor (0 * 32 + 16)(%rdx), %ymm6, %ymm6;
1045         vpxor (1 * 32 + 16)(%rdx), %ymm5, %ymm5;
1046         vpxor (2 * 32 + 16)(%rdx), %ymm4, %ymm4;
1047         vpxor (3 * 32 + 16)(%rdx), %ymm3, %ymm3;
1048         vpxor (4 * 32 + 16)(%rdx), %ymm2, %ymm2;
1049         vpxor (5 * 32 + 16)(%rdx), %ymm1, %ymm1;
1050         vpxor (6 * 32 + 16)(%rdx), %ymm0, %ymm0;
1051         vpxor (7 * 32 + 16)(%rdx), %ymm15, %ymm15;
1052         vpxor (8 * 32 + 16)(%rdx), %ymm14, %ymm14;
1053         vpxor (9 * 32 + 16)(%rdx), %ymm13, %ymm13;
1054         vpxor (10 * 32 + 16)(%rdx), %ymm12, %ymm12;
1055         vpxor (11 * 32 + 16)(%rdx), %ymm11, %ymm11;
1056         vpxor (12 * 32 + 16)(%rdx), %ymm10, %ymm10;
1057         vpxor (13 * 32 + 16)(%rdx), %ymm9, %ymm9;
1058         vpxor (14 * 32 + 16)(%rdx), %ymm8, %ymm8;
1059         write_output(%ymm7, %ymm6, %ymm5, %ymm4, %ymm3, %ymm2, %ymm1, %ymm0,
1060                      %ymm15, %ymm14, %ymm13, %ymm12, %ymm11, %ymm10, %ymm9,
1061                      %ymm8, %rsi);
1062
1063         vzeroupper;
1064
1065         FRAME_END
1066         ret;
1067 ENDPROC(camellia_cbc_dec_32way)
1068
1069 #define inc_le128(x, minus_one, tmp) \
1070         vpcmpeqq minus_one, x, tmp; \
1071         vpsubq minus_one, x, x; \
1072         vpslldq $8, tmp, tmp; \
1073         vpsubq tmp, x, x;
1074
1075 #define add2_le128(x, minus_one, minus_two, tmp1, tmp2) \
1076         vpcmpeqq minus_one, x, tmp1; \
1077         vpcmpeqq minus_two, x, tmp2; \
1078         vpsubq minus_two, x, x; \
1079         vpor tmp2, tmp1, tmp1; \
1080         vpslldq $8, tmp1, tmp1; \
1081         vpsubq tmp1, x, x;
1082
1083 ENTRY(camellia_ctr_32way)
1084         /* input:
1085          *      %rdi: ctx, CTX
1086          *      %rsi: dst (32 blocks)
1087          *      %rdx: src (32 blocks)
1088          *      %rcx: iv (little endian, 128bit)
1089          */
1090         FRAME_BEGIN
1091
1092         vzeroupper;
1093
1094         movq %rsp, %r10;
1095         cmpq %rsi, %rdx;
1096         je .Lctr_use_stack;
1097
1098         /* dst can be used as temporary storage, src is not overwritten. */
1099         movq %rsi, %rax;
1100         jmp .Lctr_continue;
1101
1102 .Lctr_use_stack:
1103         subq $(16 * 32), %rsp;
1104         movq %rsp, %rax;
1105
1106 .Lctr_continue:
1107         vpcmpeqd %ymm15, %ymm15, %ymm15;
1108         vpsrldq $8, %ymm15, %ymm15; /* ab: -1:0 ; cd: -1:0 */
1109         vpaddq %ymm15, %ymm15, %ymm12; /* ab: -2:0 ; cd: -2:0 */
1110
1111         /* load IV and byteswap */
1112         vmovdqu (%rcx), %xmm0;
1113         vmovdqa %xmm0, %xmm1;
1114         inc_le128(%xmm0, %xmm15, %xmm14);
1115         vbroadcasti128 .Lbswap128_mask, %ymm14;
1116         vinserti128 $1, %xmm0, %ymm1, %ymm0;
1117         vpshufb %ymm14, %ymm0, %ymm13;
1118         vmovdqu %ymm13, 15 * 32(%rax);
1119
1120         /* construct IVs */
