release/src/router/openssl/crypto/aes/asm/aesni-x86_64.pl

   1 #!/usr/bin/env perl
   2 #
   3 # ====================================================================
   4 # Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
   5 # project. The module is, however, dual licensed under OpenSSL and
   6 # CRYPTOGAMS licenses depending on where you obtain it. For further
   7 # details see http://www.openssl.org/~appro/cryptogams/.
   8 # ====================================================================
   9 #
  10 # This module implements support for Intel AES-NI extension. In
  11 # OpenSSL context it's used with Intel engine, but can also be used as
  12 # drop-in replacement for crypto/aes/asm/aes-x86_64.pl [see below for
  13 # details].
  14 #
  15 # Performance.
  16 #
  17 # Given aes(enc|dec) instructions' latency asymptotic performance for
  18 # non-parallelizable modes such as CBC encrypt is 3.75 cycles per byte
  19 # processed with 128-bit key. And given their throughput asymptotic
  20 # performance for parallelizable modes is 1.25 cycles per byte. Being
  21 # asymptotic limit it's not something you commonly achieve in reality,
  22 # but how close does one get? Below are results collected for
  23 # different modes and block sized. Pairs of numbers are for en-/
  24 # decryption.
  25 #
  26 #       16-byte     64-byte     256-byte    1-KB        8-KB
  27 # ECB   4.25/4.25   1.38/1.38   1.28/1.28   1.26/1.26   1.26/1.26
  28 # CTR   5.42/5.42   1.92/1.92   1.44/1.44   1.28/1.28   1.26/1.26
  29 # CBC   4.38/4.43   4.15/1.43   4.07/1.32   4.07/1.29   4.06/1.28
  30 # CCM   5.66/9.42   4.42/5.41   4.16/4.40   4.09/4.15   4.06/4.07
  31 # OFB   5.42/5.42   4.64/4.64   4.44/4.44   4.39/4.39   4.38/4.38
  32 # CFB   5.73/5.85   5.56/5.62   5.48/5.56   5.47/5.55   5.47/5.55
  33 #
  34 # ECB, CTR, CBC and CCM results are free from EVP overhead. This means
  35 # that otherwise used 'openssl speed -evp aes-128-??? -engine aesni
  36 # [-decrypt]' will exhibit 10-15% worse results for smaller blocks.
  37 # The results were collected with specially crafted speed.c benchmark
  38 # in order to compare them with results reported in "Intel Advanced
  39 # Encryption Standard (AES) New Instruction Set" White Paper Revision
  40 # 3.0 dated May 2010. All above results are consistently better. This
  41 # module also provides better performance for block sizes smaller than
  42 # 128 bytes in points *not* represented in the above table.
  43 #
  44 # Looking at the results for 8-KB buffer.
  45 #
  46 # CFB and OFB results are far from the limit, because implementation
  47 # uses "generic" CRYPTO_[c|o]fb128_encrypt interfaces relying on
  48 # single-block aesni_encrypt, which is not the most optimal way to go.
  49 # CBC encrypt result is unexpectedly high and there is no documented
  50 # explanation for it. Seemingly there is a small penalty for feeding
  51 # the result back to AES unit the way it's done in CBC mode. There is
  52 # nothing one can do and the result appears optimal. CCM result is
  53 # identical to CBC, because CBC-MAC is essentially CBC encrypt without
  54 # saving output. CCM CTR "stays invisible," because it's neatly
  55 # interleaved wih CBC-MAC. This provides ~30% improvement over
  56 # "straghtforward" CCM implementation with CTR and CBC-MAC performed
  57 # disjointly. Parallelizable modes practically achieve the theoretical
  58 # limit.
  59 #
  60 # Looking at how results vary with buffer size.
  61 #
  62 # Curves are practically saturated at 1-KB buffer size. In most cases
  63 # "256-byte" performance is >95%, and "64-byte" is ~90% of "8-KB" one.
  64 # CTR curve doesn't follow this pattern and is "slowest" changing one
  65 # with "256-byte" result being 87% of "8-KB." This is because overhead
  66 # in CTR mode is most computationally intensive. Small-block CCM
  67 # decrypt is slower than encrypt, because first CTR and last CBC-MAC
  68 # iterations can't be interleaved.
  69 #
  70 # Results for 192- and 256-bit keys.
  71 #
  72 # EVP-free results were observed to scale perfectly with number of
  73 # rounds for larger block sizes, i.e. 192-bit result being 10/12 times
  74 # lower and 256-bit one - 10/14. Well, in CBC encrypt case differences
  75 # are a tad smaller, because the above mentioned penalty biases all
  76 # results by same constant value. In similar way function call
  77 # overhead affects small-block performance, as well as OFB and CFB
  78 # results. Differences are not large, most common coefficients are
  79 # 10/11.7 and 10/13.4 (as opposite to 10/12.0 and 10/14.0), but one
  80 # observe even 10/11.2 and 10/12.4 (CTR, OFB, CFB)...
  81
  82 # January 2011
  83 #
  84 # While Westmere processor features 6 cycles latency for aes[enc|dec]
  85 # instructions, which can be scheduled every second cycle, Sandy
  86 # Bridge spends 8 cycles per instruction, but it can schedule them
  87 # every cycle. This means that code targeting Westmere would perform
  88 # suboptimally on Sandy Bridge. Therefore this update.
  89 #
  90 # In addition, non-parallelizable CBC encrypt (as well as CCM) is
  91 # optimized. Relative improvement might appear modest, 8% on Westmere,
  92 # but in absolute terms it's 3.77 cycles per byte encrypted with
  93 # 128-bit key on Westmere, and 5.07 - on Sandy Bridge. These numbers
  94 # should be compared to asymptotic limits of 3.75 for Westmere and
  95 # 5.00 for Sandy Bridge. Actually, the fact that they get this close
  96 # to asymptotic limits is quite amazing. Indeed, the limit is
  97 # calculated as latency times number of rounds, 10 for 128-bit key,
  98 # and divided by 16, the number of bytes in block, or in other words
  99 # it accounts *solely* for aesenc instructions. But there are extra
 100 # instructions, and numbers so close to the asymptotic limits mean
 101 # that it's as if it takes as little as *one* additional cycle to
 102 # execute all of them. How is it possible? It is possible thanks to
 103 # out-of-order execution logic, which manages to overlap post-
 104 # processing of previous block, things like saving the output, with
 105 # actual encryption of current block, as well as pre-processing of
 106 # current block, things like fetching input and xor-ing it with
 107 # 0-round element of the key schedule, with actual encryption of
 108 # previous block. Keep this in mind...
 109 #
 110 # For parallelizable modes, such as ECB, CBC decrypt, CTR, higher
 111 # performance is achieved by interleaving instructions working on
 112 # independent blocks. In which case asymptotic limit for such modes
 113 # can be obtained by dividing above mentioned numbers by AES
 114 # instructions' interleave factor. Westmere can execute at most 3
 115 # instructions at a time, meaning that optimal interleave factor is 3,
 116 # and that's where the "magic" number of 1.25 come from. "Optimal
 117 # interleave factor" means that increase of interleave factor does
 118 # not improve performance. The formula has proven to reflect reality
 119 # pretty well on Westmere... Sandy Bridge on the other hand can
 120 # execute up to 8 AES instructions at a time, so how does varying
 121 # interleave factor affect the performance? Here is table for ECB
 122 # (numbers are cycles per byte processed with 128-bit key):
 123 #
 124 # instruction interleave factor         3x      6x      8x
 125 # theoretical asymptotic limit          1.67    0.83    0.625
 126 # measured performance for 8KB block    1.05    0.86    0.84
 127 #
 128 # "as if" interleave factor             4.7x    5.8x    6.0x
 129 #
 130 # Further data for other parallelizable modes:
 131 #
 132 # CBC decrypt                           1.16    0.93    0.74
 133 # CTR                                   1.14    0.91    0.74
 134 #
 135 # Well, given 3x column it's probably inappropriate to call the limit
 136 # asymptotic, if it can be surpassed, isn't it? What happens there?
 137 # Rewind to CBC paragraph for the answer. Yes, out-of-order execution
 138 # magic is responsible for this. Processor overlaps not only the
 139 # additional instructions with AES ones, but even AES instuctions
 140 # processing adjacent triplets of independent blocks. In the 6x case
 141 # additional instructions  still claim disproportionally small amount
 142 # of additional cycles, but in 8x case number of instructions must be
 143 # a tad too high for out-of-order logic to cope with, and AES unit
 144 # remains underutilized... As you can see 8x interleave is hardly
 145 # justifiable, so there no need to feel bad that 32-bit aesni-x86.pl
 146 # utilizies 6x interleave because of limited register bank capacity.
 147 #
 148 # Higher interleave factors do have negative impact on Westmere
 149 # performance. While for ECB mode it's negligible ~1.5%, other
 150 # parallelizables perform ~5% worse, which is outweighed by ~25%
 151 # improvement on Sandy Bridge. To balance regression on Westmere
 152 # CTR mode was implemented with 6x aesenc interleave factor.
 153
 154 # April 2011
 155 #
 156 # Add aesni_xts_[en|de]crypt. Westmere spends 1.25 cycles processing
 157 # one byte out of 8KB with 128-bit key, Sandy Bridge - 0.90. Just like
 158 # in CTR mode AES instruction interleave factor was chosen to be 6x.
 159
 160 ######################################################################
 161 # Current large-block performance in cycles per byte processed with
 162 # 128-bit key (less is better).
 163 #
 164 #               CBC en-/decrypt CTR     XTS     ECB
 165 # Westmere      3.77/1.25       1.25    1.25    1.26
 166 # * Bridge      5.07/0.74       0.75    0.90    0.85
 167 # Haswell       4.44/0.63       0.63    0.73    0.63
 168 # Atom          5.75/3.54       3.56    4.12    3.87(*)
 169 # Bulldozer     5.77/0.70       0.72    0.90    0.70
 170 #
 171 # (*)   Atom ECB result is suboptimal because of penalties incurred
 172 #       by operations on %xmm8-15. As ECB is not considered
 173 #       critical, nothing was done to mitigate the problem.
 174
 175 $PREFIX="aesni";        # if $PREFIX is set to "AES", the script
 176                         # generates drop-in replacement for
 177                         # crypto/aes/asm/aes-x86_64.pl:-)
 178
 179 $flavour = shift;
 180 $output  = shift;
 181 if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
 182
 183 $win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);
 184
 185 $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
 186 ( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
 187 ( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or
 188 die "can't locate x86_64-xlate.pl";
 189
 190 open OUT,"| \"$^X\" $xlate $flavour $output";
 191 *STDOUT=*OUT;
 192
 193 $movkey = $PREFIX eq "aesni" ? "movups" : "movups";
 194 @_4args=$win64? ("%rcx","%rdx","%r8", "%r9") :  # Win64 order
 195                 ("%rdi","%rsi","%rdx","%rcx");  # Unix order
 196
 197 $code=".text\n";
 198 $code.=".extern OPENSSL_ia32cap_P\n";
 199
 200 $rounds="%eax"; # input to and changed by aesni_[en|de]cryptN !!!
 201 # this is natural Unix argument order for public $PREFIX_[ecb|cbc]_encrypt ...
 202 $inp="%rdi";
 203 $out="%rsi";
 204 $len="%rdx";
 205 $key="%rcx";    # input to and changed by aesni_[en|de]cryptN !!!
 206 $ivp="%r8";     # cbc, ctr, ...
 207
 208 $rnds_="%r10d"; # backup copy for $rounds
 209 $key_="%r11";   # backup copy for $key
 210
 211 # %xmm register layout
 212 $rndkey0="%xmm0";       $rndkey1="%xmm1";
 213 $inout0="%xmm2";        $inout1="%xmm3";
 214 $inout2="%xmm4";        $inout3="%xmm5";
 215 $inout4="%xmm6";        $inout5="%xmm7";
 216 $inout6="%xmm8";        $inout7="%xmm9";
 217
 218 $in2="%xmm6";           $in1="%xmm7";   # used in CBC decrypt, CTR, ...
 219 $in0="%xmm8";           $iv="%xmm9";
 220 \f
 221 # Inline version of internal aesni_[en|de]crypt1.
 222 #
 223 # Why folded loop? Because aes[enc|dec] is slow enough to accommodate
 224 # cycles which take care of loop variables...
 225 { my $sn;
 226 sub aesni_generate1 {
 227 my ($p,$key,$rounds,$inout,$ivec)=@_;   $inout=$inout0 if (!defined($inout));
 228 ++$sn;
 229 $code.=<<___;
 230         $movkey ($key),$rndkey0
 231         $movkey 16($key),$rndkey1
 232 ___
 233 $code.=<<___ if (defined($ivec));
 234         xorps   $rndkey0,$ivec
 235         lea     32($key),$key
 236         xorps   $ivec,$inout
 237 ___
 238 $code.=<<___ if (!defined($ivec));
 239         lea     32($key),$key
 240         xorps   $rndkey0,$inout
 241 ___
 242 $code.=<<___;
 243 .Loop_${p}1_$sn:
 244         aes${p} $rndkey1,$inout
 245         dec     $rounds
 246         $movkey ($key),$rndkey1
 247         lea     16($key),$key
 248         jnz     .Loop_${p}1_$sn # loop body is 16 bytes
 249         aes${p}last     $rndkey1,$inout
 250 ___
 251 }}
 252 # void $PREFIX_[en|de]crypt (const void *inp,void *out,const AES_KEY *key);
 253 #
 254 { my ($inp,$out,$key) = @_4args;
 255
 256 $code.=<<___;
 257 .globl  ${PREFIX}_encrypt
 258 .type   ${PREFIX}_encrypt,\@abi-omnipotent
 259 .align  16
 260 ${PREFIX}_encrypt:
 261         movups  ($inp),$inout0          # load input
 262         mov     240($key),$rounds       # key->rounds
 263 ___
 264         &aesni_generate1("enc",$key,$rounds);
 265 $code.=<<___;
 266         movups  $inout0,($out)          # output
 267         ret
 268 .size   ${PREFIX}_encrypt,.-${PREFIX}_encrypt
 269
 270 .globl  ${PREFIX}_decrypt
 271 .type   ${PREFIX}_decrypt,\@abi-omnipotent
 272 .align  16
 273 ${PREFIX}_decrypt:
 274         movups  ($inp),$inout0          # load input
 275         mov     240($key),$rounds       # key->rounds
 276 ___
 277         &aesni_generate1("dec",$key,$rounds);
 278 $code.=<<___;
 279         movups  $inout0,($out)          # output
 280         ret
 281 .size   ${PREFIX}_decrypt, .-${PREFIX}_decrypt
 282 ___
 283 }
 284 \f
 285 # _aesni_[en|de]cryptN are private interfaces, N denotes interleave
 286 # factor. Why 3x subroutine were originally used in loops? Even though
 287 # aes[enc|dec] latency was originally 6, it could be scheduled only
 288 # every *2nd* cycle. Thus 3x interleave was the one providing optimal
 289 # utilization, i.e. when subroutine's throughput is virtually same as
 290 # of non-interleaved subroutine [for number of input blocks up to 3].
 291 # This is why it originally made no sense to implement 2x subroutine.
 292 # But times change and it became appropriate to spend extra 192 bytes
 293 # on 2x subroutine on Atom Silvermont account. For processors that
 294 # can schedule aes[enc|dec] every cycle optimal interleave factor
 295 # equals to corresponding instructions latency. 8x is optimal for
 296 # * Bridge and "super-optimal" for other Intel CPUs...
 297
 298 sub aesni_generate2 {
 299 my $dir=shift;
 300 # As already mentioned it takes in $key and $rounds, which are *not*
 301 # preserved. $inout[0-1] is cipher/clear text...
 302 $code.=<<___;
 303 .type   _aesni_${dir}rypt2,\@abi-omnipotent
 304 .align  16
 305 _aesni_${dir}rypt2:
 306         $movkey ($key),$rndkey0
 307         shl     \$4,$rounds
 308         $movkey 16($key),$rndkey1
 309         xorps   $rndkey0,$inout0
 310         xorps   $rndkey0,$inout1
 311         $movkey 32($key),$rndkey0
 312         lea     32($key,$rounds),$key
 313         neg     %rax                            # $rounds
 314         add     \$16,%rax
 315
 316 .L${dir}_loop2:
 317         aes${dir}       $rndkey1,$inout0
 318         aes${dir}       $rndkey1,$inout1
 319         $movkey         ($key,%rax),$rndkey1
 320         add             \$32,%rax
 321         aes${dir}       $rndkey0,$inout0
 322         aes${dir}       $rndkey0,$inout1
 323         $movkey         -16($key,%rax),$rndkey0
 324         jnz             .L${dir}_loop2
 325
 326         aes${dir}       $rndkey1,$inout0
 327         aes${dir}       $rndkey1,$inout1
 328         aes${dir}last   $rndkey0,$inout0
 329         aes${dir}last   $rndkey0,$inout1
 330         ret
 331 .size   _aesni_${dir}rypt2,.-_aesni_${dir}rypt2
 332 ___
 333 }
 334 sub aesni_generate3 {
 335 my $dir=shift;
 336 # As already mentioned it takes in $key and $rounds, which are *not*
 337 # preserved. $inout[0-2] is cipher/clear text...
 338 $code.=<<___;
 339 .type   _aesni_${dir}rypt3,\@abi-omnipotent
 340 .align  16
 341 _aesni_${dir}rypt3:
 342         $movkey ($key),$rndkey0
 343         shl     \$4,$rounds
 344         $movkey 16($key),$rndkey1
 345         xorps   $rndkey0,$inout0
 346         xorps   $rndkey0,$inout1
 347         xorps   $rndkey0,$inout2
 348         $movkey 32($key),$rndkey0
 349         lea     32($key,$rounds),$key
 350         neg     %rax                            # $rounds
 351         add     \$16,%rax
 352
 353 .L${dir}_loop3:
 354         aes${dir}       $rndkey1,$inout0
 355         aes${dir}       $rndkey1,$inout1
 356         aes${dir}       $rndkey1,$inout2
 357         $movkey         ($key,%rax),$rndkey1
 358         add             \$32,%rax
 359         aes${dir}       $rndkey0,$inout0
 360         aes${dir}       $rndkey0,$inout1
 361         aes${dir}       $rndkey0,$inout2
 362         $movkey         -16($key,%rax),$rndkey0
 363         jnz             .L${dir}_loop3
 364
 365         aes${dir}       $rndkey1,$inout0
 366         aes${dir}       $rndkey1,$inout1
 367         aes${dir}       $rndkey1,$inout2
 368         aes${dir}last   $rndkey0,$inout0
 369         aes${dir}last   $rndkey0,$inout1
 370         aes${dir}last   $rndkey0,$inout2
 371         ret
 372 .size   _aesni_${dir}rypt3,.-_aesni_${dir}rypt3
 373 ___
 374 }
 375 # 4x interleave is implemented to improve small block performance,
 376 # most notably [and naturally] 4 block by ~30%. One can argue that one
 377 # should have implemented 5x as well, but improvement would be <20%,
 378 # so it's not worth it...
 379 sub aesni_generate4 {
 380 my $dir=shift;
 381 # As already mentioned it takes in $key and $rounds, which are *not*
 382 # preserved. $inout[0-3] is cipher/clear text...
