crypto/aes/asm/vpaes-x86_64.pl

   1 #! /usr/bin/env perl
   2 # Copyright 2011-2020 The OpenSSL Project Authors. All Rights Reserved.
   3 #
   4 # Licensed under the Apache License 2.0 (the "License").  You may not use
   5 # this file except in compliance with the License.  You can obtain a copy
   6 # in the file LICENSE in the source distribution or at
   7 # https://www.openssl.org/source/license.html
   8
   9
  10 ######################################################################
  11 ## Constant-time SSSE3 AES core implementation.
  12 ## version 0.1
  13 ##
  14 ## By Mike Hamburg (Stanford University), 2009
  15 ## Public domain.
  16 ##
  17 ## For details see http://shiftleft.org/papers/vector_aes/ and
  18 ## http://crypto.stanford.edu/vpaes/.
  19
  20 ######################################################################
  21 # September 2011.
  22 #
  23 # Interface to OpenSSL as "almost" drop-in replacement for
  24 # aes-x86_64.pl. "Almost" refers to the fact that AES_cbc_encrypt
  25 # doesn't handle partial vectors (doesn't have to if called from
  26 # EVP only). "Drop-in" implies that this module doesn't share key
  27 # schedule structure with the original nor does it make assumption
  28 # about its alignment...
  29 #
  30 # Performance summary. aes-x86_64.pl column lists large-block CBC
  31 # encrypt/decrypt/with-hyper-threading-off(*) results in cycles per
  32 # byte processed with 128-bit key, and vpaes-x86_64.pl column -
  33 # [also large-block CBC] encrypt/decrypt.
  34 #
  35 #               aes-x86_64.pl           vpaes-x86_64.pl
  36 #
  37 # Core 2(**)    29.6/41.1/14.3          21.9/25.2(***)
  38 # Nehalem       29.6/40.3/14.6          10.0/11.8
  39 # Atom          57.3/74.2/32.1          60.9/77.2(***)
  40 # Silvermont    52.7/64.0/19.5          48.8/60.8(***)
  41 # Goldmont      38.9/49.0/17.8          10.6/12.6
  42 #
  43 # (*)   "Hyper-threading" in the context refers rather to cache shared
  44 #       among multiple cores, than to specifically Intel HTT. As vast
  45 #       majority of contemporary cores share cache, slower code path
  46 #       is common place. In other words "with-hyper-threading-off"
  47 #       results are presented mostly for reference purposes.
  48 #
  49 # (**)  "Core 2" refers to initial 65nm design, a.k.a. Conroe.
  50 #
  51 # (***) Less impressive improvement on Core 2 and Atom is due to slow
  52 #       pshufb, yet it's respectable +36%/62% improvement on Core 2
  53 #       (as implied, over "hyper-threading-safe" code path).
  54 #
  55 #                                               <appro@openssl.org>
  56
  57 # $output is the last argument if it looks like a file (it has an extension)
  58 # $flavour is the first argument if it doesn't look like a file
  59 $output = $#ARGV >= 0 && $ARGV[$#ARGV] =~ m|\.\w+$| ? pop : undef;
  60 $flavour = $#ARGV >= 0 && $ARGV[0] !~ m|\.| ? shift : undef;
  61
  62 $win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);
  63
  64 $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
  65 ( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
  66 ( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or
  67 die "can't locate x86_64-xlate.pl";
  68
  69 open OUT,"| \"$^X\" \"$xlate\" $flavour \"$output\""
  70     or die "can't call $xlate: $!";
  71 *STDOUT=*OUT;
  72
  73 $PREFIX="vpaes";
  74
  75 $code.=<<___;
  76 .text
  77
  78 ##
  79 ##  _aes_encrypt_core
  80 ##
  81 ##  AES-encrypt %xmm0.
  82 ##
  83 ##  Inputs:
  84 ##     %xmm0 = input
  85 ##     %xmm9-%xmm15 as in _vpaes_preheat
  86 ##    (%rdx) = scheduled keys
  87 ##
  88 ##  Output in %xmm0
  89 ##  Clobbers  %xmm1-%xmm5, %r9, %r10, %r11, %rax
  90 ##  Preserves %xmm6 - %xmm8 so you get some local vectors
  91 ##
  92 ##
  93 .type   _vpaes_encrypt_core,\@abi-omnipotent
  94 .align 16
  95 _vpaes_encrypt_core:
  96 .cfi_startproc
  97         mov     %rdx,   %r9
  98         mov     \$16,   %r11
  99         mov     240(%rdx),%eax
 100         movdqa  %xmm9,  %xmm1
 101         movdqa  .Lk_ipt(%rip), %xmm2    # iptlo
 102         pandn   %xmm0,  %xmm1
 103         movdqu  (%r9),  %xmm5           # round0 key
 104         psrld   \$4,    %xmm1
 105         pand    %xmm9,  %xmm0
 106         pshufb  %xmm0,  %xmm2
 107         movdqa  .Lk_ipt+16(%rip), %xmm0 # ipthi
 108         pshufb  %xmm1,  %xmm0
 109         pxor    %xmm5,  %xmm2
 110         add     \$16,   %r9
 111         pxor    %xmm2,  %xmm0
 112         lea     .Lk_mc_backward(%rip),%r10
 113         jmp     .Lenc_entry
 114
 115 .align 16
 116 .Lenc_loop:
 117         # middle of middle round
 118         movdqa  %xmm13, %xmm4   # 4 : sb1u
 119         movdqa  %xmm12, %xmm0   # 0 : sb1t
 120         pshufb  %xmm2,  %xmm4   # 4 = sb1u
 121         pshufb  %xmm3,  %xmm0   # 0 = sb1t
 122         pxor    %xmm5,  %xmm4   # 4 = sb1u + k
 123         movdqa  %xmm15, %xmm5   # 4 : sb2u
 124         pxor    %xmm4,  %xmm0   # 0 = A
 125         movdqa  -0x40(%r11,%r10), %xmm1         # .Lk_mc_forward[]
 126         pshufb  %xmm2,  %xmm5   # 4 = sb2u
 127         movdqa  (%r11,%r10), %xmm4              # .Lk_mc_backward[]
 128         movdqa  %xmm14, %xmm2   # 2 : sb2t
 129         pshufb  %xmm3,  %xmm2   # 2 = sb2t
 130         movdqa  %xmm0,  %xmm3   # 3 = A
 131         pxor    %xmm5,  %xmm2   # 2 = 2A
 132         pshufb  %xmm1,  %xmm0   # 0 = B
 133         add     \$16,   %r9     # next key
 134         pxor    %xmm2,  %xmm0   # 0 = 2A+B
 135         pshufb  %xmm4,  %xmm3   # 3 = D
 136         add     \$16,   %r11    # next mc
 137         pxor    %xmm0,  %xmm3   # 3 = 2A+B+D
 138         pshufb  %xmm1,  %xmm0   # 0 = 2B+C
 139         and     \$0x30, %r11    # ... mod 4
 140         sub     \$1,%rax        # nr--
 141         pxor    %xmm3,  %xmm0   # 0 = 2A+3B+C+D
 142
 143 .Lenc_entry:
 144         # top of round
 145         movdqa  %xmm9,  %xmm1   # 1 : i
 146         movdqa  %xmm11, %xmm5   # 2 : a/k
 147         pandn   %xmm0,  %xmm1   # 1 = i<<4
 148         psrld   \$4,    %xmm1   # 1 = i
 149         pand    %xmm9,  %xmm0   # 0 = k
 150         pshufb  %xmm0,  %xmm5   # 2 = a/k
 151         movdqa  %xmm10, %xmm3   # 3 : 1/i
 152         pxor    %xmm1,  %xmm0   # 0 = j
 153         pshufb  %xmm1,  %xmm3   # 3 = 1/i
 154         movdqa  %xmm10, %xmm4   # 4 : 1/j
 155         pxor    %xmm5,  %xmm3   # 3 = iak = 1/i + a/k
 156         pshufb  %xmm0,  %xmm4   # 4 = 1/j
 157         movdqa  %xmm10, %xmm2   # 2 : 1/iak
 158         pxor    %xmm5,  %xmm4   # 4 = jak = 1/j + a/k
 159         pshufb  %xmm3,  %xmm2   # 2 = 1/iak
 160         movdqa  %xmm10, %xmm3   # 3 : 1/jak
 161         pxor    %xmm0,  %xmm2   # 2 = io
 162         pshufb  %xmm4,  %xmm3   # 3 = 1/jak
 163         movdqu  (%r9),  %xmm5
 164         pxor    %xmm1,  %xmm3   # 3 = jo
 165         jnz     .Lenc_loop
 166
 167         # middle of last round
 168         movdqa  -0x60(%r10), %xmm4      # 3 : sbou      .Lk_sbo
 169         movdqa  -0x50(%r10), %xmm0      # 0 : sbot      .Lk_sbo+16
 170         pshufb  %xmm2,  %xmm4   # 4 = sbou
 171         pxor    %xmm5,  %xmm4   # 4 = sb1u + k
 172         pshufb  %xmm3,  %xmm0   # 0 = sb1t
 173         movdqa  0x40(%r11,%r10), %xmm1          # .Lk_sr[]
 174         pxor    %xmm4,  %xmm0   # 0 = A
 175         pshufb  %xmm1,  %xmm0
 176         ret
 177 .cfi_endproc
 178 .size   _vpaes_encrypt_core,.-_vpaes_encrypt_core
 179
 180 ##
 181 ##  Decryption core
 182 ##
 183 ##  Same API as encryption core.