1121         add2_le128(%ymm0, %ymm15, %ymm12, %ymm11, %ymm13); /* ab:le2 ; cd:le3 */
1122         vpshufb %ymm14, %ymm0, %ymm13;
1123         vmovdqu %ymm13, 14 * 32(%rax);
1124         add2_le128(%ymm0, %ymm15, %ymm12, %ymm11, %ymm13);
1125         vpshufb %ymm14, %ymm0, %ymm13;
1126         vmovdqu %ymm13, 13 * 32(%rax);
1127         add2_le128(%ymm0, %ymm15, %ymm12, %ymm11, %ymm13);
1128         vpshufb %ymm14, %ymm0, %ymm13;
1129         vmovdqu %ymm13, 12 * 32(%rax);
1130         add2_le128(%ymm0, %ymm15, %ymm12, %ymm11, %ymm13);
1131         vpshufb %ymm14, %ymm0, %ymm13;
1132         vmovdqu %ymm13, 11 * 32(%rax);
1133         add2_le128(%ymm0, %ymm15, %ymm12, %ymm11, %ymm13);
1134         vpshufb %ymm14, %ymm0, %ymm10;
1135         add2_le128(%ymm0, %ymm15, %ymm12, %ymm11, %ymm13);
1136         vpshufb %ymm14, %ymm0, %ymm9;
1137         add2_le128(%ymm0, %ymm15, %ymm12, %ymm11, %ymm13);
1138         vpshufb %ymm14, %ymm0, %ymm8;
1139         add2_le128(%ymm0, %ymm15, %ymm12, %ymm11, %ymm13);
1140         vpshufb %ymm14, %ymm0, %ymm7;
1141         add2_le128(%ymm0, %ymm15, %ymm12, %ymm11, %ymm13);
1142         vpshufb %ymm14, %ymm0, %ymm6;
1143         add2_le128(%ymm0, %ymm15, %ymm12, %ymm11, %ymm13);
1144         vpshufb %ymm14, %ymm0, %ymm5;
1145         add2_le128(%ymm0, %ymm15, %ymm12, %ymm11, %ymm13);
1146         vpshufb %ymm14, %ymm0, %ymm4;
1147         add2_le128(%ymm0, %ymm15, %ymm12, %ymm11, %ymm13);
1148         vpshufb %ymm14, %ymm0, %ymm3;
1149         add2_le128(%ymm0, %ymm15, %ymm12, %ymm11, %ymm13);
1150         vpshufb %ymm14, %ymm0, %ymm2;
1151         add2_le128(%ymm0, %ymm15, %ymm12, %ymm11, %ymm13);
1152         vpshufb %ymm14, %ymm0, %ymm1;
1153         add2_le128(%ymm0, %ymm15, %ymm12, %ymm11, %ymm13);
1154         vextracti128 $1, %ymm0, %xmm13;
1155         vpshufb %ymm14, %ymm0, %ymm0;
1156         inc_le128(%xmm13, %xmm15, %xmm14);
1157         vmovdqu %xmm13, (%rcx);
1158
1159         /* inpack32_pre: */
1160         vpbroadcastq (key_table)(CTX), %ymm15;
1161         vpshufb .Lpack_bswap, %ymm15, %ymm15;
1162         vpxor %ymm0, %ymm15, %ymm0;
1163         vpxor %ymm1, %ymm15, %ymm1;
1164         vpxor %ymm2, %ymm15, %ymm2;
1165         vpxor %ymm3, %ymm15, %ymm3;
1166         vpxor %ymm4, %ymm15, %ymm4;
1167         vpxor %ymm5, %ymm15, %ymm5;
1168         vpxor %ymm6, %ymm15, %ymm6;
1169         vpxor %ymm7, %ymm15, %ymm7;
1170         vpxor %ymm8, %ymm15, %ymm8;
1171         vpxor %ymm9, %ymm15, %ymm9;
1172         vpxor %ymm10, %ymm15, %ymm10;
1173         vpxor 11 * 32(%rax), %ymm15, %ymm11;
1174         vpxor 12 * 32(%rax), %ymm15, %ymm12;
1175         vpxor 13 * 32(%rax), %ymm15, %ymm13;
1176         vpxor 14 * 32(%rax), %ymm15, %ymm14;
1177         vpxor 15 * 32(%rax), %ymm15, %ymm15;
1178
1179         call __camellia_enc_blk32;
1180
1181         movq %r10, %rsp;
1182