 383 $code.=<<___;
 384 .type   _aesni_${dir}rypt4,\@abi-omnipotent
 385 .align  16
 386 _aesni_${dir}rypt4:
 387         $movkey ($key),$rndkey0
 388         shl     \$4,$rounds
 389         $movkey 16($key),$rndkey1
 390         xorps   $rndkey0,$inout0
 391         xorps   $rndkey0,$inout1
 392         xorps   $rndkey0,$inout2
 393         xorps   $rndkey0,$inout3
 394         $movkey 32($key),$rndkey0
 395         lea     32($key,$rounds),$key
 396         neg     %rax                            # $rounds
 397         .byte   0x0f,0x1f,0x00
 398         add     \$16,%rax
 399
 400 .L${dir}_loop4:
 401         aes${dir}       $rndkey1,$inout0
 402         aes${dir}       $rndkey1,$inout1
 403         aes${dir}       $rndkey1,$inout2
 404         aes${dir}       $rndkey1,$inout3
 405         $movkey         ($key,%rax),$rndkey1
 406         add             \$32,%rax
 407         aes${dir}       $rndkey0,$inout0
 408         aes${dir}       $rndkey0,$inout1
 409         aes${dir}       $rndkey0,$inout2
 410         aes${dir}       $rndkey0,$inout3
 411         $movkey         -16($key,%rax),$rndkey0
 412         jnz             .L${dir}_loop4
 413
 414         aes${dir}       $rndkey1,$inout0
 415         aes${dir}       $rndkey1,$inout1
 416         aes${dir}       $rndkey1,$inout2
 417         aes${dir}       $rndkey1,$inout3
 418         aes${dir}last   $rndkey0,$inout0
 419         aes${dir}last   $rndkey0,$inout1
 420         aes${dir}last   $rndkey0,$inout2
 421         aes${dir}last   $rndkey0,$inout3
 422         ret
 423 .size   _aesni_${dir}rypt4,.-_aesni_${dir}rypt4
 424 ___
 425 }
 426 sub aesni_generate6 {
 427 my $dir=shift;
 428 # As already mentioned it takes in $key and $rounds, which are *not*
 429 # preserved. $inout[0-5] is cipher/clear text...
 430 $code.=<<___;
 431 .type   _aesni_${dir}rypt6,\@abi-omnipotent
 432 .align  16
 433 _aesni_${dir}rypt6:
 434         $movkey         ($key),$rndkey0
 435         shl             \$4,$rounds
 436         $movkey         16($key),$rndkey1
 437         xorps           $rndkey0,$inout0
 438         pxor            $rndkey0,$inout1
 439         pxor            $rndkey0,$inout2
 440         aes${dir}       $rndkey1,$inout0
 441         lea             32($key,$rounds),$key
 442         neg             %rax                    # $rounds
 443         aes${dir}       $rndkey1,$inout1
 444         pxor            $rndkey0,$inout3
 445         pxor            $rndkey0,$inout4
 446         aes${dir}       $rndkey1,$inout2
 447         pxor            $rndkey0,$inout5
 448         add             \$16,%rax
 449         aes${dir}       $rndkey1,$inout3
 450         aes${dir}       $rndkey1,$inout4
 451         aes${dir}       $rndkey1,$inout5
 452         $movkey         -16($key,%rax),$rndkey0
 453         jmp             .L${dir}_loop6_enter
 454 .align  16
 455 .L${dir}_loop6:
 456         aes${dir}       $rndkey1,$inout0
 457         aes${dir}       $rndkey1,$inout1
 458         aes${dir}       $rndkey1,$inout2
 459         aes${dir}       $rndkey1,$inout3
 460         aes${dir}       $rndkey1,$inout4
 461         aes${dir}       $rndkey1,$inout5
 462 .L${dir}_loop6_enter:
 463         $movkey         ($key,%rax),$rndkey1
 464         add             \$32,%rax
 465         aes${dir}       $rndkey0,$inout0
 466         aes${dir}       $rndkey0,$inout1
 467         aes${dir}       $rndkey0,$inout2
 468         aes${dir}       $rndkey0,$inout3
 469         aes${dir}       $rndkey0,$inout4
 470         aes${dir}       $rndkey0,$inout5
 471         $movkey         -16($key,%rax),$rndkey0
 472         jnz             .L${dir}_loop6
 473
 474         aes${dir}       $rndkey1,$inout0
 475         aes${dir}       $rndkey1,$inout1
 476         aes${dir}       $rndkey1,$inout2
 477         aes${dir}       $rndkey1,$inout3
 478         aes${dir}       $rndkey1,$inout4
 479         aes${dir}       $rndkey1,$inout5
 480         aes${dir}last   $rndkey0,$inout0
 481         aes${dir}last   $rndkey0,$inout1
 482         aes${dir}last   $rndkey0,$inout2
 483         aes${dir}last   $rndkey0,$inout3
 484         aes${dir}last   $rndkey0,$inout4
 485         aes${dir}last   $rndkey0,$inout5
 486         ret
 487 .size   _aesni_${dir}rypt6,.-_aesni_${dir}rypt6
 488 ___
 489 }
 490 sub aesni_generate8 {
 491 my $dir=shift;
 492 # As already mentioned it takes in $key and $rounds, which are *not*
 493 # preserved. $inout[0-7] is cipher/clear text...
 494 $code.=<<___;
 495 .type   _aesni_${dir}rypt8,\@abi-omnipotent
 496 .align  16
 497 _aesni_${dir}rypt8:
 498         $movkey         ($key),$rndkey0
 499         shl             \$4,$rounds
 500         $movkey         16($key),$rndkey1
 501         xorps           $rndkey0,$inout0
 502         xorps           $rndkey0,$inout1
 503         pxor            $rndkey0,$inout2
 504         pxor            $rndkey0,$inout3
 505         pxor            $rndkey0,$inout4
 506         lea             32($key,$rounds),$key
 507         neg             %rax                    # $rounds
 508         aes${dir}       $rndkey1,$inout0
 509         add             \$16,%rax
 510         pxor            $rndkey0,$inout5
 511         aes${dir}       $rndkey1,$inout1
 512         pxor            $rndkey0,$inout6
 513         pxor            $rndkey0,$inout7
 514         aes${dir}       $rndkey1,$inout2
 515         aes${dir}       $rndkey1,$inout3
 516         aes${dir}       $rndkey1,$inout4
 517         aes${dir}       $rndkey1,$inout5
 518         aes${dir}       $rndkey1,$inout6
 519         aes${dir}       $rndkey1,$inout7
 520         $movkey         -16($key,%rax),$rndkey0
 521         jmp             .L${dir}_loop8_enter
 522 .align  16
 523 .L${dir}_loop8:
 524         aes${dir}       $rndkey1,$inout0
 525         aes${dir}       $rndkey1,$inout1
 526         aes${dir}       $rndkey1,$inout2
 527         aes${dir}       $rndkey1,$inout3
 528         aes${dir}       $rndkey1,$inout4
 529         aes${dir}       $rndkey1,$inout5
 530         aes${dir}       $rndkey1,$inout6
 531         aes${dir}       $rndkey1,$inout7
 532 .L${dir}_loop8_enter:
 533         $movkey         ($key,%rax),$rndkey1
 534         add             \$32,%rax
 535         aes${dir}       $rndkey0,$inout0
 536         aes${dir}       $rndkey0,$inout1
 537         aes${dir}       $rndkey0,$inout2
 538         aes${dir}       $rndkey0,$inout3
 539         aes${dir}       $rndkey0,$inout4
 540         aes${dir}       $rndkey0,$inout5
 541         aes${dir}       $rndkey0,$inout6
 542         aes${dir}       $rndkey0,$inout7
 543         $movkey         -16($key,%rax),$rndkey0
 544         jnz             .L${dir}_loop8
 545
 546         aes${dir}       $rndkey1,$inout0
 547         aes${dir}       $rndkey1,$inout1
 548         aes${dir}       $rndkey1,$inout2
 549         aes${dir}       $rndkey1,$inout3
 550         aes${dir}       $rndkey1,$inout4
 551         aes${dir}       $rndkey1,$inout5
 552         aes${dir}       $rndkey1,$inout6
 553         aes${dir}       $rndkey1,$inout7
 554         aes${dir}last   $rndkey0,$inout0
 555         aes${dir}last   $rndkey0,$inout1
 556         aes${dir}last   $rndkey0,$inout2
 557         aes${dir}last   $rndkey0,$inout3
 558         aes${dir}last   $rndkey0,$inout4
 559         aes${dir}last   $rndkey0,$inout5
 560         aes${dir}last   $rndkey0,$inout6
 561         aes${dir}last   $rndkey0,$inout7
 562         ret
 563 .size   _aesni_${dir}rypt8,.-_aesni_${dir}rypt8
 564 ___
 565 }
 566 &aesni_generate2("enc") if ($PREFIX eq "aesni");
 567 &aesni_generate2("dec");
 568 &aesni_generate3("enc") if ($PREFIX eq "aesni");
 569 &aesni_generate3("dec");
 570 &aesni_generate4("enc") if ($PREFIX eq "aesni");
 571 &aesni_generate4("dec");
 572 &aesni_generate6("enc") if ($PREFIX eq "aesni");
 573 &aesni_generate6("dec");
 574 &aesni_generate8("enc") if ($PREFIX eq "aesni");
 575 &aesni_generate8("dec");
 576 \f
 577 if ($PREFIX eq "aesni") {
 578 ########################################################################
 579 # void aesni_ecb_encrypt (const void *in, void *out,
 580 #                         size_t length, const AES_KEY *key,
 581 #                         int enc);
 582 $code.=<<___;
 583 .globl  aesni_ecb_encrypt
 584 .type   aesni_ecb_encrypt,\@function,5
 585 .align  16
 586 aesni_ecb_encrypt:
 587 ___
 588 $code.=<<___ if ($win64);
 589         lea     -0x58(%rsp),%rsp
 590         movaps  %xmm6,(%rsp)
 591         movaps  %xmm7,0x10(%rsp)
 592         movaps  %xmm8,0x20(%rsp)
 593         movaps  %xmm9,0x30(%rsp)
 594 .Lecb_enc_body:
 595 ___
 596 $code.=<<___;
 597         and     \$-16,$len
 598         jz      .Lecb_ret
 599
 600         mov     240($key),$rounds       # key->rounds
 601         $movkey ($key),$rndkey0
 602         mov     $key,$key_              # backup $key
 603         mov     $rounds,$rnds_          # backup $rounds
 604         test    %r8d,%r8d               # 5th argument
 605         jz      .Lecb_decrypt
 606 #--------------------------- ECB ENCRYPT ------------------------------#
 607         cmp     \$0x80,$len
 608         jb      .Lecb_enc_tail
 609
 610         movdqu  ($inp),$inout0
 611         movdqu  0x10($inp),$inout1
 612         movdqu  0x20($inp),$inout2
 613         movdqu  0x30($inp),$inout3
 614         movdqu  0x40($inp),$inout4
 615         movdqu  0x50($inp),$inout5
 616         movdqu  0x60($inp),$inout6
 617         movdqu  0x70($inp),$inout7
 618         lea     0x80($inp),$inp
 619         sub     \$0x80,$len
 620         jmp     .Lecb_enc_loop8_enter
 621 .align 16
 622 .Lecb_enc_loop8:
 623         movups  $inout0,($out)
 624         mov     $key_,$key              # restore $key
 625         movdqu  ($inp),$inout0
 626         mov     $rnds_,$rounds          # restore $rounds
 627         movups  $inout1,0x10($out)
 628         movdqu  0x10($inp),$inout1
 629         movups  $inout2,0x20($out)
 630         movdqu  0x20($inp),$inout2
 631         movups  $inout3,0x30($out)
 632         movdqu  0x30($inp),$inout3
 633         movups  $inout4,0x40($out)
 634         movdqu  0x40($inp),$inout4
 635         movups  $inout5,0x50($out)
 636         movdqu  0x50($inp),$inout5
 637         movups  $inout6,0x60($out)
 638         movdqu  0x60($inp),$inout6
 639         movups  $inout7,0x70($out)
 640         lea     0x80($out),$out
 641         movdqu  0x70($inp),$inout7
 642         lea     0x80($inp),$inp
 643 .Lecb_enc_loop8_enter:
 644
 645         call    _aesni_encrypt8
 646
 647         sub     \$0x80,$len
 648         jnc     .Lecb_enc_loop8
 649
 650         movups  $inout0,($out)
 651         mov     $key_,$key              # restore $key
 652         movups  $inout1,0x10($out)
 653         mov     $rnds_,$rounds          # restore $rounds
 654         movups  $inout2,0x20($out)
 655         movups  $inout3,0x30($out)
 656         movups  $inout4,0x40($out)
 657         movups  $inout5,0x50($out)
 658         movups  $inout6,0x60($out)
 659         movups  $inout7,0x70($out)
 660         lea     0x80($out),$out
 661         add     \$0x80,$len
 662         jz      .Lecb_ret
 663
 664 .Lecb_enc_tail:
 665         movups  ($inp),$inout0
 666         cmp     \$0x20,$len
 667         jb      .Lecb_enc_one
 668         movups  0x10($inp),$inout1
 669         je      .Lecb_enc_two
 670         movups  0x20($inp),$inout2
 671         cmp     \$0x40,$len
 672         jb      .Lecb_enc_three
 673         movups  0x30($inp),$inout3
 674         je      .Lecb_enc_four
 675         movups  0x40($inp),$inout4
 676         cmp     \$0x60,$len
 677         jb      .Lecb_enc_five
 678         movups  0x50($inp),$inout5
 679         je      .Lecb_enc_six
 680         movdqu  0x60($inp),$inout6
 681         call    _aesni_encrypt8
 682         movups  $inout0,($out)
 683         movups  $inout1,0x10($out)
 684         movups  $inout2,0x20($out)
 685         movups  $inout3,0x30($out)
 686         movups  $inout4,0x40($out)
 687         movups  $inout5,0x50($out)
 688         movups  $inout6,0x60($out)
 689         jmp     .Lecb_ret
 690 .align  16
 691 .Lecb_enc_one:
 692 ___
 693         &aesni_generate1("enc",$key,$rounds);
 694 $code.=<<___;
 695         movups  $inout0,($out)
 696         jmp     .Lecb_ret
 697 .align  16
 698 .Lecb_enc_two:
 699         call    _aesni_encrypt2
 700         movups  $inout0,($out)
 701         movups  $inout1,0x10($out)
 702         jmp     .Lecb_ret
 703 .align  16
 704 .Lecb_enc_three:
 705         call    _aesni_encrypt3
 706         movups  $inout0,($out)
 707         movups  $inout1,0x10($out)
 708         movups  $inout2,0x20($out)
 709         jmp     .Lecb_ret
 710 .align  16
 711 .Lecb_enc_four:
 712         call    _aesni_encrypt4
 713         movups  $inout0,($out)
 714         movups  $inout1,0x10($out)
 715         movups  $inout2,0x20($out)
 716         movups  $inout3,0x30($out)
 717         jmp     .Lecb_ret
 718 .align  16
 719 .Lecb_enc_five:
 720         xorps   $inout5,$inout5
 721         call    _aesni_encrypt6
 722         movups  $inout0,($out)
 723         movups  $inout1,0x10($out)
 724         movups  $inout2,0x20($out)
 725         movups  $inout3,0x30($out)
 726         movups  $inout4,0x40($out)
 727         jmp     .Lecb_ret
 728 .align  16
 729 .Lecb_enc_six:
 730         call    _aesni_encrypt6
 731         movups  $inout0,($out)
 732         movups  $inout1,0x10($out)
 733         movups  $inout2,0x20($out)
 734         movups  $inout3,0x30($out)
 735         movups  $inout4,0x40($out)
 736         movups  $inout5,0x50($out)
 737         jmp     .Lecb_ret
 738 \f#--------------------------- ECB DECRYPT ------------------------------#
 739 .align  16
 740 .Lecb_decrypt:
 741         cmp     \$0x80,$len
 742         jb      .Lecb_dec_tail
 743
 744         movdqu  ($inp),$inout0
 745         movdqu  0x10($inp),$inout1
 746         movdqu  0x20($inp),$inout2
 747         movdqu  0x30($inp),$inout3
 748         movdqu  0x40($inp),$inout4
 749         movdqu  0x50($inp),$inout5
 750         movdqu  0x60($inp),$inout6
 751         movdqu  0x70($inp),$inout7
 752         lea     0x80($inp),$inp
 753         sub     \$0x80,$len
 754         jmp     .Lecb_dec_loop8_enter
 755 .align 16
 756 .Lecb_dec_loop8:
 757         movups  $inout0,($out)
 758         mov     $key_,$key              # restore $key
 759         movdqu  ($inp),$inout0
 760         mov     $rnds_,$rounds          # restore $rounds
 761         movups  $inout1,0x10($out)
 762         movdqu  0x10($inp),$inout1
 763         movups  $inout2,0x20($out)
 764         movdqu  0x20($inp),$inout2
 765         movups  $inout3,0x30($out)
 766         movdqu  0x30($inp),$inout3
 767         movups  $inout4,0x40($out)
 768         movdqu  0x40($inp),$inout4
 769         movups  $inout5,0x50($out)
 770         movdqu  0x50($inp),$inout5
 771         movups  $inout6,0x60($out)
 772         movdqu  0x60($inp),$inout6
 773         movups  $inout7,0x70($out)
 774         lea     0x80($out),$out
 775         movdqu  0x70($inp),$inout7
 776         lea     0x80($inp),$inp
 777 .Lecb_dec_loop8_enter:
 778
 779         call    _aesni_decrypt8
 780
 781         $movkey ($key_),$rndkey0
 782         sub     \$0x80,$len
 783         jnc     .Lecb_dec_loop8
 784
 785         movups  $inout0,($out)
 786         mov     $key_,$key              # restore $key
 787         movups  $inout1,0x10($out)
 788         mov     $rnds_,$rounds          # restore $rounds
 789         movups  $inout2,0x20($out)
 790         movups  $inout3,0x30($out)
 791         movups  $inout4,0x40($out)
 792         movups  $inout5,0x50($out)
 793         movups  $inout6,0x60($out)