 184 ##
 185 .type   _vpaes_decrypt_core,\@abi-omnipotent
 186 .align  16
 187 _vpaes_decrypt_core:
 188 .cfi_startproc
 189         mov     %rdx,   %r9             # load key
 190         mov     240(%rdx),%eax
 191         movdqa  %xmm9,  %xmm1
 192         movdqa  .Lk_dipt(%rip), %xmm2   # iptlo
 193         pandn   %xmm0,  %xmm1
 194         mov     %rax,   %r11
 195         psrld   \$4,    %xmm1
 196         movdqu  (%r9),  %xmm5           # round0 key
 197         shl     \$4,    %r11
 198         pand    %xmm9,  %xmm0
 199         pshufb  %xmm0,  %xmm2
 200         movdqa  .Lk_dipt+16(%rip), %xmm0 # ipthi
 201         xor     \$0x30, %r11
 202         lea     .Lk_dsbd(%rip),%r10
 203         pshufb  %xmm1,  %xmm0
 204         and     \$0x30, %r11
 205         pxor    %xmm5,  %xmm2
 206         movdqa  .Lk_mc_forward+48(%rip), %xmm5
 207         pxor    %xmm2,  %xmm0
 208         add     \$16,   %r9
 209         add     %r10,   %r11
 210         jmp     .Ldec_entry
 211
 212 .align 16
 213 .Ldec_loop:
 214 ##
 215 ##  Inverse mix columns
 216 ##
 217         movdqa  -0x20(%r10),%xmm4       # 4 : sb9u
 218         movdqa  -0x10(%r10),%xmm1       # 0 : sb9t
 219         pshufb  %xmm2,  %xmm4           # 4 = sb9u
 220         pshufb  %xmm3,  %xmm1           # 0 = sb9t
 221         pxor    %xmm4,  %xmm0
 222         movdqa  0x00(%r10),%xmm4        # 4 : sbdu
 223         pxor    %xmm1,  %xmm0           # 0 = ch
 224         movdqa  0x10(%r10),%xmm1        # 0 : sbdt
 225
 226         pshufb  %xmm2,  %xmm4           # 4 = sbdu
 227         pshufb  %xmm5,  %xmm0           # MC ch
 228         pshufb  %xmm3,  %xmm1           # 0 = sbdt
 229         pxor    %xmm4,  %xmm0           # 4 = ch
 230         movdqa  0x20(%r10),%xmm4        # 4 : sbbu
 231         pxor    %xmm1,  %xmm0           # 0 = ch
 232         movdqa  0x30(%r10),%xmm1        # 0 : sbbt
 233
 234         pshufb  %xmm2,  %xmm4           # 4 = sbbu
 235         pshufb  %xmm5,  %xmm0           # MC ch
 236         pshufb  %xmm3,  %xmm1           # 0 = sbbt
 237         pxor    %xmm4,  %xmm0           # 4 = ch
 238         movdqa  0x40(%r10),%xmm4        # 4 : sbeu
 239         pxor    %xmm1,  %xmm0           # 0 = ch
 240         movdqa  0x50(%r10),%xmm1        # 0 : sbet
 241
 242         pshufb  %xmm2,  %xmm4           # 4 = sbeu
 243         pshufb  %xmm5,  %xmm0           # MC ch
 244         pshufb  %xmm3,  %xmm1           # 0 = sbet
 245         pxor    %xmm4,  %xmm0           # 4 = ch
 246         add     \$16, %r9               # next round key
 247         palignr \$12,   %xmm5,  %xmm5
 248         pxor    %xmm1,  %xmm0           # 0 = ch
 249         sub     \$1,%rax                # nr--
 250
 251 .Ldec_entry:
 252         # top of round
 253         movdqa  %xmm9,  %xmm1   # 1 : i
 254         pandn   %xmm0,  %xmm1   # 1 = i<<4
 255         movdqa  %xmm11, %xmm2   # 2 : a/k
 256         psrld   \$4,    %xmm1   # 1 = i
 257         pand    %xmm9,  %xmm0   # 0 = k
 258         pshufb  %xmm0,  %xmm2   # 2 = a/k
 259         movdqa  %xmm10, %xmm3   # 3 : 1/i
 260         pxor    %xmm1,  %xmm0   # 0 = j
 261         pshufb  %xmm1,  %xmm3   # 3 = 1/i
 262         movdqa  %xmm10, %xmm4   # 4 : 1/j
 263         pxor    %xmm2,  %xmm3   # 3 = iak = 1/i + a/k
 264         pshufb  %xmm0,  %xmm4   # 4 = 1/j
 265         pxor    %xmm2,  %xmm4   # 4 = jak = 1/j + a/k
 266         movdqa  %xmm10, %xmm2   # 2 : 1/iak
 267         pshufb  %xmm3,  %xmm2   # 2 = 1/iak
 268         movdqa  %xmm10, %xmm3   # 3 : 1/jak
 269         pxor    %xmm0,  %xmm2   # 2 = io
 270         pshufb  %xmm4,  %xmm3   # 3 = 1/jak
 271         movdqu  (%r9),  %xmm0
 272         pxor    %xmm1,  %xmm3   # 3 = jo
 273         jnz     .Ldec_loop
 274
 275         # middle of last round
 276         movdqa  0x60(%r10), %xmm4       # 3 : sbou
 277         pshufb  %xmm2,  %xmm4   # 4 = sbou
 278         pxor    %xmm0,  %xmm4   # 4 = sb1u + k
 279         movdqa  0x70(%r10), %xmm0       # 0 : sbot
 280         movdqa  -0x160(%r11), %xmm2     # .Lk_sr-.Lk_dsbd=-0x160
 281         pshufb  %xmm3,  %xmm0   # 0 = sb1t
 282         pxor    %xmm4,  %xmm0   # 0 = A
 283         pshufb  %xmm2,  %xmm0
 284         ret
 285 .cfi_endproc
 286 .size   _vpaes_decrypt_core,.-_vpaes_decrypt_core
 287
 288 ########################################################
 289 ##                                                    ##
 290 ##                  AES key schedule                  ##
 291 ##                                                    ##
 292 ########################################################
 293 .type   _vpaes_schedule_core,\@abi-omnipotent
 294 .align  16
 295 _vpaes_schedule_core:
 296 .cfi_startproc
 297         # rdi = key
 298         # rsi = size in bits
 299         # rdx = buffer
 300         # rcx = direction.  0=encrypt, 1=decrypt
 301
 302         call    _vpaes_preheat          # load the tables
 303         movdqa  .Lk_rcon(%rip), %xmm8   # load rcon
 304         movdqu  (%rdi), %xmm0           # load key (unaligned)
 305
 306         # input transform
 307         movdqa  %xmm0,  %xmm3
 308         lea     .Lk_ipt(%rip), %r11
 309         call    _vpaes_schedule_transform
 310         movdqa  %xmm0,  %xmm7
 311
 312         lea     .Lk_sr(%rip),%r10
 313         test    %rcx,   %rcx
 314         jnz     .Lschedule_am_decrypting