1183         vpxor 0 * 32(%rdx), %ymm7, %ymm7;
1184         vpxor 1 * 32(%rdx), %ymm6, %ymm6;
1185         vpxor 2 * 32(%rdx), %ymm5, %ymm5;
1186         vpxor 3 * 32(%rdx), %ymm4, %ymm4;
1187         vpxor 4 * 32(%rdx), %ymm3, %ymm3;
1188         vpxor 5 * 32(%rdx), %ymm2, %ymm2;
1189         vpxor 6 * 32(%rdx), %ymm1, %ymm1;
1190         vpxor 7 * 32(%rdx), %ymm0, %ymm0;
1191         vpxor 8 * 32(%rdx), %ymm15, %ymm15;
1192         vpxor 9 * 32(%rdx), %ymm14, %ymm14;
1193         vpxor 10 * 32(%rdx), %ymm13, %ymm13;
1194         vpxor 11 * 32(%rdx), %ymm12, %ymm12;
1195         vpxor 12 * 32(%rdx), %ymm11, %ymm11;
1196         vpxor 13 * 32(%rdx), %ymm10, %ymm10;
1197         vpxor 14 * 32(%rdx), %ymm9, %ymm9;
1198         vpxor 15 * 32(%rdx), %ymm8, %ymm8;
1199         write_output(%ymm7, %ymm6, %ymm5, %ymm4, %ymm3, %ymm2, %ymm1, %ymm0,
1200                      %ymm15, %ymm14, %ymm13, %ymm12, %ymm11, %ymm10, %ymm9,
1201                      %ymm8, %rsi);
1202
1203         vzeroupper;
1204
1205         FRAME_END
1206         ret;
1207 ENDPROC(camellia_ctr_32way)
1208
1209 #define gf128mul_x_ble(iv, mask, tmp) \
1210         vpsrad $31, iv, tmp; \
1211         vpaddq iv, iv, iv; \
1212         vpshufd $0x13, tmp, tmp; \
1213         vpand mask, tmp, tmp; \
1214         vpxor tmp, iv, iv;
1215
1216 #define gf128mul_x2_ble(iv, mask1, mask2, tmp0, tmp1) \
1217         vpsrad $31, iv, tmp0; \
1218         vpaddq iv, iv, tmp1; \
1219         vpsllq $2, iv, iv; \
1220         vpshufd $0x13, tmp0, tmp0; \
1221         vpsrad $31, tmp1, tmp1; \
1222         vpand mask2, tmp0, tmp0; \
1223         vpshufd $0x13, tmp1, tmp1; \
1224         vpxor tmp0, iv, iv; \
1225         vpand mask1, tmp1, tmp1; \
1226         vpxor tmp1, iv, iv;
1227
1228 .align 8
1229 camellia_xts_crypt_32way:
1230         /* input:
1231          *      %rdi: ctx, CTX
1232          *      %rsi: dst (32 blocks)
1233          *      %rdx: src (32 blocks)
1234          *      %rcx: iv (t ⊕ αⁿ ∈ GF(2¹²⁸))
1235          *      %r8: index for input whitening key
1236          *      %r9: pointer to  __camellia_enc_blk32 or __camellia_dec_blk32
1237          */
1238         FRAME_BEGIN
1239
1240         vzeroupper;
1241
1242         subq $(16 * 32), %rsp;
1243         movq %rsp, %rax;
1244
1245         vbroadcasti128 .Lxts_gf128mul_and_shl1_mask_0, %ymm12;
1246
1247         /* load IV and construct second IV */
1248         vmovdqu (%rcx), %xmm0;
1249         vmovdqa %xmm0, %xmm15;
1250         gf128mul_x_ble(%xmm0, %xmm12, %xmm13);
1251         vbroadcasti128 .Lxts_gf128mul_and_shl1_mask_1, %ymm13;
1252         vinserti128 $1, %xmm0, %ymm15, %ymm0;
1253         vpxor 0 * 32(%rdx), %ymm0, %ymm15;
1254         vmovdqu %ymm15, 15 * 32(%rax);
1255         vmovdqu %ymm0, 0 * 32(%rsi);
1256
1257         /* construct IVs */
1258         gf128mul_x2_ble(%ymm0, %ymm12, %ymm13, %ymm14, %ymm15);