 794         movups  $inout7,0x70($out)
 795         lea     0x80($out),$out
 796         add     \$0x80,$len
 797         jz      .Lecb_ret
 798
 799 .Lecb_dec_tail:
 800         movups  ($inp),$inout0
 801         cmp     \$0x20,$len
 802         jb      .Lecb_dec_one
 803         movups  0x10($inp),$inout1
 804         je      .Lecb_dec_two
 805         movups  0x20($inp),$inout2
 806         cmp     \$0x40,$len
 807         jb      .Lecb_dec_three
 808         movups  0x30($inp),$inout3
 809         je      .Lecb_dec_four
 810         movups  0x40($inp),$inout4
 811         cmp     \$0x60,$len
 812         jb      .Lecb_dec_five
 813         movups  0x50($inp),$inout5
 814         je      .Lecb_dec_six
 815         movups  0x60($inp),$inout6
 816         $movkey ($key),$rndkey0
 817         call    _aesni_decrypt8
 818         movups  $inout0,($out)
 819         movups  $inout1,0x10($out)
 820         movups  $inout2,0x20($out)
 821         movups  $inout3,0x30($out)
 822         movups  $inout4,0x40($out)
 823         movups  $inout5,0x50($out)
 824         movups  $inout6,0x60($out)
 825         jmp     .Lecb_ret
 826 .align  16
 827 .Lecb_dec_one:
 828 ___
 829         &aesni_generate1("dec",$key,$rounds);
 830 $code.=<<___;
 831         movups  $inout0,($out)
 832         jmp     .Lecb_ret
 833 .align  16
 834 .Lecb_dec_two:
 835         call    _aesni_decrypt2
 836         movups  $inout0,($out)
 837         movups  $inout1,0x10($out)
 838         jmp     .Lecb_ret
 839 .align  16
 840 .Lecb_dec_three:
 841         call    _aesni_decrypt3
 842         movups  $inout0,($out)
 843         movups  $inout1,0x10($out)
 844         movups  $inout2,0x20($out)
 845         jmp     .Lecb_ret
 846 .align  16
 847 .Lecb_dec_four:
 848         call    _aesni_decrypt4
 849         movups  $inout0,($out)
 850         movups  $inout1,0x10($out)
 851         movups  $inout2,0x20($out)
 852         movups  $inout3,0x30($out)
 853         jmp     .Lecb_ret
 854 .align  16
 855 .Lecb_dec_five:
 856         xorps   $inout5,$inout5
 857         call    _aesni_decrypt6
 858         movups  $inout0,($out)
 859         movups  $inout1,0x10($out)
 860         movups  $inout2,0x20($out)
 861         movups  $inout3,0x30($out)
 862         movups  $inout4,0x40($out)
 863         jmp     .Lecb_ret
 864 .align  16
 865 .Lecb_dec_six:
 866         call    _aesni_decrypt6
 867         movups  $inout0,($out)
 868         movups  $inout1,0x10($out)
 869         movups  $inout2,0x20($out)
 870         movups  $inout3,0x30($out)
 871         movups  $inout4,0x40($out)
 872         movups  $inout5,0x50($out)
 873
 874 .Lecb_ret:
 875 ___
 876 $code.=<<___ if ($win64);
 877         movaps  (%rsp),%xmm6
 878         movaps  0x10(%rsp),%xmm7
 879         movaps  0x20(%rsp),%xmm8
 880         movaps  0x30(%rsp),%xmm9
 881         lea     0x58(%rsp),%rsp
 882 .Lecb_enc_ret:
 883 ___
 884 $code.=<<___;
 885         ret
 886 .size   aesni_ecb_encrypt,.-aesni_ecb_encrypt
 887 ___
 888 \f
 889 {
 890 ######################################################################
 891 # void aesni_ccm64_[en|de]crypt_blocks (const void *in, void *out,
 892 #                         size_t blocks, const AES_KEY *key,
 893 #                         const char *ivec,char *cmac);
 894 #
 895 # Handles only complete blocks, operates on 64-bit counter and
 896 # does not update *ivec! Nor does it finalize CMAC value
 897 # (see engine/eng_aesni.c for details)
 898 #
 899 {
 900 my $cmac="%r9"; # 6th argument
 901
 902 my $increment="%xmm9";
 903 my $iv="%xmm6";
 904 my $bswap_mask="%xmm7";
 905
 906 $code.=<<___;
 907 .globl  aesni_ccm64_encrypt_blocks
 908 .type   aesni_ccm64_encrypt_blocks,\@function,6
 909 .align  16
 910 aesni_ccm64_encrypt_blocks:
 911 ___
 912 $code.=<<___ if ($win64);
 913         lea     -0x58(%rsp),%rsp
 914         movaps  %xmm6,(%rsp)
 915         movaps  %xmm7,0x10(%rsp)
 916         movaps  %xmm8,0x20(%rsp)
 917         movaps  %xmm9,0x30(%rsp)
 918 .Lccm64_enc_body:
 919 ___
 920 $code.=<<___;
 921         mov     240($key),$rounds               # key->rounds
 922         movdqu  ($ivp),$iv
 923         movdqa  .Lincrement64(%rip),$increment
 924         movdqa  .Lbswap_mask(%rip),$bswap_mask
 925
 926         shl     \$4,$rounds
 927         mov     \$16,$rnds_
 928         lea     0($key),$key_
 929         movdqu  ($cmac),$inout1
 930         movdqa  $iv,$inout0
 931         lea     32($key,$rounds),$key           # end of key schedule
 932         pshufb  $bswap_mask,$iv
 933         sub     %rax,%r10                       # twisted $rounds
 934         jmp     .Lccm64_enc_outer
 935 .align  16
 936 .Lccm64_enc_outer:
 937         $movkey ($key_),$rndkey0
 938         mov     %r10,%rax
 939         movups  ($inp),$in0                     # load inp
 940
 941         xorps   $rndkey0,$inout0                # counter
 942         $movkey 16($key_),$rndkey1
 943         xorps   $in0,$rndkey0
 944         xorps   $rndkey0,$inout1                # cmac^=inp
 945         $movkey 32($key_),$rndkey0
 946
 947 .Lccm64_enc2_loop:
 948         aesenc  $rndkey1,$inout0
 949         aesenc  $rndkey1,$inout1
 950         $movkey ($key,%rax),$rndkey1
 951         add     \$32,%rax
 952         aesenc  $rndkey0,$inout0
 953         aesenc  $rndkey0,$inout1
 954         $movkey -16($key,%rax),$rndkey0
 955         jnz     .Lccm64_enc2_loop
 956         aesenc  $rndkey1,$inout0
 957         aesenc  $rndkey1,$inout1
 958         paddq   $increment,$iv
 959         dec     $len
 960         aesenclast      $rndkey0,$inout0
 961         aesenclast      $rndkey0,$inout1
 962
 963         lea     16($inp),$inp
 964         xorps   $inout0,$in0                    # inp ^= E(iv)
 965         movdqa  $iv,$inout0
 966         movups  $in0,($out)                     # save output
 967         pshufb  $bswap_mask,$inout0
 968         lea     16($out),$out
 969         jnz     .Lccm64_enc_outer
 970
 971         movups  $inout1,($cmac)
 972 ___
 973 $code.=<<___ if ($win64);
 974         movaps  (%rsp),%xmm6
 975         movaps  0x10(%rsp),%xmm7
 976         movaps  0x20(%rsp),%xmm8
 977         movaps  0x30(%rsp),%xmm9
 978         lea     0x58(%rsp),%rsp
 979 .Lccm64_enc_ret:
 980 ___
 981 $code.=<<___;
 982         ret
 983 .size   aesni_ccm64_encrypt_blocks,.-aesni_ccm64_encrypt_blocks
 984 ___
 985 ######################################################################
 986 $code.=<<___;
 987 .globl  aesni_ccm64_decrypt_blocks
 988 .type   aesni_ccm64_decrypt_blocks,\@function,6
 989 .align  16
 990 aesni_ccm64_decrypt_blocks:
 991 ___
 992 $code.=<<___ if ($win64);
 993         lea     -0x58(%rsp),%rsp
 994         movaps  %xmm6,(%rsp)
 995         movaps  %xmm7,0x10(%rsp)
 996         movaps  %xmm8,0x20(%rsp)
 997         movaps  %xmm9,0x30(%rsp)
 998 .Lccm64_dec_body:
 999 ___
1000 $code.=<<___;
1001         mov     240($key),$rounds               # key->rounds
1002         movups  ($ivp),$iv
1003         movdqu  ($cmac),$inout1
1004         movdqa  .Lincrement64(%rip),$increment
1005         movdqa  .Lbswap_mask(%rip),$bswap_mask
1006
1007         movaps  $iv,$inout0
1008         mov     $rounds,$rnds_
1009         mov     $key,$key_
1010         pshufb  $bswap_mask,$iv
1011 ___
1012         &aesni_generate1("enc",$key,$rounds);
1013 $code.=<<___;
1014         shl     \$4,$rnds_
1015         mov     \$16,$rounds
1016         movups  ($inp),$in0                     # load inp
1017         paddq   $increment,$iv
1018         lea     16($inp),$inp
1019         sub     %r10,%rax                       # twisted $rounds
1020         lea     32($key_,$rnds_),$key           # end of key schedule
1021         mov     %rax,%r10
1022         jmp     .Lccm64_dec_outer
1023 .align  16
1024 .Lccm64_dec_outer:
1025         xorps   $inout0,$in0                    # inp ^= E(iv)
1026         movdqa  $iv,$inout0
1027         movups  $in0,($out)                     # save output
1028         lea     16($out),$out
1029         pshufb  $bswap_mask,$inout0
1030
1031         sub     \$1,$len
1032         jz      .Lccm64_dec_break
1033
1034         $movkey ($key_),$rndkey0
1035         mov     %r10,%rax
1036         $movkey 16($key_),$rndkey1
1037         xorps   $rndkey0,$in0
1038         xorps   $rndkey0,$inout0
1039         xorps   $in0,$inout1                    # cmac^=out
1040         $movkey 32($key_),$rndkey0
1041         jmp     .Lccm64_dec2_loop
1042 .align  16
1043 .Lccm64_dec2_loop:
1044         aesenc  $rndkey1,$inout0
1045         aesenc  $rndkey1,$inout1
1046         $movkey ($key,%rax),$rndkey1
1047         add     \$32,%rax
1048         aesenc  $rndkey0,$inout0
1049         aesenc  $rndkey0,$inout1
1050         $movkey -16($key,%rax),$rndkey0
1051         jnz     .Lccm64_dec2_loop
1052         movups  ($inp),$in0                     # load inp
1053         paddq   $increment,$iv
1054         aesenc  $rndkey1,$inout0
1055         aesenc  $rndkey1,$inout1
1056         aesenclast      $rndkey0,$inout0
1057         aesenclast      $rndkey0,$inout1
1058         lea     16($inp),$inp
1059         jmp     .Lccm64_dec_outer
1060
1061 .align  16
1062 .Lccm64_dec_break:
1063         #xorps  $in0,$inout1                    # cmac^=out
1064         mov     240($key_),$rounds
1065 ___
1066         &aesni_generate1("enc",$key_,$rounds,$inout1,$in0);
1067 $code.=<<___;
1068         movups  $inout1,($cmac)
1069 ___
1070 $code.=<<___ if ($win64);
1071         movaps  (%rsp),%xmm6
1072         movaps  0x10(%rsp),%xmm7
1073         movaps  0x20(%rsp),%xmm8
1074         movaps  0x30(%rsp),%xmm9
1075         lea     0x58(%rsp),%rsp
1076 .Lccm64_dec_ret:
1077 ___
1078 $code.=<<___;
1079         ret
1080 .size   aesni_ccm64_decrypt_blocks,.-aesni_ccm64_decrypt_blocks
1081 ___
1082 }\f
1083 ######################################################################
1084 # void aesni_ctr32_encrypt_blocks (const void *in, void *out,
1085 #                         size_t blocks, const AES_KEY *key,
1086 #                         const char *ivec);
1087 #
1088 # Handles only complete blocks, operates on 32-bit counter and
1089 # does not update *ivec! (see crypto/modes/ctr128.c for details)
1090 #
1091 # Overhaul based on suggestions from Shay Gueron and Vlad Krasnov,
1092 # http://rt.openssl.org/Ticket/Display.html?id=3021&user=guest&pass=guest.
1093 # Keywords are full unroll and modulo-schedule counter calculations
1094 # with zero-round key xor.
1095 {
1096 my ($in0,$in1,$in2,$in3,$in4,$in5)=map("%xmm$_",(10..15));
1097 my ($key0,$ctr)=("${key_}d","${ivp}d");
1098 my $frame_size = 0x80 + ($win64?160:0);
1099
1100 $code.=<<___;
1101 .globl  aesni_ctr32_encrypt_blocks
1102 .type   aesni_ctr32_encrypt_blocks,\@function,5
1103 .align  16
1104 aesni_ctr32_encrypt_blocks:
1105         lea     (%rsp),%rax
1106         push    %rbp
1107         sub     \$$frame_size,%rsp
1108         and     \$-16,%rsp      # Linux kernel stack can be incorrectly seeded
1109 ___
1110 $code.=<<___ if ($win64);
1111         movaps  %xmm6,-0xa8(%rax)
1112         movaps  %xmm7,-0x98(%rax)
1113         movaps  %xmm8,-0x88(%rax)
1114         movaps  %xmm9,-0x78(%rax)
1115         movaps  %xmm10,-0x68(%rax)
1116         movaps  %xmm11,-0x58(%rax)
1117         movaps  %xmm12,-0x48(%rax)
1118         movaps  %xmm13,-0x38(%rax)
1119         movaps  %xmm14,-0x28(%rax)
1120         movaps  %xmm15,-0x18(%rax)
1121 .Lctr32_body:
1122 ___
1123 $code.=<<___;
1124         lea     -8(%rax),%rbp
1125
1126         cmp     \$1,$len
1127         je      .Lctr32_one_shortcut
1128
1129         movdqu  ($ivp),$inout0
1130         movdqu  ($key),$rndkey0
1131         mov     12($ivp),$ctr                   # counter LSB
1132         pxor    $rndkey0,$inout0
1133         mov     12($key),$key0                  # 0-round key LSB
1134         movdqa  $inout0,0x00(%rsp)              # populate counter block
1135         bswap   $ctr
1136         movdqa  $inout0,$inout1
1137         movdqa  $inout0,$inout2
1138         movdqa  $inout0,$inout3
1139         movdqa  $inout0,0x40(%rsp)
1140         movdqa  $inout0,0x50(%rsp)
1141         movdqa  $inout0,0x60(%rsp)
1142         mov     %rdx,%r10                       # borrow %rdx
1143         movdqa  $inout0,0x70(%rsp)
1144
1145         lea     1($ctr),%rax
1146          lea    2($ctr),%rdx
1147         bswap   %eax
1148          bswap  %edx
1149         xor     $key0,%eax
1150          xor    $key0,%edx
1151         pinsrd  \$3,%eax,$inout1
1152         lea     3($ctr),%rax
1153         movdqa  $inout1,0x10(%rsp)
1154          pinsrd \$3,%edx,$inout2
1155         bswap   %eax
1156          mov    %r10,%rdx                       # restore %rdx
1157          lea    4($ctr),%r10
1158          movdqa $inout2,0x20(%rsp)
1159         xor     $key0,%eax
1160          bswap  %r10d
1161         pinsrd  \$3,%eax,$inout3
1162          xor    $key0,%r10d
1163         movdqa  $inout3,0x30(%rsp)
1164         lea     5($ctr),%r9
1165          mov    %r10d,0x40+12(%rsp)
1166         bswap   %r9d
1167          lea    6($ctr),%r10
1168         mov     240($key),$rounds               # key->rounds
1169         xor     $key0,%r9d
1170          bswap  %r10d
1171         mov     %r9d,0x50+12(%rsp)
1172          xor    $key0,%r10d
1173         lea     7($ctr),%r9
1174          mov    %r10d,0x60+12(%rsp)
1175         bswap   %r9d
1176          mov    OPENSSL_ia32cap_P+4(%rip),%r10d
1177         xor     $key0,%r9d
1178          and    \$`1<<26|1<<22`,%r10d           # isolate XSAVE+MOVBE
1179         mov     %r9d,0x70+12(%rsp)
1180
1181         $movkey 0x10($key),$rndkey1
1182
1183         movdqa  0x40(%rsp),$inout4
1184         movdqa  0x50(%rsp),$inout5
1185
1186         cmp     \$8,$len
1187         jb      .Lctr32_tail
1188
1189         sub     \$6,$len
1190         cmp     \$`1<<22`,%r10d         # check for MOVBE without XSAVE
1191         je      .Lctr32_6x
1192
1193         lea     0x80($key),$key         # size optimization
1194         sub     \$2,$len
1195         jmp     .Lctr32_loop8
1196
1197 .align  16
1198 .Lctr32_6x:
1199         shl     \$4,$rounds
1200         mov     \$48,$rnds_
1201         bswap   $key0
1202         lea     32($key,$rounds),$key   # end of key schedule
1203         sub     %rax,%r10               # twisted $rounds
1204         jmp     .Lctr32_loop6
1205
1206 .align  16
1207 .Lctr32_loop6:
1208          add    \$6,$ctr
1209         $movkey -48($key,$rnds_),$rndkey0
1210         aesenc  $rndkey1,$inout0
1211          mov    $ctr,%eax
1212          xor    $key0,%eax
1213         aesenc  $rndkey1,$inout1
1214          movbe  %eax,`0x00+12`(%rsp)
1215          lea    1($ctr),%eax
1216         aesenc  $rndkey1,$inout2
1217          xor    $key0,%eax
1218          movbe  %eax,`0x10+12`(%rsp)
1219         aesenc  $rndkey1,$inout3
1220          lea    2($ctr),%eax
1221          xor    $key0,%eax
1222         aesenc  $rndkey1,$inout4
1223          movbe  %eax,`0x20+12`(%rsp)
1224          lea    3($ctr),%eax
1225         aesenc  $rndkey1,$inout5
1226         $movkey -32($key,$rnds_),$rndkey1
1227          xor    $key0,%eax
1228
1229         aesenc  $rndkey0,$inout0
1230          movbe  %eax,`0x30+12`(%rsp)
1231          lea    4($ctr),%eax
1232         aesenc  $rndkey0,$inout1
1233          xor    $key0,%eax
1234          movbe  %eax,`0x40+12`(%rsp)
1235         aesenc  $rndkey0,$inout2
1236          lea    5($ctr),%eax
1237          xor    $key0,%eax
1238         aesenc  $rndkey0,$inout3
1239          movbe  %eax,`0x50+12`(%rsp)
1240          mov    %r10,%rax               # mov   $rnds_,$rounds
1241         aesenc  $rndkey0,$inout4