 315
 316         # encrypting, output zeroth round key after transform
 317         movdqu  %xmm0,  (%rdx)
 318         jmp     .Lschedule_go
 319
 320 .Lschedule_am_decrypting:
 321         # decrypting, output zeroth round key after shiftrows
 322         movdqa  (%r8,%r10),%xmm1
 323         pshufb  %xmm1,  %xmm3
 324         movdqu  %xmm3,  (%rdx)
 325         xor     \$0x30, %r8
 326
 327 .Lschedule_go:
 328         cmp     \$192,  %esi
 329         ja      .Lschedule_256
 330         je      .Lschedule_192
 331         # 128: fall though
 332
 333 ##
 334 ##  .schedule_128
 335 ##
 336 ##  128-bit specific part of key schedule.
 337 ##
 338 ##  This schedule is really simple, because all its parts
 339 ##  are accomplished by the subroutines.
 340 ##
 341 .Lschedule_128:
 342         mov     \$10, %esi
 343
 344 .Loop_schedule_128:
 345         call    _vpaes_schedule_round
 346         dec     %rsi
 347         jz      .Lschedule_mangle_last
 348         call    _vpaes_schedule_mangle  # write output
 349         jmp     .Loop_schedule_128
 350
 351 ##
 352 ##  .aes_schedule_192
 353 ##
 354 ##  192-bit specific part of key schedule.
 355 ##
 356 ##  The main body of this schedule is the same as the 128-bit
 357 ##  schedule, but with more smearing.  The long, high side is
 358 ##  stored in %xmm7 as before, and the short, low side is in
 359 ##  the high bits of %xmm6.
 360 ##
 361 ##  This schedule is somewhat nastier, however, because each
 362 ##  round produces 192 bits of key material, or 1.5 round keys.
 363 ##  Therefore, on each cycle we do 2 rounds and produce 3 round
 364 ##  keys.
 365 ##
 366 .align  16
 367 .Lschedule_192:
 368         movdqu  8(%rdi),%xmm0           # load key part 2 (very unaligned)
 369         call    _vpaes_schedule_transform       # input transform
 370         movdqa  %xmm0,  %xmm6           # save short part
 371         pxor    %xmm4,  %xmm4           # clear 4
 372         movhlps %xmm4,  %xmm6           # clobber low side with zeros
 373         mov     \$4,    %esi
 374
 375 .Loop_schedule_192:
 376         call    _vpaes_schedule_round
 377         palignr \$8,%xmm6,%xmm0
 378         call    _vpaes_schedule_mangle  # save key n
 379         call    _vpaes_schedule_192_smear
 380         call    _vpaes_schedule_mangle  # save key n+1
 381         call    _vpaes_schedule_round
 382         dec     %rsi
 383         jz      .Lschedule_mangle_last
 384         call    _vpaes_schedule_mangle  # save key n+2
 385         call    _vpaes_schedule_192_smear
 386         jmp     .Loop_schedule_192
 387
 388 ##
 389 ##  .aes_schedule_256
 390 ##
 391 ##  256-bit specific part of key schedule.
 392 ##
 393 ##  The structure here is very similar to the 128-bit
 394 ##  schedule, but with an additional "low side" in
 395 ##  %xmm6.  The low side's rounds are the same as the
 396 ##  high side's, except no rcon and no rotation.
 397 ##
 398 .align  16
 399 .Lschedule_256:
 400         movdqu  16(%rdi),%xmm0          # load key part 2 (unaligned)
 401         call    _vpaes_schedule_transform       # input transform
 402         mov     \$7, %esi
 403
 404 .Loop_schedule_256:
 405         call    _vpaes_schedule_mangle  # output low result
 406         movdqa  %xmm0,  %xmm6           # save cur_lo in xmm6
 407
 408         # high round
 409         call    _vpaes_schedule_round
 410         dec     %rsi
 411         jz      .Lschedule_mangle_last
 412         call    _vpaes_schedule_mangle
 413
 414         # low round. swap xmm7 and xmm6
 415         pshufd  \$0xFF, %xmm0,  %xmm0
 416         movdqa  %xmm7,  %xmm5
 417         movdqa  %xmm6,  %xmm7
 418         call    _vpaes_schedule_low_round
 419         movdqa  %xmm5,  %xmm7
 420
 421         jmp     .Loop_schedule_256
 422
 423
 424 ##
 425 ##  .aes_schedule_mangle_last
 426 ##
 427 ##  Mangler for last round of key schedule
 428 ##  Mangles %xmm0
 429 ##    when encrypting, outputs out(%xmm0) ^ 63
 430 ##    when decrypting, outputs unskew(%xmm0)
 431 ##
 432 ##  Always called right before return... jumps to cleanup and exits
 433 ##
 434 .align  16
 435 .Lschedule_mangle_last:
 436         # schedule last round key from xmm0
 437         lea     .Lk_deskew(%rip),%r11   # prepare to deskew
 438         test    %rcx,   %rcx
 439         jnz     .Lschedule_mangle_last_dec
 440
 441         # encrypting
 442         movdqa  (%r8,%r10),%xmm1
 443         pshufb  %xmm1,  %xmm0           # output permute
 444         lea     .Lk_opt(%rip),  %r11    # prepare to output transform
 445         add     \$32,   %rdx
 446
 447 .Lschedule_mangle_last_dec:
 448         add     \$-16,  %rdx
 449         pxor    .Lk_s63(%rip),  %xmm0
 450         call    _vpaes_schedule_transform # output transform
 451         movdqu  %xmm0,  (%rdx)          # save last key
 452
 453         # cleanup
 454         pxor    %xmm0,  %xmm0
 455         pxor    %xmm1,  %xmm1
 456         pxor    %xmm2,  %xmm2
 457         pxor    %xmm3,  %xmm3
 458         pxor    %xmm4,  %xmm4
 459         pxor    %xmm5,  %xmm5
 460         pxor    %xmm6,  %xmm6
 461         pxor    %xmm7,  %xmm7
 462         ret
 463 .cfi_endproc
 464 .size   _vpaes_schedule_core,.-_vpaes_schedule_core
 465
 466 ##
 467 ##  .aes_schedule_192_smear
 468 ##
 469 ##  Smear the short, low side in the 192-bit key schedule.