1259         vpxor 1 * 32(%rdx), %ymm0, %ymm15;
1260         vmovdqu %ymm15, 14 * 32(%rax);
1261         vmovdqu %ymm0, 1 * 32(%rsi);
1262
1263         gf128mul_x2_ble(%ymm0, %ymm12, %ymm13, %ymm14, %ymm15);
1264         vpxor 2 * 32(%rdx), %ymm0, %ymm15;
1265         vmovdqu %ymm15, 13 * 32(%rax);
1266         vmovdqu %ymm0, 2 * 32(%rsi);
1267
1268         gf128mul_x2_ble(%ymm0, %ymm12, %ymm13, %ymm14, %ymm15);
1269         vpxor 3 * 32(%rdx), %ymm0, %ymm15;
1270         vmovdqu %ymm15, 12 * 32(%rax);
1271         vmovdqu %ymm0, 3 * 32(%rsi);
1272
1273         gf128mul_x2_ble(%ymm0, %ymm12, %ymm13, %ymm14, %ymm15);
1274         vpxor 4 * 32(%rdx), %ymm0, %ymm11;
1275         vmovdqu %ymm0, 4 * 32(%rsi);
1276
1277         gf128mul_x2_ble(%ymm0, %ymm12, %ymm13, %ymm14, %ymm15);
1278         vpxor 5 * 32(%rdx), %ymm0, %ymm10;
1279         vmovdqu %ymm0, 5 * 32(%rsi);
1280
1281         gf128mul_x2_ble(%ymm0, %ymm12, %ymm13, %ymm14, %ymm15);
1282         vpxor 6 * 32(%rdx), %ymm0, %ymm9;
1283         vmovdqu %ymm0, 6 * 32(%rsi);
1284
1285         gf128mul_x2_ble(%ymm0, %ymm12, %ymm13, %ymm14, %ymm15);
1286         vpxor 7 * 32(%rdx), %ymm0, %ymm8;
1287         vmovdqu %ymm0, 7 * 32(%rsi);
1288
1289         gf128mul_x2_ble(%ymm0, %ymm12, %ymm13, %ymm14, %ymm15);
1290         vpxor 8 * 32(%rdx), %ymm0, %ymm7;
1291         vmovdqu %ymm0, 8 * 32(%rsi);
1292
1293         gf128mul_x2_ble(%ymm0, %ymm12, %ymm13, %ymm14, %ymm15);
1294         vpxor 9 * 32(%rdx), %ymm0, %ymm6;
1295         vmovdqu %ymm0, 9 * 32(%rsi);
1296
1297         gf128mul_x2_ble(%ymm0, %ymm12, %ymm13, %ymm14, %ymm15);
1298         vpxor 10 * 32(%rdx), %ymm0, %ymm5;
1299         vmovdqu %ymm0, 10 * 32(%rsi);
1300
1301         gf128mul_x2_ble(%ymm0, %ymm12, %ymm13, %ymm14, %ymm15);
1302         vpxor 11 * 32(%rdx), %ymm0, %ymm4;
1303         vmovdqu %ymm0, 11 * 32(%rsi);
1304
1305         gf128mul_x2_ble(%ymm0, %ymm12, %ymm13, %ymm14, %ymm15);
1306         vpxor 12 * 32(%rdx), %ymm0, %ymm3;
1307         vmovdqu %ymm0, 12 * 32(%rsi);
1308
1309         gf128mul_x2_ble(%ymm0, %ymm12, %ymm13, %ymm14, %ymm15);
1310         vpxor 13 * 32(%rdx), %ymm0, %ymm2;
1311         vmovdqu %ymm0, 13 * 32(%rsi);
1312
1313         gf128mul_x2_ble(%ymm0, %ymm12, %ymm13, %ymm14, %ymm15);
1314         vpxor 14 * 32(%rdx), %ymm0, %ymm1;
1315         vmovdqu %ymm0, 14 * 32(%rsi);
1316
1317         gf128mul_x2_ble(%ymm0, %ymm12, %ymm13, %ymm14, %ymm15);
1318         vpxor 15 * 32(%rdx), %ymm0, %ymm15;
1319         vmovdqu %ymm15, 0 * 32(%rax);
1320         vmovdqu %ymm0, 15 * 32(%rsi);
1321
1322         vextracti128 $1, %ymm0, %xmm0;
1323         gf128mul_x_ble(%xmm0, %xmm12, %xmm15);
1324         vmovdqu %xmm0, (%rcx);
1325
1326         /* inpack32_pre: */
1327         vpbroadcastq (key_table)(CTX, %r8, 8), %ymm15;
1328         vpshufb .Lpack_bswap, %ymm15, %ymm15;