1242         aesenc  $rndkey0,$inout5
1243         $movkey -16($key,$rnds_),$rndkey0
1244
1245         call    .Lenc_loop6
1246
1247         movdqu  ($inp),$inout6
1248         movdqu  0x10($inp),$inout7
1249         movdqu  0x20($inp),$in0
1250         movdqu  0x30($inp),$in1
1251         movdqu  0x40($inp),$in2
1252         movdqu  0x50($inp),$in3
1253         lea     0x60($inp),$inp
1254         $movkey -64($key,$rnds_),$rndkey1
1255         pxor    $inout0,$inout6
1256         movaps  0x00(%rsp),$inout0
1257         pxor    $inout1,$inout7
1258         movaps  0x10(%rsp),$inout1
1259         pxor    $inout2,$in0
1260         movaps  0x20(%rsp),$inout2
1261         pxor    $inout3,$in1
1262         movaps  0x30(%rsp),$inout3
1263         pxor    $inout4,$in2
1264         movaps  0x40(%rsp),$inout4
1265         pxor    $inout5,$in3
1266         movaps  0x50(%rsp),$inout5
1267         movdqu  $inout6,($out)
1268         movdqu  $inout7,0x10($out)
1269         movdqu  $in0,0x20($out)
1270         movdqu  $in1,0x30($out)
1271         movdqu  $in2,0x40($out)
1272         movdqu  $in3,0x50($out)
1273         lea     0x60($out),$out
1274
1275         sub     \$6,$len
1276         jnc     .Lctr32_loop6
1277
1278         add     \$6,$len
1279         jz      .Lctr32_done
1280
1281         lea     -48($rnds_),$rounds
1282         lea     -80($key,$rnds_),$key   # restore $key
1283         neg     $rounds
1284         shr     \$4,$rounds             # restore $rounds
1285         jmp     .Lctr32_tail
1286
1287 .align  32
1288 .Lctr32_loop8:
1289          add            \$8,$ctr
1290         movdqa          0x60(%rsp),$inout6
1291         aesenc          $rndkey1,$inout0
1292          mov            $ctr,%r9d
1293         movdqa          0x70(%rsp),$inout7
1294         aesenc          $rndkey1,$inout1
1295          bswap          %r9d
1296         $movkey         0x20-0x80($key),$rndkey0
1297         aesenc          $rndkey1,$inout2
1298          xor            $key0,%r9d
1299          nop
1300         aesenc          $rndkey1,$inout3
1301          mov            %r9d,0x00+12(%rsp)
1302          lea            1($ctr),%r9
1303         aesenc          $rndkey1,$inout4
1304         aesenc          $rndkey1,$inout5
1305         aesenc          $rndkey1,$inout6
1306         aesenc          $rndkey1,$inout7
1307         $movkey         0x30-0x80($key),$rndkey1
1308 ___
1309 for($i=2;$i<8;$i++) {
1310 my $rndkeyx = ($i&1)?$rndkey1:$rndkey0;
1311 $code.=<<___;
1312          bswap          %r9d
1313         aesenc          $rndkeyx,$inout0
1314         aesenc          $rndkeyx,$inout1
1315          xor            $key0,%r9d
1316          .byte          0x66,0x90
1317         aesenc          $rndkeyx,$inout2
1318         aesenc          $rndkeyx,$inout3
1319          mov            %r9d,`0x10*($i-1)`+12(%rsp)
1320          lea            $i($ctr),%r9
1321         aesenc          $rndkeyx,$inout4
1322         aesenc          $rndkeyx,$inout5
1323         aesenc          $rndkeyx,$inout6
1324         aesenc          $rndkeyx,$inout7
1325         $movkey         `0x20+0x10*$i`-0x80($key),$rndkeyx
1326 ___
1327 }
1328 $code.=<<___;
1329          bswap          %r9d
1330         aesenc          $rndkey0,$inout0
1331         aesenc          $rndkey0,$inout1
1332         aesenc          $rndkey0,$inout2
1333          xor            $key0,%r9d
1334          movdqu         0x00($inp),$in0
1335         aesenc          $rndkey0,$inout3
1336          mov            %r9d,0x70+12(%rsp)
1337          cmp            \$11,$rounds
1338         aesenc          $rndkey0,$inout4
1339         aesenc          $rndkey0,$inout5
1340         aesenc          $rndkey0,$inout6
1341         aesenc          $rndkey0,$inout7
1342         $movkey         0xa0-0x80($key),$rndkey0
1343
1344         jb              .Lctr32_enc_done
1345
1346         aesenc          $rndkey1,$inout0
1347         aesenc          $rndkey1,$inout1
1348         aesenc          $rndkey1,$inout2
1349         aesenc          $rndkey1,$inout3
1350         aesenc          $rndkey1,$inout4
1351         aesenc          $rndkey1,$inout5
1352         aesenc          $rndkey1,$inout6
1353         aesenc          $rndkey1,$inout7
1354         $movkey         0xb0-0x80($key),$rndkey1
1355
1356         aesenc          $rndkey0,$inout0
1357         aesenc          $rndkey0,$inout1
1358         aesenc          $rndkey0,$inout2
1359         aesenc          $rndkey0,$inout3
1360         aesenc          $rndkey0,$inout4
1361         aesenc          $rndkey0,$inout5
1362         aesenc          $rndkey0,$inout6
1363         aesenc          $rndkey0,$inout7
1364         $movkey         0xc0-0x80($key),$rndkey0
1365         je              .Lctr32_enc_done
1366
1367         aesenc          $rndkey1,$inout0
1368         aesenc          $rndkey1,$inout1
1369         aesenc          $rndkey1,$inout2
1370         aesenc          $rndkey1,$inout3
1371         aesenc          $rndkey1,$inout4
1372         aesenc          $rndkey1,$inout5
1373         aesenc          $rndkey1,$inout6
1374         aesenc          $rndkey1,$inout7
1375         $movkey         0xd0-0x80($key),$rndkey1
1376
1377         aesenc          $rndkey0,$inout0
1378         aesenc          $rndkey0,$inout1
1379         aesenc          $rndkey0,$inout2
1380         aesenc          $rndkey0,$inout3
1381         aesenc          $rndkey0,$inout4
1382         aesenc          $rndkey0,$inout5
1383         aesenc          $rndkey0,$inout6
1384         aesenc          $rndkey0,$inout7
1385         $movkey         0xe0-0x80($key),$rndkey0
1386         jmp             .Lctr32_enc_done
1387
1388 .align  16
1389 .Lctr32_enc_done:
1390         movdqu          0x10($inp),$in1
1391         pxor            $rndkey0,$in0
1392         movdqu          0x20($inp),$in2
1393         pxor            $rndkey0,$in1
1394         movdqu          0x30($inp),$in3
1395         pxor            $rndkey0,$in2
1396         movdqu          0x40($inp),$in4
1397         pxor            $rndkey0,$in3
1398         movdqu          0x50($inp),$in5
1399         pxor            $rndkey0,$in4
1400         pxor            $rndkey0,$in5
1401         aesenc          $rndkey1,$inout0
1402         aesenc          $rndkey1,$inout1
1403         aesenc          $rndkey1,$inout2
1404         aesenc          $rndkey1,$inout3
1405         aesenc          $rndkey1,$inout4
1406         aesenc          $rndkey1,$inout5
1407         aesenc          $rndkey1,$inout6
1408         aesenc          $rndkey1,$inout7
1409         movdqu          0x60($inp),$rndkey1
1410         lea             0x80($inp),$inp
1411
1412         aesenclast      $in0,$inout0
1413         pxor            $rndkey0,$rndkey1
1414         movdqu          0x70-0x80($inp),$in0
1415         aesenclast      $in1,$inout1
1416         pxor            $rndkey0,$in0
1417         movdqa          0x00(%rsp),$in1         # load next counter block
1418         aesenclast      $in2,$inout2
1419         aesenclast      $in3,$inout3
1420         movdqa          0x10(%rsp),$in2
1421         movdqa          0x20(%rsp),$in3
1422         aesenclast      $in4,$inout4
1423         aesenclast      $in5,$inout5
1424         movdqa          0x30(%rsp),$in4
1425         movdqa          0x40(%rsp),$in5
1426         aesenclast      $rndkey1,$inout6
1427         movdqa          0x50(%rsp),$rndkey0
1428         $movkey         0x10-0x80($key),$rndkey1
1429         aesenclast      $in0,$inout7
1430
1431         movups          $inout0,($out)          # store output
1432         movdqa          $in1,$inout0
1433         movups          $inout1,0x10($out)
1434         movdqa          $in2,$inout1
1435         movups          $inout2,0x20($out)
1436         movdqa          $in3,$inout2
1437         movups          $inout3,0x30($out)
1438         movdqa          $in4,$inout3
1439         movups          $inout4,0x40($out)
1440         movdqa          $in5,$inout4
1441         movups          $inout5,0x50($out)
1442         movdqa          $rndkey0,$inout5
1443         movups          $inout6,0x60($out)
1444         movups          $inout7,0x70($out)
1445         lea             0x80($out),$out
1446
1447         sub     \$8,$len
1448         jnc     .Lctr32_loop8
1449
1450         add     \$8,$len
1451         jz      .Lctr32_done
1452         lea     -0x80($key),$key
1453
1454 .Lctr32_tail:
1455         lea     16($key),$key
1456         cmp     \$4,$len
1457         jb      .Lctr32_loop3
1458         je      .Lctr32_loop4
1459
1460         shl             \$4,$rounds
1461         movdqa          0x60(%rsp),$inout6
1462         pxor            $inout7,$inout7
1463
1464         $movkey         16($key),$rndkey0
1465         aesenc          $rndkey1,$inout0
1466         aesenc          $rndkey1,$inout1
1467         lea             32-16($key,$rounds),$key
1468         neg             %rax
1469         aesenc          $rndkey1,$inout2
1470         add             \$16,%rax
1471          movups         ($inp),$in0
1472         aesenc          $rndkey1,$inout3
1473         aesenc          $rndkey1,$inout4
1474          movups         0x10($inp),$in1
1475          movups         0x20($inp),$in2
1476         aesenc          $rndkey1,$inout5
1477         aesenc          $rndkey1,$inout6
1478
1479         call            .Lenc_loop8_enter
1480
1481         movdqu  0x30($inp),$in3
1482         pxor    $in0,$inout0
1483         movdqu  0x40($inp),$in0
1484         pxor    $in1,$inout1
1485         movdqu  $inout0,($out)
1486         pxor    $in2,$inout2
1487         movdqu  $inout1,0x10($out)
1488         pxor    $in3,$inout3
1489         movdqu  $inout2,0x20($out)
1490         pxor    $in0,$inout4
1491         movdqu  $inout3,0x30($out)
1492         movdqu  $inout4,0x40($out)
1493         cmp     \$6,$len
1494         jb      .Lctr32_done
1495
1496         movups  0x50($inp),$in1
1497         xorps   $in1,$inout5
1498         movups  $inout5,0x50($out)
1499         je      .Lctr32_done
1500
1501         movups  0x60($inp),$in2
1502         xorps   $in2,$inout6
1503         movups  $inout6,0x60($out)
1504         jmp     .Lctr32_done
1505
1506 .align  32
1507 .Lctr32_loop4:
1508         aesenc          $rndkey1,$inout0
1509         lea             16($key),$key
1510         dec             $rounds
1511         aesenc          $rndkey1,$inout1
1512         aesenc          $rndkey1,$inout2
1513         aesenc          $rndkey1,$inout3
1514         $movkey         ($key),$rndkey1
1515         jnz             .Lctr32_loop4
1516         aesenclast      $rndkey1,$inout0
1517         aesenclast      $rndkey1,$inout1
1518          movups         ($inp),$in0
1519          movups         0x10($inp),$in1
1520         aesenclast      $rndkey1,$inout2
1521         aesenclast      $rndkey1,$inout3
1522          movups         0x20($inp),$in2
1523          movups         0x30($inp),$in3
1524
1525         xorps   $in0,$inout0
1526         movups  $inout0,($out)
1527         xorps   $in1,$inout1
1528         movups  $inout1,0x10($out)
1529         pxor    $in2,$inout2
1530         movdqu  $inout2,0x20($out)
1531         pxor    $in3,$inout3
1532         movdqu  $inout3,0x30($out)
1533         jmp     .Lctr32_done
1534
1535 .align  32
1536 .Lctr32_loop3:
1537         aesenc          $rndkey1,$inout0
1538         lea             16($key),$key
1539         dec             $rounds
1540         aesenc          $rndkey1,$inout1
1541         aesenc          $rndkey1,$inout2
1542         $movkey         ($key),$rndkey1
1543         jnz             .Lctr32_loop3
1544         aesenclast      $rndkey1,$inout0
1545         aesenclast      $rndkey1,$inout1
1546         aesenclast      $rndkey1,$inout2
1547
1548         movups  ($inp),$in0
1549         xorps   $in0,$inout0
1550         movups  $inout0,($out)
1551         cmp     \$2,$len
1552         jb      .Lctr32_done
1553
1554         movups  0x10($inp),$in1
1555         xorps   $in1,$inout1
1556         movups  $inout1,0x10($out)
1557         je      .Lctr32_done
1558
1559         movups  0x20($inp),$in2
1560         xorps   $in2,$inout2
1561         movups  $inout2,0x20($out)
1562         jmp     .Lctr32_done
1563
1564 .align  16
1565 .Lctr32_one_shortcut:
1566         movups  ($ivp),$inout0
1567         movups  ($inp),$in0
1568         mov     240($key),$rounds               # key->rounds
1569 ___
1570         &aesni_generate1("enc",$key,$rounds);
1571 $code.=<<___;
1572         xorps   $in0,$inout0
1573         movups  $inout0,($out)
1574         jmp     .Lctr32_done
1575
1576 .align  16
1577 .Lctr32_done:
1578 ___
1579 $code.=<<___ if ($win64);
1580         movaps  -0xa0(%rbp),%xmm6
1581         movaps  -0x90(%rbp),%xmm7
1582         movaps  -0x80(%rbp),%xmm8
1583         movaps  -0x70(%rbp),%xmm9
1584         movaps  -0x60(%rbp),%xmm10
1585         movaps  -0x50(%rbp),%xmm11
1586         movaps  -0x40(%rbp),%xmm12
1587         movaps  -0x30(%rbp),%xmm13
1588         movaps  -0x20(%rbp),%xmm14
1589         movaps  -0x10(%rbp),%xmm15
1590 ___
1591 $code.=<<___;
1592         lea     (%rbp),%rsp
1593         pop     %rbp
1594 .Lctr32_epilogue:
1595         ret
1596 .size   aesni_ctr32_encrypt_blocks,.-aesni_ctr32_encrypt_blocks
1597 ___
1598 }
1599 \f
1600 ######################################################################
1601 # void aesni_xts_[en|de]crypt(const char *inp,char *out,size_t len,
1602 #       const AES_KEY *key1, const AES_KEY *key2
1603 #       const unsigned char iv[16]);
1604 #
1605 {
1606 my @tweak=map("%xmm$_",(10..15));
1607 my ($twmask,$twres,$twtmp)=("%xmm8","%xmm9",@tweak[4]);
1608 my ($key2,$ivp,$len_)=("%r8","%r9","%r9");
1609 my $frame_size = 0x70 + ($win64?160:0);
1610
1611 $code.=<<___;
1612 .globl  aesni_xts_encrypt
1613 .type   aesni_xts_encrypt,\@function,6
1614 .align  16
1615 aesni_xts_encrypt:
1616         lea     (%rsp),%rax
1617         push    %rbp
1618         sub     \$$frame_size,%rsp
1619         and     \$-16,%rsp      # Linux kernel stack can be incorrectly seeded
1620 ___
1621 $code.=<<___ if ($win64);
1622         movaps  %xmm6,-0xa8(%rax)
1623         movaps  %xmm7,-0x98(%rax)
1624         movaps  %xmm8,-0x88(%rax)
1625         movaps  %xmm9,-0x78(%rax)
1626         movaps  %xmm10,-0x68(%rax)
1627         movaps  %xmm11,-0x58(%rax)
1628         movaps  %xmm12,-0x48(%rax)
1629         movaps  %xmm13,-0x38(%rax)
1630         movaps  %xmm14,-0x28(%rax)
1631         movaps  %xmm15,-0x18(%rax)
1632 .Lxts_enc_body:
1633 ___
1634 $code.=<<___;
1635         lea     -8(%rax),%rbp
1636         movups  ($ivp),$inout0                  # load clear-text tweak
1637         mov     240(%r8),$rounds                # key2->rounds
1638         mov     240($key),$rnds_                # key1->rounds
1639 ___
1640         # generate the tweak
1641         &aesni_generate1("enc",$key2,$rounds,$inout0);
1642 $code.=<<___;
1643         $movkey ($key),$rndkey0                 # zero round key
1644         mov     $key,$key_                      # backup $key
1645         mov     $rnds_,$rounds                  # backup $rounds
1646         shl     \$4,$rnds_
1647         mov     $len,$len_                      # backup $len
1648         and     \$-16,$len
1649
1650         $movkey 16($key,$rnds_),$rndkey1        # last round key
1651
1652         movdqa  .Lxts_magic(%rip),$twmask
1653         movdqa  $inout0,@tweak[5]
1654         pshufd  \$0x5f,$inout0,$twres
1655         pxor    $rndkey0,$rndkey1
1656 ___
1657     # alternative tweak calculation algorithm is based on suggestions
1658     # by Shay Gueron. psrad doesn't conflict with AES-NI instructions
1659     # and should help in the future...