 470 ##
 471 ##  Inputs:
 472 ##    %xmm7: high side, b  a  x  y
 473 ##    %xmm6:  low side, d  c  0  0
 474 ##    %xmm13: 0
 475 ##
 476 ##  Outputs:
 477 ##    %xmm6: b+c+d  b+c  0  0
 478 ##    %xmm0: b+c+d  b+c  b  a
 479 ##
 480 .type   _vpaes_schedule_192_smear,\@abi-omnipotent
 481 .align  16
 482 _vpaes_schedule_192_smear:
 483 .cfi_startproc
 484         pshufd  \$0x80, %xmm6,  %xmm1   # d c 0 0 -> c 0 0 0
 485         pshufd  \$0xFE, %xmm7,  %xmm0   # b a _ _ -> b b b a
 486         pxor    %xmm1,  %xmm6           # -> c+d c 0 0
 487         pxor    %xmm1,  %xmm1
 488         pxor    %xmm0,  %xmm6           # -> b+c+d b+c b a
 489         movdqa  %xmm6,  %xmm0
 490         movhlps %xmm1,  %xmm6           # clobber low side with zeros
 491         ret
 492 .cfi_endproc
 493 .size   _vpaes_schedule_192_smear,.-_vpaes_schedule_192_smear
 494
 495 ##
 496 ##  .aes_schedule_round
 497 ##
 498 ##  Runs one main round of the key schedule on %xmm0, %xmm7
 499 ##
 500 ##  Specifically, runs subbytes on the high dword of %xmm0
 501 ##  then rotates it by one byte and xors into the low dword of
 502 ##  %xmm7.
 503 ##
 504 ##  Adds rcon from low byte of %xmm8, then rotates %xmm8 for
 505 ##  next rcon.
 506 ##
 507 ##  Smears the dwords of %xmm7 by xoring the low into the
 508 ##  second low, result into third, result into highest.
 509 ##
 510 ##  Returns results in %xmm7 = %xmm0.
 511 ##  Clobbers %xmm1-%xmm4, %r11.
 512 ##
 513 .type   _vpaes_schedule_round,\@abi-omnipotent
 514 .align  16
 515 _vpaes_schedule_round:
 516 .cfi_startproc
 517         # extract rcon from xmm8
 518         pxor    %xmm1,  %xmm1
 519         palignr \$15,   %xmm8,  %xmm1
 520         palignr \$15,   %xmm8,  %xmm8
 521         pxor    %xmm1,  %xmm7
 522
 523         # rotate
 524         pshufd  \$0xFF, %xmm0,  %xmm0
 525         palignr \$1,    %xmm0,  %xmm0
 526
 527         # fall through...
 528
 529         # low round: same as high round, but no rotation and no rcon.
 530 _vpaes_schedule_low_round:
 531         # smear xmm7
 532         movdqa  %xmm7,  %xmm1
 533         pslldq  \$4,    %xmm7
 534         pxor    %xmm1,  %xmm7
 535         movdqa  %xmm7,  %xmm1
 536         pslldq  \$8,    %xmm7
 537         pxor    %xmm1,  %xmm7
 538         pxor    .Lk_s63(%rip), %xmm7
 539
 540         # subbytes
 541         movdqa  %xmm9,  %xmm1
 542         pandn   %xmm0,  %xmm1
 543         psrld   \$4,    %xmm1           # 1 = i
 544         pand    %xmm9,  %xmm0           # 0 = k
 545         movdqa  %xmm11, %xmm2           # 2 : a/k
 546         pshufb  %xmm0,  %xmm2           # 2 = a/k
 547         pxor    %xmm1,  %xmm0           # 0 = j
 548         movdqa  %xmm10, %xmm3           # 3 : 1/i
 549         pshufb  %xmm1,  %xmm3           # 3 = 1/i
 550         pxor    %xmm2,  %xmm3           # 3 = iak = 1/i + a/k
 551         movdqa  %xmm10, %xmm4           # 4 : 1/j
 552         pshufb  %xmm0,  %xmm4           # 4 = 1/j
 553         pxor    %xmm2,  %xmm4           # 4 = jak = 1/j + a/k
 554         movdqa  %xmm10, %xmm2           # 2 : 1/iak
 555         pshufb  %xmm3,  %xmm2           # 2 = 1/iak
 556         pxor    %xmm0,  %xmm2           # 2 = io
 557         movdqa  %xmm10, %xmm3           # 3 : 1/jak
 558         pshufb  %xmm4,  %xmm3           # 3 = 1/jak
 559         pxor    %xmm1,  %xmm3           # 3 = jo
 560         movdqa  %xmm13, %xmm4           # 4 : sbou
 561         pshufb  %xmm2,  %xmm4           # 4 = sbou
 562         movdqa  %xmm12, %xmm0           # 0 : sbot
 563         pshufb  %xmm3,  %xmm0           # 0 = sb1t
 564         pxor    %xmm4,  %xmm0           # 0 = sbox output
 565
 566         # add in smeared stuff
 567         pxor    %xmm7,  %xmm0
 568         movdqa  %xmm0,  %xmm7
 569         ret
 570 .cfi_endproc
 571 .size   _vpaes_schedule_round,.-_vpaes_schedule_round
 572
 573 ##
 574 ##  .aes_schedule_transform
 575 ##
 576 ##  Linear-transform %xmm0 according to tables at (%r11)
 577 ##
 578 ##  Requires that %xmm9 = 0x0F0F... as in preheat
 579 ##  Output in %xmm0
 580 ##  Clobbers %xmm1, %xmm2
 581 ##
 582 .type   _vpaes_schedule_transform,\@abi-omnipotent
 583 .align  16
 584 _vpaes_schedule_transform:
 585 .cfi_startproc
 586         movdqa  %xmm9,  %xmm1
 587         pandn   %xmm0,  %xmm1
 588         psrld   \$4,    %xmm1
 589         pand    %xmm9,  %xmm0
 590         movdqa  (%r11), %xmm2   # lo
 591         pshufb  %xmm0,  %xmm2
 592         movdqa  16(%r11), %xmm0 # hi
 593         pshufb  %xmm1,  %xmm0
 594         pxor    %xmm2,  %xmm0
 595         ret
 596 .cfi_endproc
 597 .size   _vpaes_schedule_transform,.-_vpaes_schedule_transform
 598
 599 ##
 600 ##  .aes_schedule_mangle
 601 ##
 602 ##  Mangle xmm0 from (basis-transformed) standard version
 603 ##  to our version.