1329         vpxor 0 * 32(%rax), %ymm15, %ymm0;
1330         vpxor %ymm1, %ymm15, %ymm1;
1331         vpxor %ymm2, %ymm15, %ymm2;
1332         vpxor %ymm3, %ymm15, %ymm3;
1333         vpxor %ymm4, %ymm15, %ymm4;
1334         vpxor %ymm5, %ymm15, %ymm5;
1335         vpxor %ymm6, %ymm15, %ymm6;
1336         vpxor %ymm7, %ymm15, %ymm7;
1337         vpxor %ymm8, %ymm15, %ymm8;
1338         vpxor %ymm9, %ymm15, %ymm9;
1339         vpxor %ymm10, %ymm15, %ymm10;
1340         vpxor %ymm11, %ymm15, %ymm11;
1341         vpxor 12 * 32(%rax), %ymm15, %ymm12;
1342         vpxor 13 * 32(%rax), %ymm15, %ymm13;
1343         vpxor 14 * 32(%rax), %ymm15, %ymm14;
1344         vpxor 15 * 32(%rax), %ymm15, %ymm15;
1345
1346         call *%r9;
1347
1348         addq $(16 * 32), %rsp;
1349
1350         vpxor 0 * 32(%rsi), %ymm7, %ymm7;
1351         vpxor 1 * 32(%rsi), %ymm6, %ymm6;
1352         vpxor 2 * 32(%rsi), %ymm5, %ymm5;
1353         vpxor 3 * 32(%rsi), %ymm4, %ymm4;
1354         vpxor 4 * 32(%rsi), %ymm3, %ymm3;
1355         vpxor 5 * 32(%rsi), %ymm2, %ymm2;
1356         vpxor 6 * 32(%rsi), %ymm1, %ymm1;
1357         vpxor 7 * 32(%rsi), %ymm0, %ymm0;
1358         vpxor 8 * 32(%rsi), %ymm15, %ymm15;
1359         vpxor 9 * 32(%rsi), %ymm14, %ymm14;
1360         vpxor 10 * 32(%rsi), %ymm13, %ymm13;
1361         vpxor 11 * 32(%rsi), %ymm12, %ymm12;
1362         vpxor 12 * 32(%rsi), %ymm11, %ymm11;
1363         vpxor 13 * 32(%rsi), %ymm10, %ymm10;
1364         vpxor 14 * 32(%rsi), %ymm9, %ymm9;
1365         vpxor 15 * 32(%rsi), %ymm8, %ymm8;
1366         write_output(%ymm7, %ymm6, %ymm5, %ymm4, %ymm3, %ymm2, %ymm1, %ymm0,
1367                      %ymm15, %ymm14, %ymm13, %ymm12, %ymm11, %ymm10, %ymm9,
1368                      %ymm8, %rsi);
1369
1370         vzeroupper;
1371
1372         FRAME_END
1373         ret;
1374 ENDPROC(camellia_xts_crypt_32way)
1375
1376 ENTRY(camellia_xts_enc_32way)
1377         /* input:
1378          *      %rdi: ctx, CTX
1379          *      %rsi: dst (32 blocks)
1380          *      %rdx: src (32 blocks)
1381          *      %rcx: iv (t ⊕ αⁿ ∈ GF(2¹²⁸))
1382          */
1383
1384         xorl %r8d, %r8d; /* input whitening key, 0 for enc */
1385
1386         leaq __camellia_enc_blk32, %r9;
1387
1388         jmp camellia_xts_crypt_32way;
1389 ENDPROC(camellia_xts_enc_32way)
1390
1391 ENTRY(camellia_xts_dec_32way)
1392         /* input:
1393          *      %rdi: ctx, CTX
1394          *      %rsi: dst (32 blocks)
1395          *      %rdx: src (32 blocks)
1396          *      %rcx: iv (t ⊕ αⁿ ∈ GF(2¹²⁸))
1397          */
1398
1399         cmpl $16, key_length(CTX);
1400         movl $32, %r8d;
1401         movl $24, %eax;
1402         cmovel %eax, %r8d;  /* input whitening key, last for dec */
1403
1404         leaq __camellia_dec_blk32, %r9;
1405
1406         jmp camellia_xts_crypt_32way;
1407 ENDPROC(camellia_xts_dec_32way)