1660     for ($i=0;$i<4;$i++) {
1661     $code.=<<___;
1662         movdqa  $twres,$twtmp
1663         paddd   $twres,$twres
1664         movdqa  @tweak[5],@tweak[$i]
1665         psrad   \$31,$twtmp                     # broadcast upper bits
1666         paddq   @tweak[5],@tweak[5]
1667         pand    $twmask,$twtmp
1668         pxor    $rndkey0,@tweak[$i]
1669         pxor    $twtmp,@tweak[5]
1670 ___
1671     }
1672 $code.=<<___;
1673         movdqa  @tweak[5],@tweak[4]
1674         psrad   \$31,$twres
1675         paddq   @tweak[5],@tweak[5]
1676         pand    $twmask,$twres
1677         pxor    $rndkey0,@tweak[4]
1678         pxor    $twres,@tweak[5]
1679         movaps  $rndkey1,0x60(%rsp)             # save round[0]^round[last]
1680
1681         sub     \$16*6,$len
1682         jc      .Lxts_enc_short
1683
1684         mov     \$16+96,$rounds
1685         lea     32($key_,$rnds_),$key           # end of key schedule
1686         sub     %r10,%rax                       # twisted $rounds
1687         $movkey 16($key_),$rndkey1
1688         mov     %rax,%r10                       # backup twisted $rounds
1689         lea     .Lxts_magic(%rip),%r8
1690         jmp     .Lxts_enc_grandloop
1691
1692 .align  32
1693 .Lxts_enc_grandloop:
1694         movdqu  `16*0`($inp),$inout0            # load input
1695         movdqa  $rndkey0,$twmask
1696         movdqu  `16*1`($inp),$inout1
1697         pxor    @tweak[0],$inout0
1698         movdqu  `16*2`($inp),$inout2
1699         pxor    @tweak[1],$inout1
1700          aesenc         $rndkey1,$inout0
1701         movdqu  `16*3`($inp),$inout3
1702         pxor    @tweak[2],$inout2
1703          aesenc         $rndkey1,$inout1
1704         movdqu  `16*4`($inp),$inout4
1705         pxor    @tweak[3],$inout3
1706          aesenc         $rndkey1,$inout2
1707         movdqu  `16*5`($inp),$inout5
1708         pxor    @tweak[5],$twmask               # round[0]^=tweak[5]
1709          movdqa 0x60(%rsp),$twres               # load round[0]^round[last]
1710         pxor    @tweak[4],$inout4
1711          aesenc         $rndkey1,$inout3
1712         $movkey 32($key_),$rndkey0
1713         lea     `16*6`($inp),$inp
1714         pxor    $twmask,$inout5
1715
1716          pxor   $twres,@tweak[0]
1717         aesenc          $rndkey1,$inout4
1718          pxor   $twres,@tweak[1]
1719          movdqa @tweak[0],`16*0`(%rsp)          # put aside tweaks^last round key
1720         aesenc          $rndkey1,$inout5
1721         $movkey         48($key_),$rndkey1
1722          pxor   $twres,@tweak[2]
1723
1724         aesenc          $rndkey0,$inout0
1725          pxor   $twres,@tweak[3]
1726          movdqa @tweak[1],`16*1`(%rsp)
1727         aesenc          $rndkey0,$inout1
1728          pxor   $twres,@tweak[4]
1729          movdqa @tweak[2],`16*2`(%rsp)
1730         aesenc          $rndkey0,$inout2
1731         aesenc          $rndkey0,$inout3
1732          pxor   $twres,$twmask
1733          movdqa @tweak[4],`16*4`(%rsp)
1734         aesenc          $rndkey0,$inout4
1735         aesenc          $rndkey0,$inout5
1736         $movkey         64($key_),$rndkey0
1737          movdqa $twmask,`16*5`(%rsp)
1738         pshufd  \$0x5f,@tweak[5],$twres
1739         jmp     .Lxts_enc_loop6
1740 .align  32
1741 .Lxts_enc_loop6:
1742         aesenc          $rndkey1,$inout0
1743         aesenc          $rndkey1,$inout1
1744         aesenc          $rndkey1,$inout2
1745         aesenc          $rndkey1,$inout3
1746         aesenc          $rndkey1,$inout4
1747         aesenc          $rndkey1,$inout5
1748         $movkey         -64($key,%rax),$rndkey1
1749         add             \$32,%rax
1750
1751         aesenc          $rndkey0,$inout0
1752         aesenc          $rndkey0,$inout1
1753         aesenc          $rndkey0,$inout2
1754         aesenc          $rndkey0,$inout3
1755         aesenc          $rndkey0,$inout4
1756         aesenc          $rndkey0,$inout5
1757         $movkey         -80($key,%rax),$rndkey0
1758         jnz             .Lxts_enc_loop6
1759
1760         movdqa  (%r8),$twmask
1761         movdqa  $twres,$twtmp
1762         paddd   $twres,$twres
1763          aesenc         $rndkey1,$inout0
1764         paddq   @tweak[5],@tweak[5]
1765         psrad   \$31,$twtmp
1766          aesenc         $rndkey1,$inout1
1767         pand    $twmask,$twtmp
1768         $movkey ($key_),@tweak[0]               # load round[0]
1769          aesenc         $rndkey1,$inout2
1770          aesenc         $rndkey1,$inout3
1771          aesenc         $rndkey1,$inout4
1772         pxor    $twtmp,@tweak[5]
1773         movaps  @tweak[0],@tweak[1]             # copy round[0]
1774          aesenc         $rndkey1,$inout5
1775          $movkey        -64($key),$rndkey1
1776
1777         movdqa  $twres,$twtmp
1778          aesenc         $rndkey0,$inout0
1779         paddd   $twres,$twres
1780         pxor    @tweak[5],@tweak[0]
1781          aesenc         $rndkey0,$inout1
1782         psrad   \$31,$twtmp
1783         paddq   @tweak[5],@tweak[5]
1784          aesenc         $rndkey0,$inout2
1785          aesenc         $rndkey0,$inout3
1786         pand    $twmask,$twtmp
1787         movaps  @tweak[1],@tweak[2]
1788          aesenc         $rndkey0,$inout4
1789         pxor    $twtmp,@tweak[5]
1790         movdqa  $twres,$twtmp
1791          aesenc         $rndkey0,$inout5
1792          $movkey        -48($key),$rndkey0
1793
1794         paddd   $twres,$twres
1795          aesenc         $rndkey1,$inout0
1796         pxor    @tweak[5],@tweak[1]
1797         psrad   \$31,$twtmp
1798          aesenc         $rndkey1,$inout1
1799         paddq   @tweak[5],@tweak[5]
1800         pand    $twmask,$twtmp
1801          aesenc         $rndkey1,$inout2
1802          aesenc         $rndkey1,$inout3
1803          movdqa @tweak[3],`16*3`(%rsp)
1804         pxor    $twtmp,@tweak[5]
1805          aesenc         $rndkey1,$inout4
1806         movaps  @tweak[2],@tweak[3]
1807         movdqa  $twres,$twtmp
1808          aesenc         $rndkey1,$inout5
1809          $movkey        -32($key),$rndkey1
1810
1811         paddd   $twres,$twres
1812          aesenc         $rndkey0,$inout0
1813         pxor    @tweak[5],@tweak[2]
1814         psrad   \$31,$twtmp
1815          aesenc         $rndkey0,$inout1
1816         paddq   @tweak[5],@tweak[5]
1817         pand    $twmask,$twtmp
1818          aesenc         $rndkey0,$inout2
1819          aesenc         $rndkey0,$inout3
1820          aesenc         $rndkey0,$inout4
1821         pxor    $twtmp,@tweak[5]
1822         movaps  @tweak[3],@tweak[4]
1823          aesenc         $rndkey0,$inout5
1824
1825         movdqa  $twres,$rndkey0
1826         paddd   $twres,$twres
1827          aesenc         $rndkey1,$inout0
1828         pxor    @tweak[5],@tweak[3]
1829         psrad   \$31,$rndkey0
1830          aesenc         $rndkey1,$inout1
1831         paddq   @tweak[5],@tweak[5]
1832         pand    $twmask,$rndkey0
1833          aesenc         $rndkey1,$inout2
1834          aesenc         $rndkey1,$inout3
1835         pxor    $rndkey0,@tweak[5]
1836         $movkey         ($key_),$rndkey0
1837          aesenc         $rndkey1,$inout4
1838          aesenc         $rndkey1,$inout5
1839         $movkey         16($key_),$rndkey1
1840
1841         pxor    @tweak[5],@tweak[4]
1842          aesenclast     `16*0`(%rsp),$inout0
1843         psrad   \$31,$twres
1844         paddq   @tweak[5],@tweak[5]
1845          aesenclast     `16*1`(%rsp),$inout1
1846          aesenclast     `16*2`(%rsp),$inout2
1847         pand    $twmask,$twres
1848         mov     %r10,%rax                       # restore $rounds
1849          aesenclast     `16*3`(%rsp),$inout3
1850          aesenclast     `16*4`(%rsp),$inout4
1851          aesenclast     `16*5`(%rsp),$inout5
1852         pxor    $twres,@tweak[5]
1853
1854         lea     `16*6`($out),$out
1855         movups  $inout0,`-16*6`($out)           # write output
1856         movups  $inout1,`-16*5`($out)
1857         movups  $inout2,`-16*4`($out)
1858         movups  $inout3,`-16*3`($out)
1859         movups  $inout4,`-16*2`($out)
1860         movups  $inout5,`-16*1`($out)
1861         sub     \$16*6,$len
1862         jnc     .Lxts_enc_grandloop
1863
1864         mov     \$16+96,$rounds
1865         sub     $rnds_,$rounds
1866         mov     $key_,$key                      # restore $key
1867         shr     \$4,$rounds                     # restore original value
1868
1869 .Lxts_enc_short:
1870         mov     $rounds,$rnds_                  # backup $rounds
1871         pxor    $rndkey0,@tweak[0]
1872         add     \$16*6,$len
1873         jz      .Lxts_enc_done
1874
1875         pxor    $rndkey0,@tweak[1]
1876         cmp     \$0x20,$len
1877         jb      .Lxts_enc_one
1878         pxor    $rndkey0,@tweak[2]
1879         je      .Lxts_enc_two
1880
1881         pxor    $rndkey0,@tweak[3]
1882         cmp     \$0x40,$len
1883         jb      .Lxts_enc_three
1884         pxor    $rndkey0,@tweak[4]
1885         je      .Lxts_enc_four
1886
1887         movdqu  ($inp),$inout0
1888         movdqu  16*1($inp),$inout1
1889         movdqu  16*2($inp),$inout2
1890         pxor    @tweak[0],$inout0
1891         movdqu  16*3($inp),$inout3
1892         pxor    @tweak[1],$inout1
1893         movdqu  16*4($inp),$inout4
1894         lea     16*5($inp),$inp
1895         pxor    @tweak[2],$inout2
1896         pxor    @tweak[3],$inout3
1897         pxor    @tweak[4],$inout4
1898
1899         call    _aesni_encrypt6
1900
1901         xorps   @tweak[0],$inout0
1902         movdqa  @tweak[5],@tweak[0]
1903         xorps   @tweak[1],$inout1
1904         xorps   @tweak[2],$inout2
1905         movdqu  $inout0,($out)
1906         xorps   @tweak[3],$inout3
1907         movdqu  $inout1,16*1($out)
1908         xorps   @tweak[4],$inout4
1909         movdqu  $inout2,16*2($out)
1910         movdqu  $inout3,16*3($out)
1911         movdqu  $inout4,16*4($out)
1912         lea     16*5($out),$out
1913         jmp     .Lxts_enc_done
1914
1915 .align  16
1916 .Lxts_enc_one:
1917         movups  ($inp),$inout0
1918         lea     16*1($inp),$inp
1919         xorps   @tweak[0],$inout0
1920 ___
1921         &aesni_generate1("enc",$key,$rounds);
1922 $code.=<<___;
1923         xorps   @tweak[0],$inout0
1924         movdqa  @tweak[1],@tweak[0]
1925         movups  $inout0,($out)
1926         lea     16*1($out),$out
1927         jmp     .Lxts_enc_done
1928
1929 .align  16
1930 .Lxts_enc_two:
1931         movups  ($inp),$inout0
1932         movups  16($inp),$inout1
1933         lea     32($inp),$inp
1934         xorps   @tweak[0],$inout0
1935         xorps   @tweak[1],$inout1
1936
1937         call    _aesni_encrypt2
1938
1939         xorps   @tweak[0],$inout0
1940         movdqa  @tweak[2],@tweak[0]
1941         xorps   @tweak[1],$inout1
1942         movups  $inout0,($out)
1943         movups  $inout1,16*1($out)
1944         lea     16*2($out),$out
1945         jmp     .Lxts_enc_done
1946
1947 .align  16
1948 .Lxts_enc_three:
1949         movups  ($inp),$inout0
1950         movups  16*1($inp),$inout1
1951         movups  16*2($inp),$inout2
1952         lea     16*3($inp),$inp
1953         xorps   @tweak[0],$inout0
1954         xorps   @tweak[1],$inout1
1955         xorps   @tweak[2],$inout2
1956
1957         call    _aesni_encrypt3
1958
1959         xorps   @tweak[0],$inout0
1960         movdqa  @tweak[3],@tweak[0]
1961         xorps   @tweak[1],$inout1
1962         xorps   @tweak[2],$inout2
1963         movups  $inout0,($out)
1964         movups  $inout1,16*1($out)
1965         movups  $inout2,16*2($out)
1966         lea     16*3($out),$out
1967         jmp     .Lxts_enc_done
1968
1969 .align  16
1970 .Lxts_enc_four:
1971         movups  ($inp),$inout0
1972         movups  16*1($inp),$inout1
1973         movups  16*2($inp),$inout2
1974         xorps   @tweak[0],$inout0
1975         movups  16*3($inp),$inout3
1976         lea     16*4($inp),$inp
1977         xorps   @tweak[1],$inout1
1978         xorps   @tweak[2],$inout2
1979         xorps   @tweak[3],$inout3
1980
1981         call    _aesni_encrypt4
1982
1983         pxor    @tweak[0],$inout0
1984         movdqa  @tweak[4],@tweak[0]
1985         pxor    @tweak[1],$inout1
1986         pxor    @tweak[2],$inout2
1987         movdqu  $inout0,($out)
1988         pxor    @tweak[3],$inout3
1989         movdqu  $inout1,16*1($out)
1990         movdqu  $inout2,16*2($out)
1991         movdqu  $inout3,16*3($out)
1992         lea     16*4($out),$out
1993         jmp     .Lxts_enc_done
1994
1995 .align  16
1996 .Lxts_enc_done:
1997         and     \$15,$len_
1998         jz      .Lxts_enc_ret
1999         mov     $len_,$len
2000
2001 .Lxts_enc_steal:
2002         movzb   ($inp),%eax                     # borrow $rounds ...
2003         movzb   -16($out),%ecx                  # ... and $key
2004         lea     1($inp),$inp
2005         mov     %al,-16($out)
2006         mov     %cl,0($out)
2007         lea     1($out),$out
2008         sub     \$1,$len
2009         jnz     .Lxts_enc_steal
2010
2011         sub     $len_,$out                      # rewind $out
2012         mov     $key_,$key                      # restore $key
2013         mov     $rnds_,$rounds                  # restore $rounds
2014
2015         movups  -16($out),$inout0
2016         xorps   @tweak[0],$inout0
2017 ___
2018         &aesni_generate1("enc",$key,$rounds);
2019 $code.=<<___;
2020         xorps   @tweak[0],$inout0
2021         movups  $inout0,-16($out)
2022
2023 .Lxts_enc_ret:
2024 ___
2025 $code.=<<___ if ($win64);
2026         movaps  -0xa0(%rbp),%xmm6
2027         movaps  -0x90(%rbp),%xmm7
2028         movaps  -0x80(%rbp),%xmm8
2029         movaps  -0x70(%rbp),%xmm9
2030         movaps  -0x60(%rbp),%xmm10
2031         movaps  -0x50(%rbp),%xmm11
2032         movaps  -0x40(%rbp),%xmm12
2033         movaps  -0x30(%rbp),%xmm13
2034         movaps  -0x20(%rbp),%xmm14
2035         movaps  -0x10(%rbp),%xmm15
2036 ___
2037 $code.=<<___;
2038         lea     (%rbp),%rsp
2039         pop     %rbp
2040 .Lxts_enc_epilogue:
2041         ret
2042 .size   aesni_xts_encrypt,.-aesni_xts_encrypt
2043 ___
2044
2045 $code.=<<___;
2046 .globl  aesni_xts_decrypt
2047 .type   aesni_xts_decrypt,\@function,6
2048 .align  16
2049 aesni_xts_decrypt:
2050         lea     (%rsp),%rax
2051         push    %rbp
2052         sub     \$$frame_size,%rsp
2053         and     \$-16,%rsp      # Linux kernel stack can be incorrectly seeded
2054 ___
2055 $code.=<<___ if ($win64);
2056         movaps  %xmm6,-0xa8(%rax)
2057         movaps  %xmm7,-0x98(%rax)
2058         movaps  %xmm8,-0x88(%rax)
2059         movaps  %xmm9,-0x78(%rax)
2060         movaps  %xmm10,-0x68(%rax)
2061         movaps  %xmm11,-0x58(%rax)
2062         movaps  %xmm12,-0x48(%rax)
2063         movaps  %xmm13,-0x38(%rax)
2064         movaps  %xmm14,-0x28(%rax)
2065         movaps  %xmm15,-0x18(%rax)
2066 .Lxts_dec_body:
2067 ___
2068 $code.=<<___;
2069         lea     -8(%rax),%rbp
2070         movups  ($ivp),$inout0                  # load clear-text tweak
2071         mov     240($key2),$rounds              # key2->rounds
2072         mov     240($key),$rnds_                # key1->rounds
2073 ___
2074         # generate the tweak
2075         &aesni_generate1("enc",$key2,$rounds,$inout0);
2076 $code.=<<___;
2077         xor     %eax,%eax                       # if ($len%16) len-=16;
2078         test    \$15,$len
2079         setnz   %al
2080         shl     \$4,%rax
2081         sub     %rax,$len
2082
2083         $movkey ($key),$rndkey0                 # zero round key
2084         mov     $key,$key_                      # backup $key
2085         mov     $rnds_,$rounds                  # backup $rounds
2086         shl     \$4,$rnds_
2087         mov     $len,$len_                      # backup $len
2088         and     \$-16,$len
2089
2090         $movkey 16($key,$rnds_),$rndkey1        # last round key
2091
2092         movdqa  .Lxts_magic(%rip),$twmask
2093         movdqa  $inout0,@tweak[5]
2094         pshufd  \$0x5f,$inout0,$twres
2095         pxor    $rndkey0,$rndkey1
2096 ___
2097     for ($i=0;$i<4;$i++) {
2098     $code.=<<___;
2099         movdqa  $twres,$twtmp
2100         paddd   $twres,$twres
2101         movdqa  @tweak[5],@tweak[$i]
2102         psrad   \$31,$twtmp                     # broadcast upper bits
2103         paddq   @tweak[5],@tweak[5]
2104         pand    $twmask,$twtmp
2105         pxor    $rndkey0,@tweak[$i]
2106         pxor    $twtmp,@tweak[5]
2107 ___
2108     }
2109 $code.=<<___;
2110         movdqa  @tweak[5],@tweak[4]
2111         psrad   \$31,$twres
2112         paddq   @tweak[5],@tweak[5]
2113         pand    $twmask,$twres
2114         pxor    $rndkey0,@tweak[4]
2115         pxor    $twres,@tweak[5]
2116         movaps  $rndkey1,0x60(%rsp)             # save round[0]^round[last]
2117
2118         sub     \$16*6,$len
2119         jc      .Lxts_dec_short
2120
2121         mov     \$16+96,$rounds
2122         lea     32($key_,$rnds_),$key           # end of key schedule
2123         sub     %r10,%rax                       # twisted $rounds
2124         $movkey 16($key_),$rndkey1
2125         mov     %rax,%r10                       # backup twisted $rounds
2126         lea     .Lxts_magic(%rip),%r8
2127         jmp     .Lxts_dec_grandloop
2128
2129 .align  32
2130 .Lxts_dec_grandloop:
2131         movdqu  `16*0`($inp),$inout0            # load input
2132         movdqa  $rndkey0,$twmask
2133         movdqu  `16*1`($inp),$inout1
2134         pxor    @tweak[0],$inout0
2135         movdqu  `16*2`($inp),$inout2
2136         pxor    @tweak[1],$inout1
2137          aesdec         $rndkey1,$inout0
2138         movdqu  `16*3`($inp),$inout3
2139         pxor    @tweak[2],$inout2
2140          aesdec         $rndkey1,$inout1
2141         movdqu  `16*4`($inp),$inout4
2142         pxor    @tweak[3],$inout3
2143          aesdec         $rndkey1,$inout2
2144         movdqu  `16*5`($inp),$inout5
2145         pxor    @tweak[5],$twmask               # round[0]^=tweak[5]
2146          movdqa 0x60(%rsp),$twres               # load round[0]^round[last]