 604 ##
 605 ##  On encrypt,
 606 ##    xor with 0x63
 607 ##    multiply by circulant 0,1,1,1
 608 ##    apply shiftrows transform
 609 ##
 610 ##  On decrypt,
 611 ##    xor with 0x63
 612 ##    multiply by "inverse mixcolumns" circulant E,B,D,9
 613 ##    deskew
 614 ##    apply shiftrows transform
 615 ##
 616 ##
 617 ##  Writes out to (%rdx), and increments or decrements it
 618 ##  Keeps track of round number mod 4 in %r8
 619 ##  Preserves xmm0
 620 ##  Clobbers xmm1-xmm5
 621 ##
 622 .type   _vpaes_schedule_mangle,\@abi-omnipotent
 623 .align  16
 624 _vpaes_schedule_mangle:
 625 .cfi_startproc
 626         movdqa  %xmm0,  %xmm4   # save xmm0 for later
 627         movdqa  .Lk_mc_forward(%rip),%xmm5
 628         test    %rcx,   %rcx
 629         jnz     .Lschedule_mangle_dec
 630
 631         # encrypting
 632         add     \$16,   %rdx
 633         pxor    .Lk_s63(%rip),%xmm4
 634         pshufb  %xmm5,  %xmm4
 635         movdqa  %xmm4,  %xmm3
 636         pshufb  %xmm5,  %xmm4
 637         pxor    %xmm4,  %xmm3
 638         pshufb  %xmm5,  %xmm4
 639         pxor    %xmm4,  %xmm3
 640
 641         jmp     .Lschedule_mangle_both
 642 .align  16
 643 .Lschedule_mangle_dec:
 644         # inverse mix columns
 645         lea     .Lk_dksd(%rip),%r11
 646         movdqa  %xmm9,  %xmm1
 647         pandn   %xmm4,  %xmm1
 648         psrld   \$4,    %xmm1   # 1 = hi
 649         pand    %xmm9,  %xmm4   # 4 = lo
 650
 651         movdqa  0x00(%r11), %xmm2
 652         pshufb  %xmm4,  %xmm2
 653         movdqa  0x10(%r11), %xmm3
 654         pshufb  %xmm1,  %xmm3
 655         pxor    %xmm2,  %xmm3
 656         pshufb  %xmm5,  %xmm3
 657
 658         movdqa  0x20(%r11), %xmm2
 659         pshufb  %xmm4,  %xmm2
 660         pxor    %xmm3,  %xmm2
 661         movdqa  0x30(%r11), %xmm3
 662         pshufb  %xmm1,  %xmm3
 663         pxor    %xmm2,  %xmm3
 664         pshufb  %xmm5,  %xmm3
 665
 666         movdqa  0x40(%r11), %xmm2
 667         pshufb  %xmm4,  %xmm2
 668         pxor    %xmm3,  %xmm2
 669         movdqa  0x50(%r11), %xmm3
 670         pshufb  %xmm1,  %xmm3
 671         pxor    %xmm2,  %xmm3
 672         pshufb  %xmm5,  %xmm3
 673
 674         movdqa  0x60(%r11), %xmm2
 675         pshufb  %xmm4,  %xmm2
 676         pxor    %xmm3,  %xmm2
 677         movdqa  0x70(%r11), %xmm3
 678         pshufb  %xmm1,  %xmm3
 679         pxor    %xmm2,  %xmm3
 680
 681         add     \$-16,  %rdx
 682
 683 .Lschedule_mangle_both:
 684         movdqa  (%r8,%r10),%xmm1
 685         pshufb  %xmm1,%xmm3
 686         add     \$-16,  %r8
 687         and     \$0x30, %r8
 688         movdqu  %xmm3,  (%rdx)
 689         ret
 690 .cfi_endproc
 691 .size   _vpaes_schedule_mangle,.-_vpaes_schedule_mangle
 692
 693 #
 694 # Interface to OpenSSL
 695 #
 696 .globl  ${PREFIX}_set_encrypt_key
 697 .type   ${PREFIX}_set_encrypt_key,\@function,3
 698 .align  16
 699 ${PREFIX}_set_encrypt_key:
 700 .cfi_startproc
 701         endbranch
 702 ___
 703 $code.=<<___ if ($win64);
 704         lea     -0xb8(%rsp),%rsp
 705         movaps  %xmm6,0x10(%rsp)
 706         movaps  %xmm7,0x20(%rsp)
 707         movaps  %xmm8,0x30(%rsp)
 708         movaps  %xmm9,0x40(%rsp)
 709         movaps  %xmm10,0x50(%rsp)
 710         movaps  %xmm11,0x60(%rsp)
 711         movaps  %xmm12,0x70(%rsp)
 712         movaps  %xmm13,0x80(%rsp)
 713         movaps  %xmm14,0x90(%rsp)
 714         movaps  %xmm15,0xa0(%rsp)
 715 .Lenc_key_body:
 716 ___
 717 $code.=<<___;
 718         mov     %esi,%eax
 719         shr     \$5,%eax
 720         add     \$5,%eax
 721         mov     %eax,240(%rdx)  # AES_KEY->rounds = nbits/32+5;
 722
 723         mov     \$0,%ecx
 724         mov     \$0x30,%r8d
 725         call    _vpaes_schedule_core
 726 ___
 727 $code.=<<___ if ($win64);
 728         movaps  0x10(%rsp),%xmm6
 729         movaps  0x20(%rsp),%xmm7
 730         movaps  0x30(%rsp),%xmm8
 731         movaps  0x40(%rsp),%xmm9
 732         movaps  0x50(%rsp),%xmm10
 733         movaps  0x60(%rsp),%xmm11
 734         movaps  0x70(%rsp),%xmm12
 735         movaps  0x80(%rsp),%xmm13
 736         movaps  0x90(%rsp),%xmm14
 737         movaps  0xa0(%rsp),%xmm15
 738         lea     0xb8(%rsp),%rsp
 739 .Lenc_key_epilogue:
 740 ___
 741 $code.=<<___;
 742         xor     %eax,%eax
 743         ret
 744 .cfi_endproc
 745 .size   ${PREFIX}_set_encrypt_key,.-${PREFIX}_set_encrypt_key
 746
 747 .globl  ${PREFIX}_set_decrypt_key
 748 .type   ${PREFIX}_set_decrypt_key,\@function,3
 749 .align  16
 750 ${PREFIX}_set_decrypt_key:
 751 .cfi_startproc
 752         endbranch
 753 ___
 754 $code.=<<___ if ($win64);
 755         lea     -0xb8(%rsp),%rsp
 756         movaps  %xmm6,0x10(%rsp)
 757         movaps  %xmm7,0x20(%rsp)
 758         movaps  %xmm8,0x30(%rsp)
 759         movaps  %xmm9,0x40(%rsp)
 760         movaps  %xmm10,0x50(%rsp)
 761         movaps  %xmm11,0x60(%rsp)
 762         movaps  %xmm12,0x70(%rsp)
 763         movaps  %xmm13,0x80(%rsp)
 764         movaps  %xmm14,0x90(%rsp)
 765         movaps  %xmm15,0xa0(%rsp)
 766 .Ldec_key_body:
 767 ___
 768 $code.=<<___;
 769         mov     %esi,%eax
 770         shr     \$5,%eax
 771         add     \$5,%eax
 772         mov     %eax,240(%rdx)  # AES_KEY->rounds = nbits/32+5;
 773         shl     \$4,%eax
 774         lea     16(%rdx,%rax),%rdx
 775
 776         mov     \$1,%ecx
 777         mov     %esi,%r8d
 778         shr     \$1,%r8d
 779         and     \$32,%r8d
 780         xor     \$32,%r8d       # nbits==192?0:32
 781         call    _vpaes_schedule_core
 782 ___
 783 $code.=<<___ if ($win64);
 784         movaps  0x10(%rsp),%xmm6
 785         movaps  0x20(%rsp),%xmm7
 786         movaps  0x30(%rsp),%xmm8
 787         movaps  0x40(%rsp),%xmm9
 788         movaps  0x50(%rsp),%xmm10
 789         movaps  0x60(%rsp),%xmm11
 790         movaps  0x70(%rsp),%xmm12
 791         movaps  0x80(%rsp),%xmm13
 792         movaps  0x90(%rsp),%xmm14
 793         movaps  0xa0(%rsp),%xmm15
 794         lea     0xb8(%rsp),%rsp
 795 .Ldec_key_epilogue:
 796 ___
 797 $code.=<<___;
 798         xor     %eax,%eax
 799         ret
 800 .cfi_endproc
 801 .size   ${PREFIX}_set_decrypt_key,.-${PREFIX}_set_decrypt_key
 802
 803 .globl  ${PREFIX}_encrypt
 804 .type   ${PREFIX}_encrypt,\@function,3
 805 .align  16
 806 ${PREFIX}_encrypt:
 807 .cfi_startproc
 808         endbranch
 809 ___
 810 $code.=<<___ if ($win64);
 811         lea     -0xb8(%rsp),%rsp
 812         movaps  %xmm6,0x10(%rsp)
 813         movaps  %xmm7,0x20(%rsp)
 814         movaps  %xmm8,0x30(%rsp)
 815         movaps  %xmm9,0x40(%rsp)
 816         movaps  %xmm10,0x50(%rsp)
 817         movaps  %xmm11,0x60(%rsp)
 818         movaps  %xmm12,0x70(%rsp)
 819         movaps  %xmm13,0x80(%rsp)
 820         movaps  %xmm14,0x90(%rsp)
 821         movaps  %xmm15,0xa0(%rsp)
 822 .Lenc_body:
 823 ___
 824 $code.=<<___;
 825         movdqu  (%rdi),%xmm0
 826         call    _vpaes_preheat
 827         call    _vpaes_encrypt_core
 828         movdqu  %xmm0,(%rsi)
 829 ___
 830 $code.=<<___ if ($win64);
 831         movaps  0x10(%rsp),%xmm6
 832         movaps  0x20(%rsp),%xmm7
 833         movaps  0x30(%rsp),%xmm8
 834         movaps  0x40(%rsp),%xmm9
 835         movaps  0x50(%rsp),%xmm10
 836         movaps  0x60(%rsp),%xmm11
 837         movaps  0x70(%rsp),%xmm12
 838         movaps  0x80(%rsp),%xmm13
 839         movaps  0x90(%rsp),%xmm14
 840         movaps  0xa0(%rsp),%xmm15
 841         lea     0xb8(%rsp),%rsp
 842 .Lenc_epilogue:
 843 ___
 844 $code.=<<___;
 845         ret
 846 .cfi_endproc
 847 .size   ${PREFIX}_encrypt,.-${PREFIX}_encrypt
 848
 849 .globl  ${PREFIX}_decrypt
 850 .type   ${PREFIX}_decrypt,\@function,3
 851 .align  16
 852 ${PREFIX}_decrypt:
 853 .cfi_startproc
 854         endbranch
 855 ___
 856 $code.=<<___ if ($win64);
 857         lea     -0xb8(%rsp),%rsp
 858         movaps  %xmm6,0x10(%rsp)
 859         movaps  %xmm7,0x20(%rsp)
 860         movaps  %xmm8,0x30(%rsp)
 861         movaps  %xmm9,0x40(%rsp)
 862         movaps  %xmm10,0x50(%rsp)
 863         movaps  %xmm11,0x60(%rsp)
 864         movaps  %xmm12,0x70(%rsp)
 865         movaps  %xmm13,0x80(%rsp)
 866         movaps  %xmm14,0x90(%rsp)
 867         movaps  %xmm15,0xa0(%rsp)
 868 .Ldec_body:
 869 ___
 870 $code.=<<___;
 871         movdqu  (%rdi),%xmm0
 872         call    _vpaes_preheat
 873         call    _vpaes_decrypt_core
 874         movdqu  %xmm0,(%rsi)
 875 ___
 876 $code.=<<___ if ($win64);
 877         movaps  0x10(%rsp),%xmm6
 878         movaps  0x20(%rsp),%xmm7
 879         movaps  0x30(%rsp),%xmm8
 880         movaps  0x40(%rsp),%xmm9
 881         movaps  0x50(%rsp),%xmm10
 882         movaps  0x60(%rsp),%xmm11
 883         movaps  0x70(%rsp),%xmm12
 884         movaps  0x80(%rsp),%xmm13
 885         movaps  0x90(%rsp),%xmm14
 886         movaps  0xa0(%rsp),%xmm15
 887         lea     0xb8(%rsp),%rsp
 888 .Ldec_epilogue:
 889 ___
 890 $code.=<<___;
 891         ret
 892 .cfi_endproc
 893 .size   ${PREFIX}_decrypt,.-${PREFIX}_decrypt
 894 ___
 895 {
 896 my ($inp,$out,$len,$key,$ivp,$enc)=("%rdi","%rsi","%rdx","%rcx","%r8","%r9");
 897 # void AES_cbc_encrypt (const void char *inp, unsigned char *out,
 898 #                       size_t length, const AES_KEY *key,
 899 #                       unsigned char *ivp,const int enc);
 900 $code.=<<___;
 901 .globl  ${PREFIX}_cbc_encrypt
 902 .type   ${PREFIX}_cbc_encrypt,\@function,6
 903 .align  16
 904 ${PREFIX}_cbc_encrypt:
 905 .cfi_startproc
 906         endbranch
 907         xchg    $key,$len
 908 ___
 909 ($len,$key)=($key,$len);
 910 $code.=<<___;
 911         sub     \$16,$len
 912         jc      .Lcbc_abort
 913 ___
 914 $code.=<<___ if ($win64);
 915         lea     -0xb8(%rsp),%rsp
 916         movaps  %xmm6,0x10(%rsp)
 917         movaps  %xmm7,0x20(%rsp)
 918         movaps  %xmm8,0x30(%rsp)
 919         movaps  %xmm9,0x40(%rsp)
 920         movaps  %xmm10,0x50(%rsp)
 921         movaps  %xmm11,0x60(%rsp)
 922         movaps  %xmm12,0x70(%rsp)
 923         movaps  %xmm13,0x80(%rsp)
 924         movaps  %xmm14,0x90(%rsp)
 925         movaps  %xmm15,0xa0(%rsp)
 926 .Lcbc_body:
 927 ___
 928 $code.=<<___;
 929         movdqu  ($ivp),%xmm6            # load IV
 930         sub     $inp,$out
 931         call    _vpaes_preheat
 932         cmp     \$0,${enc}d
 933         je      .Lcbc_dec_loop
 934         jmp     .Lcbc_enc_loop
 935 .align  16
 936 .Lcbc_enc_loop:
 937         movdqu  ($inp),%xmm0
 938         pxor    %xmm6,%xmm0
 939         call    _vpaes_encrypt_core
 940         movdqa  %xmm0,%xmm6
 941         movdqu  %xmm0,($out,$inp)
 942         lea     16($inp),$inp
 943         sub     \$16,$len
 944         jnc     .Lcbc_enc_loop
 945         jmp     .Lcbc_done
 946 .align  16
 947 .Lcbc_dec_loop:
 948         movdqu  ($inp),%xmm0
 949         movdqa  %xmm0,%xmm7
 950         call    _vpaes_decrypt_core
 951         pxor    %xmm6,%xmm0
 952         movdqa  %xmm7,%xmm6
 953         movdqu  %xmm0,($out,$inp)
 954         lea     16($inp),$inp
 955         sub     \$16,$len
 956         jnc     .Lcbc_dec_loop
 957 .Lcbc_done:
 958         movdqu  %xmm6,($ivp)            # save IV
 959 ___
 960 $code.=<<___ if ($win64);
 961         movaps  0x10(%rsp),%xmm6
 962         movaps  0x20(%rsp),%xmm7
 963         movaps  0x30(%rsp),%xmm8
 964         movaps  0x40(%rsp),%xmm9
 965         movaps  0x50(%rsp),%xmm10
 966         movaps  0x60(%rsp),%xmm11
 967         movaps  0x70(%rsp),%xmm12
 968         movaps  0x80(%rsp),%xmm13
 969         movaps  0x90(%rsp),%xmm14
 970         movaps  0xa0(%rsp),%xmm15
 971         lea     0xb8(%rsp),%rsp
 972 .Lcbc_epilogue:
 973 ___
 974 $code.=<<___;
 975 .Lcbc_abort:
 976         ret
 977 .cfi_endproc
 978 .size   ${PREFIX}_cbc_encrypt,.-${PREFIX}_cbc_encrypt
 979 ___
 980 }
 981 $code.=<<___;
 982 ##
 983 ##  _aes_preheat
 984 ##
 985 ##  Fills register %r10 -> .aes_consts (so you can -fPIC)
 986 ##  and %xmm9-%xmm15 as specified below.