2147         pxor    @tweak[4],$inout4
2148          aesdec         $rndkey1,$inout3
2149         $movkey 32($key_),$rndkey0
2150         lea     `16*6`($inp),$inp
2151         pxor    $twmask,$inout5
2152
2153          pxor   $twres,@tweak[0]
2154         aesdec          $rndkey1,$inout4
2155          pxor   $twres,@tweak[1]
2156          movdqa @tweak[0],`16*0`(%rsp)          # put aside tweaks^last round key
2157         aesdec          $rndkey1,$inout5
2158         $movkey         48($key_),$rndkey1
2159          pxor   $twres,@tweak[2]
2160
2161         aesdec          $rndkey0,$inout0
2162          pxor   $twres,@tweak[3]
2163          movdqa @tweak[1],`16*1`(%rsp)
2164         aesdec          $rndkey0,$inout1
2165          pxor   $twres,@tweak[4]
2166          movdqa @tweak[2],`16*2`(%rsp)
2167         aesdec          $rndkey0,$inout2
2168         aesdec          $rndkey0,$inout3
2169          pxor   $twres,$twmask
2170          movdqa @tweak[4],`16*4`(%rsp)
2171         aesdec          $rndkey0,$inout4
2172         aesdec          $rndkey0,$inout5
2173         $movkey         64($key_),$rndkey0
2174          movdqa $twmask,`16*5`(%rsp)
2175         pshufd  \$0x5f,@tweak[5],$twres
2176         jmp     .Lxts_dec_loop6
2177 .align  32
2178 .Lxts_dec_loop6:
2179         aesdec          $rndkey1,$inout0
2180         aesdec          $rndkey1,$inout1
2181         aesdec          $rndkey1,$inout2
2182         aesdec          $rndkey1,$inout3
2183         aesdec          $rndkey1,$inout4
2184         aesdec          $rndkey1,$inout5
2185         $movkey         -64($key,%rax),$rndkey1
2186         add             \$32,%rax
2187
2188         aesdec          $rndkey0,$inout0
2189         aesdec          $rndkey0,$inout1
2190         aesdec          $rndkey0,$inout2
2191         aesdec          $rndkey0,$inout3
2192         aesdec          $rndkey0,$inout4
2193         aesdec          $rndkey0,$inout5
2194         $movkey         -80($key,%rax),$rndkey0
2195         jnz             .Lxts_dec_loop6
2196
2197         movdqa  (%r8),$twmask
2198         movdqa  $twres,$twtmp
2199         paddd   $twres,$twres
2200          aesdec         $rndkey1,$inout0
2201         paddq   @tweak[5],@tweak[5]
2202         psrad   \$31,$twtmp
2203          aesdec         $rndkey1,$inout1
2204         pand    $twmask,$twtmp
2205         $movkey ($key_),@tweak[0]               # load round[0]
2206          aesdec         $rndkey1,$inout2
2207          aesdec         $rndkey1,$inout3
2208          aesdec         $rndkey1,$inout4
2209         pxor    $twtmp,@tweak[5]
2210         movaps  @tweak[0],@tweak[1]             # copy round[0]
2211          aesdec         $rndkey1,$inout5
2212          $movkey        -64($key),$rndkey1
2213
2214         movdqa  $twres,$twtmp
2215          aesdec         $rndkey0,$inout0
2216         paddd   $twres,$twres
2217         pxor    @tweak[5],@tweak[0]
2218          aesdec         $rndkey0,$inout1
2219         psrad   \$31,$twtmp
2220         paddq   @tweak[5],@tweak[5]
2221          aesdec         $rndkey0,$inout2
2222          aesdec         $rndkey0,$inout3
2223         pand    $twmask,$twtmp
2224         movaps  @tweak[1],@tweak[2]
2225          aesdec         $rndkey0,$inout4
2226         pxor    $twtmp,@tweak[5]
2227         movdqa  $twres,$twtmp
2228          aesdec         $rndkey0,$inout5
2229          $movkey        -48($key),$rndkey0
2230
2231         paddd   $twres,$twres
2232          aesdec         $rndkey1,$inout0
2233         pxor    @tweak[5],@tweak[1]
2234         psrad   \$31,$twtmp
2235          aesdec         $rndkey1,$inout1
2236         paddq   @tweak[5],@tweak[5]
2237         pand    $twmask,$twtmp
2238          aesdec         $rndkey1,$inout2
2239          aesdec         $rndkey1,$inout3
2240          movdqa @tweak[3],`16*3`(%rsp)
2241         pxor    $twtmp,@tweak[5]
2242          aesdec         $rndkey1,$inout4
2243         movaps  @tweak[2],@tweak[3]
2244         movdqa  $twres,$twtmp
2245          aesdec         $rndkey1,$inout5
2246          $movkey        -32($key),$rndkey1
2247
2248         paddd   $twres,$twres
2249          aesdec         $rndkey0,$inout0
2250         pxor    @tweak[5],@tweak[2]
2251         psrad   \$31,$twtmp
2252          aesdec         $rndkey0,$inout1
2253         paddq   @tweak[5],@tweak[5]
2254         pand    $twmask,$twtmp
2255          aesdec         $rndkey0,$inout2
2256          aesdec         $rndkey0,$inout3
2257          aesdec         $rndkey0,$inout4
2258         pxor    $twtmp,@tweak[5]
2259         movaps  @tweak[3],@tweak[4]
2260          aesdec         $rndkey0,$inout5
2261
2262         movdqa  $twres,$rndkey0
2263         paddd   $twres,$twres
2264          aesdec         $rndkey1,$inout0
2265         pxor    @tweak[5],@tweak[3]
2266         psrad   \$31,$rndkey0
2267          aesdec         $rndkey1,$inout1
2268         paddq   @tweak[5],@tweak[5]
2269         pand    $twmask,$rndkey0
2270          aesdec         $rndkey1,$inout2
2271          aesdec         $rndkey1,$inout3
2272         pxor    $rndkey0,@tweak[5]
2273         $movkey         ($key_),$rndkey0
2274          aesdec         $rndkey1,$inout4
2275          aesdec         $rndkey1,$inout5
2276         $movkey         16($key_),$rndkey1
2277
2278         pxor    @tweak[5],@tweak[4]
2279          aesdeclast     `16*0`(%rsp),$inout0
2280         psrad   \$31,$twres
2281         paddq   @tweak[5],@tweak[5]
2282          aesdeclast     `16*1`(%rsp),$inout1
2283          aesdeclast     `16*2`(%rsp),$inout2
2284         pand    $twmask,$twres
2285         mov     %r10,%rax                       # restore $rounds
2286          aesdeclast     `16*3`(%rsp),$inout3
2287          aesdeclast     `16*4`(%rsp),$inout4
2288          aesdeclast     `16*5`(%rsp),$inout5
2289         pxor    $twres,@tweak[5]
2290
2291         lea     `16*6`($out),$out
2292         movups  $inout0,`-16*6`($out)           # write output
2293         movups  $inout1,`-16*5`($out)
2294         movups  $inout2,`-16*4`($out)
2295         movups  $inout3,`-16*3`($out)
2296         movups  $inout4,`-16*2`($out)
2297         movups  $inout5,`-16*1`($out)
2298         sub     \$16*6,$len
2299         jnc     .Lxts_dec_grandloop
2300
2301         mov     \$16+96,$rounds
2302         sub     $rnds_,$rounds
2303         mov     $key_,$key                      # restore $key
2304         shr     \$4,$rounds                     # restore original value
2305
2306 .Lxts_dec_short:
2307         mov     $rounds,$rnds_                  # backup $rounds
2308         pxor    $rndkey0,@tweak[0]
2309         pxor    $rndkey0,@tweak[1]
2310         add     \$16*6,$len
2311         jz      .Lxts_dec_done
2312
2313         pxor    $rndkey0,@tweak[2]
2314         cmp     \$0x20,$len
2315         jb      .Lxts_dec_one
2316         pxor    $rndkey0,@tweak[3]
2317         je      .Lxts_dec_two
2318
2319         pxor    $rndkey0,@tweak[4]
2320         cmp     \$0x40,$len
2321         jb      .Lxts_dec_three
2322         je      .Lxts_dec_four
2323
2324         movdqu  ($inp),$inout0
2325         movdqu  16*1($inp),$inout1
2326         movdqu  16*2($inp),$inout2
2327         pxor    @tweak[0],$inout0
2328         movdqu  16*3($inp),$inout3
2329         pxor    @tweak[1],$inout1
2330         movdqu  16*4($inp),$inout4
2331         lea     16*5($inp),$inp
2332         pxor    @tweak[2],$inout2
2333         pxor    @tweak[3],$inout3
2334         pxor    @tweak[4],$inout4
2335
2336         call    _aesni_decrypt6
2337
2338         xorps   @tweak[0],$inout0
2339         xorps   @tweak[1],$inout1
2340         xorps   @tweak[2],$inout2
2341         movdqu  $inout0,($out)
2342         xorps   @tweak[3],$inout3
2343         movdqu  $inout1,16*1($out)
2344         xorps   @tweak[4],$inout4
2345         movdqu  $inout2,16*2($out)
2346          pxor           $twtmp,$twtmp
2347         movdqu  $inout3,16*3($out)
2348          pcmpgtd        @tweak[5],$twtmp
2349         movdqu  $inout4,16*4($out)
2350         lea     16*5($out),$out
2351          pshufd         \$0x13,$twtmp,@tweak[1] # $twres
2352         and     \$15,$len_
2353         jz      .Lxts_dec_ret
2354
2355         movdqa  @tweak[5],@tweak[0]
2356         paddq   @tweak[5],@tweak[5]             # psllq 1,$tweak
2357         pand    $twmask,@tweak[1]               # isolate carry and residue
2358         pxor    @tweak[5],@tweak[1]
2359         jmp     .Lxts_dec_done2
2360
2361 .align  16
2362 .Lxts_dec_one:
2363         movups  ($inp),$inout0
2364         lea     16*1($inp),$inp
2365         xorps   @tweak[0],$inout0
2366 ___
2367         &aesni_generate1("dec",$key,$rounds);
2368 $code.=<<___;
2369         xorps   @tweak[0],$inout0
2370         movdqa  @tweak[1],@tweak[0]
2371         movups  $inout0,($out)
2372         movdqa  @tweak[2],@tweak[1]
2373         lea     16*1($out),$out
2374         jmp     .Lxts_dec_done
2375
2376 .align  16
2377 .Lxts_dec_two:
2378         movups  ($inp),$inout0
2379         movups  16($inp),$inout1
2380         lea     32($inp),$inp
2381         xorps   @tweak[0],$inout0
2382         xorps   @tweak[1],$inout1
2383
2384         call    _aesni_decrypt2
2385
2386         xorps   @tweak[0],$inout0
2387         movdqa  @tweak[2],@tweak[0]
2388         xorps   @tweak[1],$inout1
2389         movdqa  @tweak[3],@tweak[1]
2390         movups  $inout0,($out)
2391         movups  $inout1,16*1($out)
2392         lea     16*2($out),$out
2393         jmp     .Lxts_dec_done
2394
2395 .align  16
2396 .Lxts_dec_three:
2397         movups  ($inp),$inout0
2398         movups  16*1($inp),$inout1
2399         movups  16*2($inp),$inout2
2400         lea     16*3($inp),$inp
2401         xorps   @tweak[0],$inout0
2402         xorps   @tweak[1],$inout1
2403         xorps   @tweak[2],$inout2
2404
2405         call    _aesni_decrypt3
2406
2407         xorps   @tweak[0],$inout0
2408         movdqa  @tweak[3],@tweak[0]
2409         xorps   @tweak[1],$inout1
2410         movdqa  @tweak[4],@tweak[1]
2411         xorps   @tweak[2],$inout2
2412         movups  $inout0,($out)
2413         movups  $inout1,16*1($out)
2414         movups  $inout2,16*2($out)
2415         lea     16*3($out),$out
2416         jmp     .Lxts_dec_done
2417
2418 .align  16
2419 .Lxts_dec_four:
2420         movups  ($inp),$inout0
2421         movups  16*1($inp),$inout1
2422         movups  16*2($inp),$inout2
2423         xorps   @tweak[0],$inout0
2424         movups  16*3($inp),$inout3
2425         lea     16*4($inp),$inp
2426         xorps   @tweak[1],$inout1
2427         xorps   @tweak[2],$inout2
2428         xorps   @tweak[3],$inout3
2429
2430         call    _aesni_decrypt4
2431
2432         pxor    @tweak[0],$inout0
2433         movdqa  @tweak[4],@tweak[0]
2434         pxor    @tweak[1],$inout1
2435         movdqa  @tweak[5],@tweak[1]
2436         pxor    @tweak[2],$inout2
2437         movdqu  $inout0,($out)
2438         pxor    @tweak[3],$inout3
2439         movdqu  $inout1,16*1($out)
2440         movdqu  $inout2,16*2($out)
2441         movdqu  $inout3,16*3($out)
2442         lea     16*4($out),$out
2443         jmp     .Lxts_dec_done
2444
2445 .align  16
2446 .Lxts_dec_done:
2447         and     \$15,$len_
2448         jz      .Lxts_dec_ret
2449 .Lxts_dec_done2:
2450         mov     $len_,$len
2451         mov     $key_,$key                      # restore $key
2452         mov     $rnds_,$rounds                  # restore $rounds
2453
2454         movups  ($inp),$inout0
2455         xorps   @tweak[1],$inout0
2456 ___
2457         &aesni_generate1("dec",$key,$rounds);
2458 $code.=<<___;
2459         xorps   @tweak[1],$inout0
2460         movups  $inout0,($out)
2461
2462 .Lxts_dec_steal:
2463         movzb   16($inp),%eax                   # borrow $rounds ...
2464         movzb   ($out),%ecx                     # ... and $key
2465         lea     1($inp),$inp
2466         mov     %al,($out)
2467         mov     %cl,16($out)
2468         lea     1($out),$out
2469         sub     \$1,$len
2470         jnz     .Lxts_dec_steal
2471
2472         sub     $len_,$out                      # rewind $out
2473         mov     $key_,$key                      # restore $key
2474         mov     $rnds_,$rounds                  # restore $rounds
2475
2476         movups  ($out),$inout0
2477         xorps   @tweak[0],$inout0
2478 ___
2479         &aesni_generate1("dec",$key,$rounds);
2480 $code.=<<___;
2481         xorps   @tweak[0],$inout0
2482         movups  $inout0,($out)
2483
2484 .Lxts_dec_ret:
2485 ___
2486 $code.=<<___ if ($win64);
2487         movaps  -0xa0(%rbp),%xmm6
2488         movaps  -0x90(%rbp),%xmm7
2489         movaps  -0x80(%rbp),%xmm8
2490         movaps  -0x70(%rbp),%xmm9
2491         movaps  -0x60(%rbp),%xmm10
2492         movaps  -0x50(%rbp),%xmm11
2493         movaps  -0x40(%rbp),%xmm12
2494         movaps  -0x30(%rbp),%xmm13
2495         movaps  -0x20(%rbp),%xmm14
2496         movaps  -0x10(%rbp),%xmm15
2497 ___
2498 $code.=<<___;
2499         lea     (%rbp),%rsp
2500         pop     %rbp
2501 .Lxts_dec_epilogue:
2502         ret
2503 .size   aesni_xts_decrypt,.-aesni_xts_decrypt
2504 ___
2505 } }}
2506 \f
2507 ########################################################################
2508 # void $PREFIX_cbc_encrypt (const void *inp, void *out,
2509 #                           size_t length, const AES_KEY *key,
2510 #                           unsigned char *ivp,const int enc);
2511 {
2512 my $frame_size = 0x10 + ($win64?0xa0:0);        # used in decrypt
2513 my ($iv,$in0,$in1,$in2,$in3,$in4)=map("%xmm$_",(10..15));
2514 my $inp_=$key_;
2515
2516 $code.=<<___;
2517 .globl  ${PREFIX}_cbc_encrypt
2518 .type   ${PREFIX}_cbc_encrypt,\@function,6
2519 .align  16
2520 ${PREFIX}_cbc_encrypt:
2521         test    $len,$len               # check length
2522         jz      .Lcbc_ret
2523
2524         mov     240($key),$rnds_        # key->rounds
2525         mov     $key,$key_              # backup $key
2526         test    %r9d,%r9d               # 6th argument
2527         jz      .Lcbc_decrypt
2528 #--------------------------- CBC ENCRYPT ------------------------------#
2529         movups  ($ivp),$inout0          # load iv as initial state
2530         mov     $rnds_,$rounds
2531         cmp     \$16,$len
2532         jb      .Lcbc_enc_tail
2533         sub     \$16,$len
2534         jmp     .Lcbc_enc_loop
2535 .align  16
2536 .Lcbc_enc_loop:
2537         movups  ($inp),$inout1          # load input
2538         lea     16($inp),$inp
2539         #xorps  $inout1,$inout0
2540 ___
2541         &aesni_generate1("enc",$key,$rounds,$inout0,$inout1);
2542 $code.=<<___;
2543         mov     $rnds_,$rounds          # restore $rounds
2544         mov     $key_,$key              # restore $key
2545         movups  $inout0,0($out)         # store output
2546         lea     16($out),$out
2547         sub     \$16,$len
2548         jnc     .Lcbc_enc_loop
2549         add     \$16,$len
2550         jnz     .Lcbc_enc_tail
2551         movups  $inout0,($ivp)
2552         jmp     .Lcbc_ret
2553
2554 .Lcbc_enc_tail:
2555         mov     $len,%rcx       # zaps $key
2556         xchg    $inp,$out       # $inp is %rsi and $out is %rdi now
2557         .long   0x9066A4F3      # rep movsb
2558         mov     \$16,%ecx       # zero tail
2559         sub     $len,%rcx
2560         xor     %eax,%eax
2561         .long   0x9066AAF3      # rep stosb
2562         lea     -16(%rdi),%rdi  # rewind $out by 1 block
2563         mov     $rnds_,$rounds  # restore $rounds
2564         mov     %rdi,%rsi       # $inp and $out are the same
2565         mov     $key_,$key      # restore $key
2566         xor     $len,$len       # len=16
2567         jmp     .Lcbc_enc_loop  # one more spin
2568 \f#--------------------------- CBC DECRYPT ------------------------------#
2569 .align  16
2570 .Lcbc_decrypt:
2571         lea     (%rsp),%rax
2572         push    %rbp
2573         sub     \$$frame_size,%rsp
2574         and     \$-16,%rsp      # Linux kernel stack can be incorrectly seeded
2575 ___
2576 $code.=<<___ if ($win64);
2577         movaps  %xmm6,0x10(%rsp)
2578         movaps  %xmm7,0x20(%rsp)
2579         movaps  %xmm8,0x30(%rsp)
2580         movaps  %xmm9,0x40(%rsp)
2581         movaps  %xmm10,0x50(%rsp)
2582         movaps  %xmm11,0x60(%rsp)
2583         movaps  %xmm12,0x70(%rsp)
2584         movaps  %xmm13,0x80(%rsp)
2585         movaps  %xmm14,0x90(%rsp)
2586         movaps  %xmm15,0xa0(%rsp)
2587 .Lcbc_decrypt_body:
2588 ___
2589 $code.=<<___;
2590         lea     -8(%rax),%rbp
2591         movups  ($ivp),$iv
2592         mov     $rnds_,$rounds
2593         cmp     \$0x50,$len
2594         jbe     .Lcbc_dec_tail
2595
2596         $movkey ($key),$rndkey0
2597         movdqu  0x00($inp),$inout0      # load input
2598         movdqu  0x10($inp),$inout1
2599         movdqa  $inout0,$in0
2600         movdqu  0x20($inp),$inout2
2601         movdqa  $inout1,$in1
2602         movdqu  0x30($inp),$inout3
2603         movdqa  $inout2,$in2
2604         movdqu  0x40($inp),$inout4
2605         movdqa  $inout3,$in3
2606         movdqu  0x50($inp),$inout5
2607         movdqa  $inout4,$in4
2608         mov     OPENSSL_ia32cap_P+4(%rip),%r9d
2609         cmp     \$0x70,$len
2610         jbe     .Lcbc_dec_six_or_seven
2611
2612         and     \$`1<<26|1<<22`,%r9d    # isolate XSAVE+MOVBE
2613         sub     \$0x50,$len
2614         cmp     \$`1<<22`,%r9d          # check for MOVBE without XSAVE
2615         je      .Lcbc_dec_loop6_enter
2616         sub     \$0x20,$len
2617         lea     0x70($key),$key         # size optimization
2618         jmp     .Lcbc_dec_loop8_enter
2619 .align  16
2620 .Lcbc_dec_loop8:
2621         movups  $inout7,($out)
2622         lea     0x10($out),$out
2623 .Lcbc_dec_loop8_enter:
2624         movdqu          0x60($inp),$inout6
2625         pxor            $rndkey0,$inout0
2626         movdqu          0x70($inp),$inout7
2627         pxor            $rndkey0,$inout1
2628         $movkey         0x10-0x70($key),$rndkey1
2629         pxor            $rndkey0,$inout2
2630         xor             $inp_,$inp_
2631         cmp             \$0x70,$len     # is there at least 0x60 bytes ahead?