 987 ##
 988 .type   _vpaes_preheat,\@abi-omnipotent
 989 .align  16
 990 _vpaes_preheat:
 991 .cfi_startproc
 992         lea     .Lk_s0F(%rip), %r10
 993         movdqa  -0x20(%r10), %xmm10     # .Lk_inv
 994         movdqa  -0x10(%r10), %xmm11     # .Lk_inv+16
 995         movdqa  0x00(%r10), %xmm9       # .Lk_s0F
 996         movdqa  0x30(%r10), %xmm13      # .Lk_sb1
 997         movdqa  0x40(%r10), %xmm12      # .Lk_sb1+16
 998         movdqa  0x50(%r10), %xmm15      # .Lk_sb2
 999         movdqa  0x60(%r10), %xmm14      # .Lk_sb2+16
1000         ret
1001 .cfi_endproc
1002 .size   _vpaes_preheat,.-_vpaes_preheat
1003 ########################################################
1004 ##                                                    ##
1005 ##                     Constants                      ##
1006 ##                                                    ##
1007 ########################################################
1008 .type   _vpaes_consts,\@object
1009 .align  64
1010 _vpaes_consts:
1011 .Lk_inv:        # inv, inva
1012         .quad   0x0E05060F0D080180, 0x040703090A0B0C02
1013         .quad   0x01040A060F0B0780, 0x030D0E0C02050809
1014
1015 .Lk_s0F:        # s0F
1016         .quad   0x0F0F0F0F0F0F0F0F, 0x0F0F0F0F0F0F0F0F
1017
1018 .Lk_ipt:        # input transform (lo, hi)
1019         .quad   0xC2B2E8985A2A7000, 0xCABAE09052227808
1020         .quad   0x4C01307D317C4D00, 0xCD80B1FCB0FDCC81
1021
1022 .Lk_sb1:        # sb1u, sb1t
1023         .quad   0xB19BE18FCB503E00, 0xA5DF7A6E142AF544
1024         .quad   0x3618D415FAE22300, 0x3BF7CCC10D2ED9EF
1025 .Lk_sb2:        # sb2u, sb2t
1026         .quad   0xE27A93C60B712400, 0x5EB7E955BC982FCD
1027         .quad   0x69EB88400AE12900, 0xC2A163C8AB82234A
1028 .Lk_sbo:        # sbou, sbot
1029         .quad   0xD0D26D176FBDC700, 0x15AABF7AC502A878
1030         .quad   0xCFE474A55FBB6A00, 0x8E1E90D1412B35FA
1031
1032 .Lk_mc_forward: # mc_forward
1033         .quad   0x0407060500030201, 0x0C0F0E0D080B0A09
1034         .quad   0x080B0A0904070605, 0x000302010C0F0E0D
1035         .quad   0x0C0F0E0D080B0A09, 0x0407060500030201
1036         .quad   0x000302010C0F0E0D, 0x080B0A0904070605
1037
1038 .Lk_mc_backward:# mc_backward
1039         .quad   0x0605040702010003, 0x0E0D0C0F0A09080B
1040         .quad   0x020100030E0D0C0F, 0x0A09080B06050407
1041         .quad   0x0E0D0C0F0A09080B, 0x0605040702010003
1042         .quad   0x0A09080B06050407, 0x020100030E0D0C0F
1043
1044 .Lk_sr:         # sr
1045         .quad   0x0706050403020100, 0x0F0E0D0C0B0A0908
1046         .quad   0x030E09040F0A0500, 0x0B06010C07020D08
1047         .quad   0x0F060D040B020900, 0x070E050C030A0108
1048         .quad   0x0B0E0104070A0D00, 0x0306090C0F020508
1049
1050 .Lk_rcon:       # rcon
1051         .quad   0x1F8391B9AF9DEEB6, 0x702A98084D7C7D81
1052
1053 .Lk_s63:        # s63: all equal to 0x63 transformed
1054         .quad   0x5B5B5B5B5B5B5B5B, 0x5B5B5B5B5B5B5B5B
1055
1056 .Lk_opt:        # output transform
1057         .quad   0xFF9F4929D6B66000, 0xF7974121DEBE6808
1058         .quad   0x01EDBD5150BCEC00, 0xE10D5DB1B05C0CE0
1059
1060 .Lk_deskew:     # deskew tables: inverts the sbox's "skew"
1061         .quad   0x07E4A34047A4E300, 0x1DFEB95A5DBEF91A
1062         .quad   0x5F36B5DC83EA6900, 0x2841C2ABF49D1E77
1063
1064 ##
1065 ##  Decryption stuff
1066 ##  Key schedule constants
1067 ##
1068 .Lk_dksd:       # decryption key schedule: invskew x*D
1069         .quad   0xFEB91A5DA3E44700, 0x0740E3A45A1DBEF9
1070         .quad   0x41C277F4B5368300, 0x5FDC69EAAB289D1E
1071 .Lk_dksb:       # decryption key schedule: invskew x*B
1072         .quad   0x9A4FCA1F8550D500, 0x03D653861CC94C99
1073         .quad   0x115BEDA7B6FC4A00, 0xD993256F7E3482C8
1074 .Lk_dkse:       # decryption key schedule: invskew x*E + 0x63
1075         .quad   0xD5031CCA1FC9D600, 0x53859A4C994F5086
1076         .quad   0xA23196054FDC7BE8, 0xCD5EF96A20B31487
1077 .Lk_dks9:       # decryption key schedule: invskew x*9
1078         .quad   0xB6116FC87ED9A700, 0x4AED933482255BFC
1079         .quad   0x4576516227143300, 0x8BB89FACE9DAFDCE
1080
1081 ##
1082 ##  Decryption stuff
1083 ##  Round function constants
1084 ##
1085 .Lk_dipt:       # decryption input transform
1086         .quad   0x0F505B040B545F00, 0x154A411E114E451A
1087         .quad   0x86E383E660056500, 0x12771772F491F194
1088
1089 .Lk_dsb9:       # decryption sbox output *9*u, *9*t
1090         .quad   0x851C03539A86D600, 0xCAD51F504F994CC9
1091         .quad   0xC03B1789ECD74900, 0x725E2C9EB2FBA565
1092 .Lk_dsbd:       # decryption sbox output *D*u, *D*t
1093         .quad   0x7D57CCDFE6B1A200, 0xF56E9B13882A4439
1094         .quad   0x3CE2FAF724C6CB00, 0x2931180D15DEEFD3
1095 .Lk_dsbb:       # decryption sbox output *B*u, *B*t
1096         .quad   0xD022649296B44200, 0x602646F6B0F2D404
1097         .quad   0xC19498A6CD596700, 0xF3FF0C3E3255AA6B
1098 .Lk_dsbe:       # decryption sbox output *E*u, *E*t
1099         .quad   0x46F2929626D4D000, 0x2242600464B4F6B0
1100         .quad   0x0C55A6CDFFAAC100, 0x9467F36B98593E32
1101 .Lk_dsbo:       # decryption sbox final output
1102         .quad   0x1387EA537EF94000, 0xC7AA6DB9D4943E2D
1103         .quad   0x12D7560F93441D00, 0xCA4B8159D8C58E9C
1104 .asciz  "Vector Permutation AES for x86_64/SSSE3, Mike Hamburg (Stanford University)"
1105 .align  64
1106 .size   _vpaes_consts,.-_vpaes_consts
1107 ___
1108
1109 if ($win64) {
1110 # EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
1111 #               CONTEXT *context,DISPATCHER_CONTEXT *disp)
1112 $rec="%rcx";
1113 $frame="%rdx";
1114 $context="%r8";
1115 $disp="%r9";
1116
1117 $code.=<<___;
1118 .extern __imp_RtlVirtualUnwind
1119 .type   se_handler,\@abi-omnipotent
1120 .align  16
1121 se_handler:
1122         push    %rsi
1123         push    %rdi
1124         push    %rbx
1125         push    %rbp
1126         push    %r12
1127         push    %r13
1128         push    %r14
1129         push    %r15
1130         pushfq
1131         sub     \$64,%rsp
1132
1133         mov     120($context),%rax      # pull context->Rax
1134         mov     248($context),%rbx      # pull context->Rip
1135
1136         mov     8($disp),%rsi           # disp->ImageBase
1137         mov     56($disp),%r11          # disp->HandlerData
1138
1139         mov     0(%r11),%r10d           # HandlerData[0]
1140         lea     (%rsi,%r10),%r10        # prologue label
1141         cmp     %r10,%rbx               # context->Rip<prologue label
1142         jb      .Lin_prologue
1143
1144         mov     152($context),%rax      # pull context->Rsp
1145
1146         mov     4(%r11),%r10d           # HandlerData[1]
1147         lea     (%rsi,%r10),%r10        # epilogue label
1148         cmp     %r10,%rbx               # context->Rip>=epilogue label
1149         jae     .Lin_prologue
1150
1151         lea     16(%rax),%rsi           # %xmm save area
1152         lea     512($context),%rdi      # &context.Xmm6
1153         mov     \$20,%ecx               # 10*sizeof(%xmm0)/sizeof(%rax)
1154         .long   0xa548f3fc              # cld; rep movsq
1155         lea     0xb8(%rax),%rax         # adjust stack pointer
1156
1157 .Lin_prologue:
1158         mov     8(%rax),%rdi
1159         mov     16(%rax),%rsi
1160         mov     %rax,152($context)      # restore context->Rsp
1161         mov     %rsi,168($context)      # restore context->Rsi
1162         mov     %rdi,176($context)      # restore context->Rdi
1163
1164         mov     40($disp),%rdi          # disp->ContextRecord
1165         mov     $context,%rsi           # context
1166         mov     \$`1232/8`,%ecx         # sizeof(CONTEXT)
1167         .long   0xa548f3fc              # cld; rep movsq
1168
1169         mov     $disp,%rsi
1170         xor     %rcx,%rcx               # arg1, UNW_FLAG_NHANDLER
1171         mov     8(%rsi),%rdx            # arg2, disp->ImageBase
1172         mov     0(%rsi),%r8             # arg3, disp->ControlPc
1173         mov     16(%rsi),%r9            # arg4, disp->FunctionEntry
1174         mov     40(%rsi),%r10           # disp->ContextRecord
1175         lea     56(%rsi),%r11           # &disp->HandlerData
1176         lea     24(%rsi),%r12           # &disp->EstablisherFrame
1177         mov     %r10,32(%rsp)           # arg5
1178         mov     %r11,40(%rsp)           # arg6
1179         mov     %r12,48(%rsp)           # arg7
1180         mov     %rcx,56(%rsp)           # arg8, (NULL)
1181         call    *__imp_RtlVirtualUnwind(%rip)
1182
1183         mov     \$1,%eax                # ExceptionContinueSearch
1184         add     \$64,%rsp
1185         popfq
1186         pop     %r15
1187         pop     %r14
1188         pop     %r13
1189         pop     %r12
1190         pop     %rbp
1191         pop     %rbx
1192         pop     %rdi
1193         pop     %rsi
1194         ret
1195 .size   se_handler,.-se_handler
1196
1197 .section        .pdata
1198 .align  4
1199         .rva    .LSEH_begin_${PREFIX}_set_encrypt_key
1200         .rva    .LSEH_end_${PREFIX}_set_encrypt_key
1201         .rva    .LSEH_info_${PREFIX}_set_encrypt_key
1202
1203         .rva    .LSEH_begin_${PREFIX}_set_decrypt_key
1204         .rva    .LSEH_end_${PREFIX}_set_decrypt_key
1205         .rva    .LSEH_info_${PREFIX}_set_decrypt_key
1206
1207         .rva    .LSEH_begin_${PREFIX}_encrypt
1208         .rva    .LSEH_end_${PREFIX}_encrypt
1209         .rva    .LSEH_info_${PREFIX}_encrypt
1210
1211         .rva    .LSEH_begin_${PREFIX}_decrypt
1212         .rva    .LSEH_end_${PREFIX}_decrypt
1213         .rva    .LSEH_info_${PREFIX}_decrypt
1214
1215         .rva    .LSEH_begin_${PREFIX}_cbc_encrypt
1216         .rva    .LSEH_end_${PREFIX}_cbc_encrypt
1217         .rva    .LSEH_info_${PREFIX}_cbc_encrypt
1218
1219 .section        .xdata
1220 .align  8
1221 .LSEH_info_${PREFIX}_set_encrypt_key:
1222         .byte   9,0,0,0
1223         .rva    se_handler
1224         .rva    .Lenc_key_body,.Lenc_key_epilogue       # HandlerData[]
1225 .LSEH_info_${PREFIX}_set_decrypt_key:
1226         .byte   9,0,0,0
1227         .rva    se_handler
1228         .rva    .Ldec_key_body,.Ldec_key_epilogue       # HandlerData[]
1229 .LSEH_info_${PREFIX}_encrypt:
1230         .byte   9,0,0,0
1231         .rva    se_handler
1232         .rva    .Lenc_body,.Lenc_epilogue               # HandlerData[]
1233 .LSEH_info_${PREFIX}_decrypt:
1234         .byte   9,0,0,0
1235         .rva    se_handler
1236         .rva    .Ldec_body,.Ldec_epilogue               # HandlerData[]
1237 .LSEH_info_${PREFIX}_cbc_encrypt:
1238         .byte   9,0,0,0
1239         .rva    se_handler
1240         .rva    .Lcbc_body,.Lcbc_epilogue               # HandlerData[]
1241 ___
1242 }
1243
1244 $code =~ s/\`([^\`]*)\`/eval($1)/gem;
1245
1246 print $code;
1247
1248 close STDOUT or die "error closing STDOUT: $!";