2632         pxor            $rndkey0,$inout3
2633         pxor            $rndkey0,$inout4
2634         pxor            $rndkey0,$inout5
2635         pxor            $rndkey0,$inout6
2636
2637         aesdec          $rndkey1,$inout0
2638         pxor            $rndkey0,$inout7
2639         $movkey         0x20-0x70($key),$rndkey0
2640         aesdec          $rndkey1,$inout1
2641         aesdec          $rndkey1,$inout2
2642         aesdec          $rndkey1,$inout3
2643         aesdec          $rndkey1,$inout4
2644         aesdec          $rndkey1,$inout5
2645         aesdec          $rndkey1,$inout6
2646         setnc           ${inp_}b
2647         shl             \$7,$inp_
2648         aesdec          $rndkey1,$inout7
2649         add             $inp,$inp_
2650         $movkey         0x30-0x70($key),$rndkey1
2651 ___
2652 for($i=1;$i<12;$i++) {
2653 my $rndkeyx = ($i&1)?$rndkey0:$rndkey1;
2654 $code.=<<___    if ($i==7);
2655         cmp             \$11,$rounds
2656 ___
2657 $code.=<<___;
2658         aesdec          $rndkeyx,$inout0
2659         aesdec          $rndkeyx,$inout1
2660         aesdec          $rndkeyx,$inout2
2661         aesdec          $rndkeyx,$inout3
2662         aesdec          $rndkeyx,$inout4
2663         aesdec          $rndkeyx,$inout5
2664         aesdec          $rndkeyx,$inout6
2665         aesdec          $rndkeyx,$inout7
2666         $movkey         `0x30+0x10*$i`-0x70($key),$rndkeyx
2667 ___
2668 $code.=<<___    if ($i<6 || (!($i&1) && $i>7));
2669         nop
2670 ___
2671 $code.=<<___    if ($i==7);
2672         jb              .Lcbc_dec_done
2673 ___
2674 $code.=<<___    if ($i==9);
2675         je              .Lcbc_dec_done
2676 ___
2677 $code.=<<___    if ($i==11);
2678         jmp             .Lcbc_dec_done
2679 ___
2680 }
2681 $code.=<<___;
2682 .align  16
2683 .Lcbc_dec_done:
2684         aesdec          $rndkey1,$inout0
2685         aesdec          $rndkey1,$inout1
2686         pxor            $rndkey0,$iv
2687         pxor            $rndkey0,$in0
2688         aesdec          $rndkey1,$inout2
2689         aesdec          $rndkey1,$inout3
2690         pxor            $rndkey0,$in1
2691         pxor            $rndkey0,$in2
2692         aesdec          $rndkey1,$inout4
2693         aesdec          $rndkey1,$inout5
2694         pxor            $rndkey0,$in3
2695         pxor            $rndkey0,$in4
2696         aesdec          $rndkey1,$inout6
2697         aesdec          $rndkey1,$inout7
2698         movdqu          0x50($inp),$rndkey1
2699
2700         aesdeclast      $iv,$inout0
2701         movdqu          0x60($inp),$iv          # borrow $iv
2702         pxor            $rndkey0,$rndkey1
2703         aesdeclast      $in0,$inout1
2704         pxor            $rndkey0,$iv
2705         movdqu          0x70($inp),$rndkey0     # next IV
2706         aesdeclast      $in1,$inout2
2707         lea             0x80($inp),$inp
2708         movdqu          0x00($inp_),$in0
2709         aesdeclast      $in2,$inout3
2710         aesdeclast      $in3,$inout4
2711         movdqu          0x10($inp_),$in1
2712         movdqu          0x20($inp_),$in2
2713         aesdeclast      $in4,$inout5
2714         aesdeclast      $rndkey1,$inout6
2715         movdqu          0x30($inp_),$in3
2716         movdqu          0x40($inp_),$in4
2717         aesdeclast      $iv,$inout7
2718         movdqa          $rndkey0,$iv            # return $iv
2719         movdqu          0x50($inp_),$rndkey1
2720         $movkey         -0x70($key),$rndkey0
2721
2722         movups          $inout0,($out)          # store output
2723         movdqa          $in0,$inout0
2724         movups          $inout1,0x10($out)
2725         movdqa          $in1,$inout1
2726         movups          $inout2,0x20($out)
2727         movdqa          $in2,$inout2
2728         movups          $inout3,0x30($out)
2729         movdqa          $in3,$inout3
2730         movups          $inout4,0x40($out)
2731         movdqa          $in4,$inout4
2732         movups          $inout5,0x50($out)
2733         movdqa          $rndkey1,$inout5
2734         movups          $inout6,0x60($out)
2735         lea             0x70($out),$out
2736
2737         sub     \$0x80,$len
2738         ja      .Lcbc_dec_loop8
2739
2740         movaps  $inout7,$inout0
2741         lea     -0x70($key),$key
2742         add     \$0x70,$len
2743         jle     .Lcbc_dec_tail_collected
2744         movups  $inout7,($out)
2745         lea     0x10($out),$out
2746         cmp     \$0x50,$len
2747         jbe     .Lcbc_dec_tail
2748
2749         movaps  $in0,$inout0
2750 .Lcbc_dec_six_or_seven:
2751         cmp     \$0x60,$len
2752         ja      .Lcbc_dec_seven
2753
2754         movaps  $inout5,$inout6
2755         call    _aesni_decrypt6
2756         pxor    $iv,$inout0             # ^= IV
2757         movaps  $inout6,$iv
2758         pxor    $in0,$inout1
2759         movdqu  $inout0,($out)
2760         pxor    $in1,$inout2
2761         movdqu  $inout1,0x10($out)
2762         pxor    $in2,$inout3
2763         movdqu  $inout2,0x20($out)
2764         pxor    $in3,$inout4
2765         movdqu  $inout3,0x30($out)
2766         pxor    $in4,$inout5
2767         movdqu  $inout4,0x40($out)
2768         lea     0x50($out),$out
2769         movdqa  $inout5,$inout0
2770         jmp     .Lcbc_dec_tail_collected
2771
2772 .align  16
2773 .Lcbc_dec_seven:
2774         movups  0x60($inp),$inout6
2775         xorps   $inout7,$inout7
2776         call    _aesni_decrypt8
2777         movups  0x50($inp),$inout7
2778         pxor    $iv,$inout0             # ^= IV
2779         movups  0x60($inp),$iv
2780         pxor    $in0,$inout1
2781         movdqu  $inout0,($out)
2782         pxor    $in1,$inout2
2783         movdqu  $inout1,0x10($out)
2784         pxor    $in2,$inout3
2785         movdqu  $inout2,0x20($out)
2786         pxor    $in3,$inout4
2787         movdqu  $inout3,0x30($out)
2788         pxor    $in4,$inout5
2789         movdqu  $inout4,0x40($out)
2790         pxor    $inout7,$inout6
2791         movdqu  $inout5,0x50($out)
2792         lea     0x60($out),$out
2793         movdqa  $inout6,$inout0
2794         jmp     .Lcbc_dec_tail_collected
2795
2796 .align  16
2797 .Lcbc_dec_loop6:
2798         movups  $inout5,($out)
2799         lea     0x10($out),$out
2800         movdqu  0x00($inp),$inout0      # load input
2801         movdqu  0x10($inp),$inout1
2802         movdqa  $inout0,$in0
2803         movdqu  0x20($inp),$inout2
2804         movdqa  $inout1,$in1
2805         movdqu  0x30($inp),$inout3
2806         movdqa  $inout2,$in2
2807         movdqu  0x40($inp),$inout4
2808         movdqa  $inout3,$in3
2809         movdqu  0x50($inp),$inout5
2810         movdqa  $inout4,$in4
2811 .Lcbc_dec_loop6_enter:
2812         lea     0x60($inp),$inp
2813         movdqa  $inout5,$inout6
2814
2815         call    _aesni_decrypt6
2816
2817         pxor    $iv,$inout0             # ^= IV
2818         movdqa  $inout6,$iv
2819         pxor    $in0,$inout1
2820         movdqu  $inout0,($out)
2821         pxor    $in1,$inout2
2822         movdqu  $inout1,0x10($out)
2823         pxor    $in2,$inout3
2824         movdqu  $inout2,0x20($out)
2825         pxor    $in3,$inout4
2826         mov     $key_,$key
2827         movdqu  $inout3,0x30($out)
2828         pxor    $in4,$inout5
2829         mov     $rnds_,$rounds
2830         movdqu  $inout4,0x40($out)
2831         lea     0x50($out),$out
2832         sub     \$0x60,$len
2833         ja      .Lcbc_dec_loop6
2834
2835         movdqa  $inout5,$inout0
2836         add     \$0x50,$len
2837         jle     .Lcbc_dec_tail_collected
2838         movups  $inout5,($out)
2839         lea     0x10($out),$out
2840
2841 .Lcbc_dec_tail:
2842         movups  ($inp),$inout0
2843         sub     \$0x10,$len
2844         jbe     .Lcbc_dec_one
2845
2846         movups  0x10($inp),$inout1
2847         movaps  $inout0,$in0
2848         sub     \$0x10,$len
2849         jbe     .Lcbc_dec_two
2850
2851         movups  0x20($inp),$inout2
2852         movaps  $inout1,$in1
2853         sub     \$0x10,$len
2854         jbe     .Lcbc_dec_three
2855
2856         movups  0x30($inp),$inout3
2857         movaps  $inout2,$in2
2858         sub     \$0x10,$len
2859         jbe     .Lcbc_dec_four
2860
2861         movups  0x40($inp),$inout4
2862         movaps  $inout3,$in3
2863         movaps  $inout4,$in4
2864         xorps   $inout5,$inout5
2865         call    _aesni_decrypt6
2866         pxor    $iv,$inout0
2867         movaps  $in4,$iv
2868         pxor    $in0,$inout1
2869         movdqu  $inout0,($out)
2870         pxor    $in1,$inout2
2871         movdqu  $inout1,0x10($out)
2872         pxor    $in2,$inout3
2873         movdqu  $inout2,0x20($out)
2874         pxor    $in3,$inout4
2875         movdqu  $inout3,0x30($out)
2876         lea     0x40($out),$out
2877         movdqa  $inout4,$inout0
2878         sub     \$0x10,$len
2879         jmp     .Lcbc_dec_tail_collected
2880
2881 .align  16
2882 .Lcbc_dec_one:
2883         movaps  $inout0,$in0
2884 ___
2885         &aesni_generate1("dec",$key,$rounds);
2886 $code.=<<___;
2887         xorps   $iv,$inout0
2888         movaps  $in0,$iv
2889         jmp     .Lcbc_dec_tail_collected
2890 .align  16
2891 .Lcbc_dec_two:
2892         movaps  $inout1,$in1
2893         call    _aesni_decrypt2
2894         pxor    $iv,$inout0
2895         movaps  $in1,$iv
2896         pxor    $in0,$inout1
2897         movdqu  $inout0,($out)
2898         movdqa  $inout1,$inout0
2899         lea     0x10($out),$out
2900         jmp     .Lcbc_dec_tail_collected
2901 .align  16
2902 .Lcbc_dec_three:
2903         movaps  $inout2,$in2
2904         call    _aesni_decrypt3
2905         pxor    $iv,$inout0
2906         movaps  $in2,$iv
2907         pxor    $in0,$inout1
2908         movdqu  $inout0,($out)
2909         pxor    $in1,$inout2
2910         movdqu  $inout1,0x10($out)
2911         movdqa  $inout2,$inout0
2912         lea     0x20($out),$out
2913         jmp     .Lcbc_dec_tail_collected
2914 .align  16
2915 .Lcbc_dec_four:
2916         movaps  $inout3,$in3
2917         call    _aesni_decrypt4
2918         pxor    $iv,$inout0
2919         movaps  $in3,$iv
2920         pxor    $in0,$inout1
2921         movdqu  $inout0,($out)
2922         pxor    $in1,$inout2
2923         movdqu  $inout1,0x10($out)
2924         pxor    $in2,$inout3
2925         movdqu  $inout2,0x20($out)
2926         movdqa  $inout3,$inout0
2927         lea     0x30($out),$out
2928         jmp     .Lcbc_dec_tail_collected
2929
2930 .align  16
2931 .Lcbc_dec_tail_collected:
2932         movups  $iv,($ivp)
2933         and     \$15,$len
2934         jnz     .Lcbc_dec_tail_partial
2935         movups  $inout0,($out)
2936         jmp     .Lcbc_dec_ret
2937 .align  16
2938 .Lcbc_dec_tail_partial:
2939         movaps  $inout0,(%rsp)
2940         mov     \$16,%rcx
2941         mov     $out,%rdi
2942         sub     $len,%rcx
2943         lea     (%rsp),%rsi
2944         .long   0x9066A4F3      # rep movsb
2945
2946 .Lcbc_dec_ret:
2947 ___
2948 $code.=<<___ if ($win64);
2949         movaps  0x10(%rsp),%xmm6
2950         movaps  0x20(%rsp),%xmm7
2951         movaps  0x30(%rsp),%xmm8
2952         movaps  0x40(%rsp),%xmm9
2953         movaps  0x50(%rsp),%xmm10
2954         movaps  0x60(%rsp),%xmm11
2955         movaps  0x70(%rsp),%xmm12
2956         movaps  0x80(%rsp),%xmm13
2957         movaps  0x90(%rsp),%xmm14
2958         movaps  0xa0(%rsp),%xmm15
2959 ___
2960 $code.=<<___;
2961         lea     (%rbp),%rsp
2962         pop     %rbp
2963 .Lcbc_ret:
2964         ret
2965 .size   ${PREFIX}_cbc_encrypt,.-${PREFIX}_cbc_encrypt
2966 ___
2967 } \f
2968 # int $PREFIX_set_[en|de]crypt_key (const unsigned char *userKey,
2969 #                               int bits, AES_KEY *key)
2970 { my ($inp,$bits,$key) = @_4args;
2971   $bits =~ s/%r/%e/;
2972
2973 $code.=<<___;
2974 .globl  ${PREFIX}_set_decrypt_key
2975 .type   ${PREFIX}_set_decrypt_key,\@abi-omnipotent
2976 .align  16
2977 ${PREFIX}_set_decrypt_key:
2978         .byte   0x48,0x83,0xEC,0x08     # sub rsp,8
2979         call    __aesni_set_encrypt_key
2980         shl     \$4,$bits               # rounds-1 after _aesni_set_encrypt_key
2981         test    %eax,%eax
2982         jnz     .Ldec_key_ret
2983         lea     16($key,$bits),$inp     # points at the end of key schedule
2984
2985         $movkey ($key),%xmm0            # just swap
2986         $movkey ($inp),%xmm1
2987         $movkey %xmm0,($inp)
2988         $movkey %xmm1,($key)
2989         lea     16($key),$key
2990         lea     -16($inp),$inp
2991
2992 .Ldec_key_inverse:
2993         $movkey ($key),%xmm0            # swap and inverse
2994         $movkey ($inp),%xmm1
2995         aesimc  %xmm0,%xmm0
2996         aesimc  %xmm1,%xmm1
2997         lea     16($key),$key
2998         lea     -16($inp),$inp
2999         $movkey %xmm0,16($inp)
3000         $movkey %xmm1,-16($key)
3001         cmp     $key,$inp
3002         ja      .Ldec_key_inverse
3003
3004         $movkey ($key),%xmm0            # inverse middle
3005         aesimc  %xmm0,%xmm0
3006         $movkey %xmm0,($inp)
3007 .Ldec_key_ret:
3008         add     \$8,%rsp
3009         ret
3010 .LSEH_end_set_decrypt_key:
3011 .size   ${PREFIX}_set_decrypt_key,.-${PREFIX}_set_decrypt_key
3012 ___
3013 \f
3014 # This is based on submission by
3015 #
3016 #       Huang Ying <ying.huang@intel.com>
3017 #       Vinodh Gopal <vinodh.gopal@intel.com>
3018 #       Kahraman Akdemir
3019 #
3020 # Agressively optimized in respect to aeskeygenassist's critical path
3021 # and is contained in %xmm0-5 to meet Win64 ABI requirement.
3022 #
3023 $code.=<<___;
3024 .globl  ${PREFIX}_set_encrypt_key
3025 .type   ${PREFIX}_set_encrypt_key,\@abi-omnipotent
3026 .align  16
3027 ${PREFIX}_set_encrypt_key:
3028 __aesni_set_encrypt_key:
3029         .byte   0x48,0x83,0xEC,0x08     # sub rsp,8
3030         mov     \$-1,%rax
3031         test    $inp,$inp
3032         jz      .Lenc_key_ret
3033         test    $key,$key
3034         jz      .Lenc_key_ret
3035
3036         movups  ($inp),%xmm0            # pull first 128 bits of *userKey
3037         xorps   %xmm4,%xmm4             # low dword of xmm4 is assumed 0
3038         lea     16($key),%rax
3039         cmp     \$256,$bits
3040         je      .L14rounds
3041         cmp     \$192,$bits
3042         je      .L12rounds
3043         cmp     \$128,$bits
3044         jne     .Lbad_keybits
3045
3046 .L10rounds:
3047         mov     \$9,$bits                       # 10 rounds for 128-bit key
3048         $movkey %xmm0,($key)                    # round 0
3049         aeskeygenassist \$0x1,%xmm0,%xmm1       # round 1
3050         call            .Lkey_expansion_128_cold
3051         aeskeygenassist \$0x2,%xmm0,%xmm1       # round 2
3052         call            .Lkey_expansion_128
3053         aeskeygenassist \$0x4,%xmm0,%xmm1       # round 3
3054         call            .Lkey_expansion_128
3055         aeskeygenassist \$0x8,%xmm0,%xmm1       # round 4
3056         call            .Lkey_expansion_128
3057         aeskeygenassist \$0x10,%xmm0,%xmm1      # round 5
3058         call            .Lkey_expansion_128
3059         aeskeygenassist \$0x20,%xmm0,%xmm1      # round 6
3060         call            .Lkey_expansion_128
3061         aeskeygenassist \$0x40,%xmm0,%xmm1      # round 7
3062         call            .Lkey_expansion_128
3063         aeskeygenassist \$0x80,%xmm0,%xmm1      # round 8
3064         call            .Lkey_expansion_128
3065         aeskeygenassist \$0x1b,%xmm0,%xmm1      # round 9
3066         call            .Lkey_expansion_128
3067         aeskeygenassist \$0x36,%xmm0,%xmm1      # round 10
3068         call            .Lkey_expansion_128
3069         $movkey %xmm0,(%rax)
3070         mov     $bits,80(%rax)  # 240(%rdx)
3071         xor     %eax,%eax
3072         jmp     .Lenc_key_ret
3073
3074 .align  16
3075 .L12rounds:
3076         movq    16($inp),%xmm2                  # remaining 1/3 of *userKey
3077         mov     \$11,$bits                      # 12 rounds for 192
3078         $movkey %xmm0,($key)                    # round 0
3079         aeskeygenassist \$0x1,%xmm2,%xmm1       # round 1,2
3080         call            .Lkey_expansion_192a_cold
3081         aeskeygenassist \$0x2,%xmm2,%xmm1       # round 2,3
3082         call            .Lkey_expansion_192b
3083         aeskeygenassist \$0x4,%xmm2,%xmm1       # round 4,5
3084         call            .Lkey_expansion_192a
3085         aeskeygenassist \$0x8,%xmm2,%xmm1       # round 5,6
3086         call            .Lkey_expansion_192b
3087         aeskeygenassist \$0x10,%xmm2,%xmm1      # round 7,8
3088         call            .Lkey_expansion_192a
3089         aeskeygenassist \$0x20,%xmm2,%xmm1      # round 8,9
3090         call            .Lkey_expansion_192b
3091         aeskeygenassist \$0x40,%xmm2,%xmm1      # round 10,11
3092         call            .Lkey_expansion_192a
3093         aeskeygenassist \$0x80,%xmm2,%xmm1      # round 11,12
3094         call            .Lkey_expansion_192b
3095         $movkey %xmm0,(%rax)
3096         mov     $bits,48(%rax)  # 240(%rdx)
3097         xor     %rax, %rax
3098         jmp     .Lenc_key_ret
3099
3100 .align  16
3101 .L14rounds:
3102         movups  16($inp),%xmm2                  # remaning half of *userKey
3103         mov     \$13,$bits                      # 14 rounds for 256
3104         lea     16(%rax),%rax
3105         $movkey %xmm0,($key)                    # round 0
3106         $movkey %xmm2,16($key)                  # round 1
3107         aeskeygenassist \$0x1,%xmm2,%xmm1       # round 2
3108         call            .Lkey_expansion_256a_cold
3109         aeskeygenassist \$0x1,%xmm0,%xmm1       # round 3
3110         call            .Lkey_expansion_256b
3111         aeskeygenassist \$0x2,%xmm2,%xmm1       # round 4
3112         call            .Lkey_expansion_256a
3113         aeskeygenassist \$0x2,%xmm0,%xmm1       # round 5
3114         call            .Lkey_expansion_256b
3115         aeskeygenassist \$0x4,%xmm2,%xmm1       # round 6
3116         call            .Lkey_expansion_256a
3117         aeskeygenassist \$0x4,%xmm0,%xmm1       # round 7
3118         call            .Lkey_expansion_256b
3119         aeskeygenassist \$0x8,%xmm2,%xmm1       # round 8
3120         call            .Lkey_expansion_256a
3121         aeskeygenassist \$0x8,%xmm0,%xmm1       # round 9
3122         call            .Lkey_expansion_256b
3123         aeskeygenassist \$0x10,%xmm2,%xmm1      # round 10
3124         call            .Lkey_expansion_256a
3125         aeskeygenassist \$0x10,%xmm0,%xmm1      # round 11
3126         call            .Lkey_expansion_256b
3127         aeskeygenassist \$0x20,%xmm2,%xmm1      # round 12
3128         call            .Lkey_expansion_256a
3129         aeskeygenassist \$0x20,%xmm0,%xmm1      # round 13
3130         call            .Lkey_expansion_256b
3131         aeskeygenassist \$0x40,%xmm2,%xmm1      # round 14
3132         call            .Lkey_expansion_256a
3133         $movkey %xmm0,(%rax)
3134         mov     $bits,16(%rax)  # 240(%rdx)
3135         xor     %rax,%rax
3136         jmp     .Lenc_key_ret
3137
3138 .align  16
3139 .Lbad_keybits:
3140         mov     \$-2,%rax
3141 .Lenc_key_ret:
3142         add     \$8,%rsp
3143         ret
3144 .LSEH_end_set_encrypt_key:
3145 \f
3146 .align  16
3147 .Lkey_expansion_128:
3148         $movkey %xmm0,(%rax)
3149         lea     16(%rax),%rax
3150 .Lkey_expansion_128_cold:
3151         shufps  \$0b00010000,%xmm0,%xmm4
3152         xorps   %xmm4, %xmm0
3153         shufps  \$0b10001100,%xmm0,%xmm4
3154         xorps   %xmm4, %xmm0
3155         shufps  \$0b11111111,%xmm1,%xmm1        # critical path
3156         xorps   %xmm1,%xmm0
3157         ret
3158
3159 .align 16
3160 .Lkey_expansion_192a:
3161         $movkey %xmm0,(%rax)
3162         lea     16(%rax),%rax
3163 .Lkey_expansion_192a_cold:
3164         movaps  %xmm2, %xmm5
3165 .Lkey_expansion_192b_warm:
3166         shufps  \$0b00010000,%xmm0,%xmm4
3167         movdqa  %xmm2,%xmm3
3168         xorps   %xmm4,%xmm0
3169         shufps  \$0b10001100,%xmm0,%xmm4
3170         pslldq  \$4,%xmm3
3171         xorps   %xmm4,%xmm0
3172         pshufd  \$0b01010101,%xmm1,%xmm1        # critical path
3173         pxor    %xmm3,%xmm2
3174         pxor    %xmm1,%xmm0
3175         pshufd  \$0b11111111,%xmm0,%xmm3
3176         pxor    %xmm3,%xmm2
3177         ret
3178
3179 .align 16
3180 .Lkey_expansion_192b:
3181         movaps  %xmm0,%xmm3
3182         shufps  \$0b01000100,%xmm0,%xmm5
3183         $movkey %xmm5,(%rax)
3184         shufps  \$0b01001110,%xmm2,%xmm3
3185         $movkey %xmm3,16(%rax)
3186         lea     32(%rax),%rax
3187         jmp     .Lkey_expansion_192b_warm
3188
3189 .align  16
3190 .Lkey_expansion_256a:
3191         $movkey %xmm2,(%rax)
3192         lea     16(%rax),%rax
3193 .Lkey_expansion_256a_cold:
3194         shufps  \$0b00010000,%xmm0,%xmm4
3195         xorps   %xmm4,%xmm0
3196         shufps  \$0b10001100,%xmm0,%xmm4
3197         xorps   %xmm4,%xmm0
3198         shufps  \$0b11111111,%xmm1,%xmm1        # critical path
3199         xorps   %xmm1,%xmm0
3200         ret
3201
3202 .align 16
3203 .Lkey_expansion_256b:
3204         $movkey %xmm0,(%rax)
3205         lea     16(%rax),%rax
3206
3207         shufps  \$0b00010000,%xmm2,%xmm4
3208         xorps   %xmm4,%xmm2
3209         shufps  \$0b10001100,%xmm2,%xmm4
3210         xorps   %xmm4,%xmm2
3211         shufps  \$0b10101010,%xmm1,%xmm1        # critical path
3212         xorps   %xmm1,%xmm2
3213         ret
3214 .size   ${PREFIX}_set_encrypt_key,.-${PREFIX}_set_encrypt_key
3215 .size   __aesni_set_encrypt_key,.-__aesni_set_encrypt_key
3216 ___
3217 }
3218 \f
3219 $code.=<<___;
3220 .align  64
3221 .Lbswap_mask:
3222         .byte   15,14,13,12,11,10,9,8,7,6,5,4,3,2,1,0
3223 .Lincrement32:
3224         .long   6,6,6,0
3225 .Lincrement64:
3226         .long   1,0,0,0
3227 .Lxts_magic:
3228         .long   0x87,0,1,0
3229 .Lincrement1:
3230         .byte   0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1
3231
3232 .asciz  "AES for Intel AES-NI, CRYPTOGAMS by <appro\@openssl.org>"
3233 .align  64
3234 ___
3235
3236 # EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
3237 #               CONTEXT *context,DISPATCHER_CONTEXT *disp)
3238 if ($win64) {
3239 $rec="%rcx";
3240 $frame="%rdx";
3241 $context="%r8";
3242 $disp="%r9";
3243
3244 $code.=<<___;
3245 .extern __imp_RtlVirtualUnwind
3246 ___
3247 $code.=<<___ if ($PREFIX eq "aesni");
3248 .type   ecb_ccm64_se_handler,\@abi-omnipotent
3249 .align  16
3250 ecb_ccm64_se_handler:
3251         push    %rsi
3252         push    %rdi
3253         push    %rbx
3254         push    %rbp
3255         push    %r12
3256         push    %r13
3257         push    %r14
3258         push    %r15
3259         pushfq
3260         sub     \$64,%rsp
3261
3262         mov     120($context),%rax      # pull context->Rax
3263         mov     248($context),%rbx      # pull context->Rip
3264
3265         mov     8($disp),%rsi           # disp->ImageBase
3266         mov     56($disp),%r11          # disp->HandlerData
3267
3268         mov     0(%r11),%r10d           # HandlerData[0]
3269         lea     (%rsi,%r10),%r10        # prologue label
3270         cmp     %r10,%rbx               # context->Rip<prologue label
3271         jb      .Lcommon_seh_tail
3272
3273         mov     152($context),%rax      # pull context->Rsp
3274
3275         mov     4(%r11),%r10d           # HandlerData[1]
3276         lea     (%rsi,%r10),%r10        # epilogue label
3277         cmp     %r10,%rbx               # context->Rip>=epilogue label
3278         jae     .Lcommon_seh_tail
3279
3280         lea     0(%rax),%rsi            # %xmm save area
3281         lea     512($context),%rdi      # &context.Xmm6
3282         mov     \$8,%ecx                # 4*sizeof(%xmm0)/sizeof(%rax)
3283         .long   0xa548f3fc              # cld; rep movsq
3284         lea     0x58(%rax),%rax         # adjust stack pointer
3285
3286         jmp     .Lcommon_seh_tail
3287 .size   ecb_ccm64_se_handler,.-ecb_ccm64_se_handler
3288
3289 .type   ctr_xts_se_handler,\@abi-omnipotent
3290 .align  16
3291 ctr_xts_se_handler:
3292         push    %rsi
3293         push    %rdi
3294         push    %rbx
3295         push    %rbp
3296         push    %r12
3297         push    %r13
3298         push    %r14
3299         push    %r15
3300         pushfq
3301         sub     \$64,%rsp
3302
3303         mov     120($context),%rax      # pull context->Rax
3304         mov     248($context),%rbx      # pull context->Rip
3305
3306         mov     8($disp),%rsi           # disp->ImageBase
3307         mov     56($disp),%r11          # disp->HandlerData
3308
3309         mov     0(%r11),%r10d           # HandlerData[0]
3310         lea     (%rsi,%r10),%r10        # prologue lable
3311         cmp     %r10,%rbx               # context->Rip<prologue label
3312         jb      .Lcommon_seh_tail
3313
3314         mov     152($context),%rax      # pull context->Rsp
3315
3316         mov     4(%r11),%r10d           # HandlerData[1]
3317         lea     (%rsi,%r10),%r10        # epilogue label
3318         cmp     %r10,%rbx               # context->Rip>=epilogue label
3319         jae     .Lcommon_seh_tail
3320
3321         mov     160($context),%rax      # pull context->Rbp
3322         lea     -0xa0(%rax),%rsi        # %xmm save area
3323         lea     512($context),%rdi      # & context.Xmm6
3324         mov     \$20,%ecx               # 10*sizeof(%xmm0)/sizeof(%rax)
3325         .long   0xa548f3fc              # cld; rep movsq
3326
3327         jmp     .Lcommon_rbp_tail
3328 .size   ctr_xts_se_handler,.-ctr_xts_se_handler
3329 ___
3330 $code.=<<___;
3331 .type   cbc_se_handler,\@abi-omnipotent
3332 .align  16
3333 cbc_se_handler:
3334         push    %rsi
3335         push    %rdi
3336         push    %rbx
3337         push    %rbp
3338         push    %r12
3339         push    %r13
3340         push    %r14
3341         push    %r15
3342         pushfq
3343         sub     \$64,%rsp
3344
3345         mov     152($context),%rax      # pull context->Rsp
3346         mov     248($context),%rbx      # pull context->Rip
3347
3348         lea     .Lcbc_decrypt(%rip),%r10
3349         cmp     %r10,%rbx               # context->Rip<"prologue" label
3350         jb      .Lcommon_seh_tail
3351
3352         lea     .Lcbc_decrypt_body(%rip),%r10
3353         cmp     %r10,%rbx               # context->Rip<cbc_decrypt_body
3354         jb      .Lrestore_cbc_rax
3355
3356         lea     .Lcbc_ret(%rip),%r10
3357         cmp     %r10,%rbx               # context->Rip>="epilogue" label
3358         jae     .Lcommon_seh_tail
3359
3360         lea     16(%rax),%rsi           # %xmm save area
3361         lea     512($context),%rdi      # &context.Xmm6
3362         mov     \$20,%ecx               # 10*sizeof(%xmm0)/sizeof(%rax)
3363         .long   0xa548f3fc              # cld; rep movsq
3364
3365 .Lcommon_rbp_tail:
3366         mov     160($context),%rax      # pull context->Rbp
3367         mov     (%rax),%rbp             # restore saved %rbp
3368         lea     8(%rax),%rax            # adjust stack pointer
3369         mov     %rbp,160($context)      # restore context->Rbp
3370         jmp     .Lcommon_seh_tail
3371
3372 .Lrestore_cbc_rax:
3373         mov     120($context),%rax
3374
3375 .Lcommon_seh_tail:
3376         mov     8(%rax),%rdi
3377         mov     16(%rax),%rsi
3378         mov     %rax,152($context)      # restore context->Rsp
3379         mov     %rsi,168($context)      # restore context->Rsi
3380         mov     %rdi,176($context)      # restore context->Rdi
3381
3382         mov     40($disp),%rdi          # disp->ContextRecord
3383         mov     $context,%rsi           # context
3384         mov     \$154,%ecx              # sizeof(CONTEXT)
3385         .long   0xa548f3fc              # cld; rep movsq
3386
3387         mov     $disp,%rsi
3388         xor     %rcx,%rcx               # arg1, UNW_FLAG_NHANDLER
3389         mov     8(%rsi),%rdx            # arg2, disp->ImageBase
3390         mov     0(%rsi),%r8             # arg3, disp->ControlPc
3391         mov     16(%rsi),%r9            # arg4, disp->FunctionEntry
3392         mov     40(%rsi),%r10           # disp->ContextRecord
3393         lea     56(%rsi),%r11           # &disp->HandlerData
3394         lea     24(%rsi),%r12           # &disp->EstablisherFrame
3395         mov     %r10,32(%rsp)           # arg5
3396         mov     %r11,40(%rsp)           # arg6
3397         mov     %r12,48(%rsp)           # arg7
3398         mov     %rcx,56(%rsp)           # arg8, (NULL)
3399         call    *__imp_RtlVirtualUnwind(%rip)
3400
3401         mov     \$1,%eax                # ExceptionContinueSearch
3402         add     \$64,%rsp
3403         popfq
3404         pop     %r15
3405         pop     %r14
3406         pop     %r13
3407         pop     %r12
3408         pop     %rbp
3409         pop     %rbx
3410         pop     %rdi
3411         pop     %rsi
3412         ret
3413 .size   cbc_se_handler,.-cbc_se_handler
3414
3415 .section        .pdata
3416 .align  4
3417 ___
3418 $code.=<<___ if ($PREFIX eq "aesni");
3419         .rva    .LSEH_begin_aesni_ecb_encrypt
3420         .rva    .LSEH_end_aesni_ecb_encrypt
3421         .rva    .LSEH_info_ecb
3422
3423         .rva    .LSEH_begin_aesni_ccm64_encrypt_blocks
3424         .rva    .LSEH_end_aesni_ccm64_encrypt_blocks
3425         .rva    .LSEH_info_ccm64_enc
3426
3427         .rva    .LSEH_begin_aesni_ccm64_decrypt_blocks
3428         .rva    .LSEH_end_aesni_ccm64_decrypt_blocks
3429         .rva    .LSEH_info_ccm64_dec
3430
3431         .rva    .LSEH_begin_aesni_ctr32_encrypt_blocks
3432         .rva    .LSEH_end_aesni_ctr32_encrypt_blocks
3433         .rva    .LSEH_info_ctr32
3434
3435         .rva    .LSEH_begin_aesni_xts_encrypt
3436         .rva    .LSEH_end_aesni_xts_encrypt
3437         .rva    .LSEH_info_xts_enc
3438
3439         .rva    .LSEH_begin_aesni_xts_decrypt
3440         .rva    .LSEH_end_aesni_xts_decrypt
3441         .rva    .LSEH_info_xts_dec
3442 ___
3443 $code.=<<___;
3444         .rva    .LSEH_begin_${PREFIX}_cbc_encrypt
3445         .rva    .LSEH_end_${PREFIX}_cbc_encrypt
3446         .rva    .LSEH_info_cbc
3447
3448         .rva    ${PREFIX}_set_decrypt_key
3449         .rva    .LSEH_end_set_decrypt_key
3450         .rva    .LSEH_info_key
3451
3452         .rva    ${PREFIX}_set_encrypt_key
3453         .rva    .LSEH_end_set_encrypt_key
3454         .rva    .LSEH_info_key
3455 .section        .xdata
3456 .align  8
3457 ___
3458 $code.=<<___ if ($PREFIX eq "aesni");
3459 .LSEH_info_ecb:
3460         .byte   9,0,0,0
3461         .rva    ecb_ccm64_se_handler
3462         .rva    .Lecb_enc_body,.Lecb_enc_ret            # HandlerData[]
3463 .LSEH_info_ccm64_enc:
3464         .byte   9,0,0,0
3465         .rva    ecb_ccm64_se_handler
3466         .rva    .Lccm64_enc_body,.Lccm64_enc_ret        # HandlerData[]
3467 .LSEH_info_ccm64_dec:
3468         .byte   9,0,0,0
3469         .rva    ecb_ccm64_se_handler
3470         .rva    .Lccm64_dec_body,.Lccm64_dec_ret        # HandlerData[]
3471 .LSEH_info_ctr32:
3472         .byte   9,0,0,0
3473         .rva    ctr_xts_se_handler
3474         .rva    .Lctr32_body,.Lctr32_epilogue           # HandlerData[]
3475 .LSEH_info_xts_enc:
3476         .byte   9,0,0,0
3477         .rva    ctr_xts_se_handler
3478         .rva    .Lxts_enc_body,.Lxts_enc_epilogue       # HandlerData[]
3479 .LSEH_info_xts_dec:
3480         .byte   9,0,0,0
3481         .rva    ctr_xts_se_handler
3482         .rva    .Lxts_dec_body,.Lxts_dec_epilogue       # HandlerData[]
3483 ___
3484 $code.=<<___;
3485 .LSEH_info_cbc:
3486         .byte   9,0,0,0
3487         .rva    cbc_se_handler
3488 .LSEH_info_key:
3489         .byte   0x01,0x04,0x01,0x00
3490         .byte   0x04,0x02,0x00,0x00     # sub rsp,8
3491 ___
3492 }
3493
3494 sub rex {
3495   local *opcode=shift;
3496   my ($dst,$src)=@_;
3497   my $rex=0;
3498
3499     $rex|=0x04                  if($dst>=8);
3500     $rex|=0x01                  if($src>=8);
3501     push @opcode,$rex|0x40      if($rex);
3502 }
3503
3504 sub aesni {
3505   my $line=shift;
3506   my @opcode=(0x66);
3507
3508     if ($line=~/(aeskeygenassist)\s+\$([x0-9a-f]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) {
3509         rex(\@opcode,$4,$3);
3510         push @opcode,0x0f,0x3a,0xdf;
3511         push @opcode,0xc0|($3&7)|(($4&7)<<3);   # ModR/M
3512         my $c=$2;
3513         push @opcode,$c=~/^0/?oct($c):$c;
3514         return ".byte\t".join(',',@opcode);
3515     }
3516     elsif ($line=~/(aes[a-z]+)\s+%xmm([0-9]+),\s*%xmm([0-9]+)/) {
3517         my %opcodelet = (
3518                 "aesimc" => 0xdb,
3519                 "aesenc" => 0xdc,       "aesenclast" => 0xdd,
3520                 "aesdec" => 0xde,       "aesdeclast" => 0xdf
3521         );
3522         return undef if (!defined($opcodelet{$1}));
3523         rex(\@opcode,$3,$2);
3524         push @opcode,0x0f,0x38,$opcodelet{$1};
3525         push @opcode,0xc0|($2&7)|(($3&7)<<3);   # ModR/M
3526         return ".byte\t".join(',',@opcode);
3527     }
3528     elsif ($line=~/(aes[a-z]+)\s+([0x1-9a-fA-F]*)\(%rsp\),\s*%xmm([0-9]+)/) {
3529         my %opcodelet = (
3530                 "aesenc" => 0xdc,       "aesenclast" => 0xdd,
3531                 "aesdec" => 0xde,       "aesdeclast" => 0xdf
3532         );
3533         return undef if (!defined($opcodelet{$1}));
3534         my $off = $2;
3535         push @opcode,0x44 if ($3>=8);
3536         push @opcode,0x0f,0x38,$opcodelet{$1};
3537         push @opcode,0x44|(($3&7)<<3),0x24;     # ModR/M
3538         push @opcode,($off=~/^0/?oct($off):$off)&0xff;
3539         return ".byte\t".join(',',@opcode);
3540     }
3541     return $line;
3542 }
3543
3544 sub movbe {
3545         ".byte  0x0f,0x38,0xf1,0x44,0x24,".shift;
3546 }
3547
3548 $code =~ s/\`([^\`]*)\`/eval($1)/gem;
3549 $code =~ s/\b(aes.*%xmm[0-9]+).*$/aesni($1)/gem;
3550 #$code =~ s/\bmovbe\s+%eax/bswap %eax; mov %eax/gm;     # debugging artefact
3551 $code =~ s/\bmovbe\s+%eax,\s*([0-9]+)\(%rsp\)/movbe($1)/gem;
3552
3553 print $code;
3554
3555 close STDOUT;