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

   1 #!/usr/bin/env perl
   2 #
   3 # ====================================================================
   4 # Written by Andy Polyakov <appro@fy.chalmers.se> 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    0.93
 133 # CTR                                   1.14    0.91    n/a
 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 1.33 cycles processing
 157 # one byte out of 8KB with 128-bit key, Sandy Bridge - 0.97. Just like
 158 # in CTR mode AES instruction interleave factor was chosen to be 6x.
 159
 160 $PREFIX="aesni";        # if $PREFIX is set to "AES", the script
 161                         # generates drop-in replacement for
 162                         # crypto/aes/asm/aes-x86_64.pl:-)
 163
 164 $flavour = shift;
 165 $output  = shift;
 166 if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
 167
 168 $win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);
 169
 170 $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
 171 ( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
 172 ( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or
 173 die "can't locate x86_64-xlate.pl";
 174
 175 open STDOUT,"| $^X $xlate $flavour $output";
 176
 177 $movkey = $PREFIX eq "aesni" ? "movups" : "movups";
 178 @_4args=$win64? ("%rcx","%rdx","%r8", "%r9") :  # Win64 order
 179                 ("%rdi","%rsi","%rdx","%rcx");  # Unix order
 180
 181 $code=".text\n";
 182
 183 $rounds="%eax"; # input to and changed by aesni_[en|de]cryptN !!!
 184 # this is natural Unix argument order for public $PREFIX_[ecb|cbc]_encrypt ...
 185 $inp="%rdi";
 186 $out="%rsi";
 187 $len="%rdx";
 188 $key="%rcx";    # input to and changed by aesni_[en|de]cryptN !!!
 189 $ivp="%r8";     # cbc, ctr, ...
 190
 191 $rnds_="%r10d"; # backup copy for $rounds
 192 $key_="%r11";   # backup copy for $key
 193
 194 # %xmm register layout
 195 $rndkey0="%xmm0";       $rndkey1="%xmm1";
 196 $inout0="%xmm2";        $inout1="%xmm3";
 197 $inout2="%xmm4";        $inout3="%xmm5";
 198 $inout4="%xmm6";        $inout5="%xmm7";
 199 $inout6="%xmm8";        $inout7="%xmm9";
 200
 201 $in2="%xmm6";           $in1="%xmm7";   # used in CBC decrypt, CTR, ...
 202 $in0="%xmm8";           $iv="%xmm9";
 203 \f
 204 # Inline version of internal aesni_[en|de]crypt1.
 205 #
 206 # Why folded loop? Because aes[enc|dec] is slow enough to accommodate
 207 # cycles which take care of loop variables...
 208 { my $sn;
 209 sub aesni_generate1 {
 210 my ($p,$key,$rounds,$inout,$ivec)=@_;   $inout=$inout0 if (!defined($inout));
 211 ++$sn;
 212 $code.=<<___;
 213         $movkey ($key),$rndkey0
 214         $movkey 16($key),$rndkey1
 215 ___
 216 $code.=<<___ if (defined($ivec));
 217         xorps   $rndkey0,$ivec
 218         lea     32($key),$key
 219         xorps   $ivec,$inout
 220 ___
 221 $code.=<<___ if (!defined($ivec));
 222         lea     32($key),$key
 223         xorps   $rndkey0,$inout
 224 ___
 225 $code.=<<___;
 226 .Loop_${p}1_$sn:
 227         aes${p} $rndkey1,$inout
 228         dec     $rounds
 229         $movkey ($key),$rndkey1
 230         lea     16($key),$key
 231         jnz     .Loop_${p}1_$sn # loop body is 16 bytes
 232         aes${p}last     $rndkey1,$inout
 233 ___
 234 }}
 235 # void $PREFIX_[en|de]crypt (const void *inp,void *out,const AES_KEY *key);
 236 #
 237 { my ($inp,$out,$key) = @_4args;
 238
 239 $code.=<<___;
 240 .globl  ${PREFIX}_encrypt
 241 .type   ${PREFIX}_encrypt,\@abi-omnipotent
 242 .align  16
 243 ${PREFIX}_encrypt:
 244         movups  ($inp),$inout0          # load input
 245         mov     240($key),$rounds       # key->rounds
 246 ___
 247         &aesni_generate1("enc",$key,$rounds);
 248 $code.=<<___;
 249         movups  $inout0,($out)          # output
 250         ret
 251 .size   ${PREFIX}_encrypt,.-${PREFIX}_encrypt
 252
 253 .globl  ${PREFIX}_decrypt
 254 .type   ${PREFIX}_decrypt,\@abi-omnipotent
 255 .align  16
 256 ${PREFIX}_decrypt:
 257         movups  ($inp),$inout0          # load input
 258         mov     240($key),$rounds       # key->rounds
 259 ___
 260         &aesni_generate1("dec",$key,$rounds);
 261 $code.=<<___;
 262         movups  $inout0,($out)          # output
 263         ret
 264 .size   ${PREFIX}_decrypt, .-${PREFIX}_decrypt
 265 ___
 266 }
 267 \f
 268 # _aesni_[en|de]cryptN are private interfaces, N denotes interleave
 269 # factor. Why 3x subroutine were originally used in loops? Even though
 270 # aes[enc|dec] latency was originally 6, it could be scheduled only
 271 # every *2nd* cycle. Thus 3x interleave was the one providing optimal
 272 # utilization, i.e. when subroutine's throughput is virtually same as
 273 # of non-interleaved subroutine [for number of input blocks up to 3].
 274 # This is why it makes no sense to implement 2x subroutine.
 275 # aes[enc|dec] latency in next processor generation is 8, but the
 276 # instructions can be scheduled every cycle. Optimal interleave for
 277 # new processor is therefore 8x...
 278 sub aesni_generate3 {
 279 my $dir=shift;
 280 # As already mentioned it takes in $key and $rounds, which are *not*
 281 # preserved. $inout[0-2] is cipher/clear text...
 282 $code.=<<___;
 283 .type   _aesni_${dir}rypt3,\@abi-omnipotent
 284 .align  16
 285 _aesni_${dir}rypt3:
 286         $movkey ($key),$rndkey0
 287         shr     \$1,$rounds
 288         $movkey 16($key),$rndkey1
 289         lea     32($key),$key
 290         xorps   $rndkey0,$inout0
 291         xorps   $rndkey0,$inout1
 292         xorps   $rndkey0,$inout2
 293         $movkey         ($key),$rndkey0
 294
 295 .L${dir}_loop3:
 296         aes${dir}       $rndkey1,$inout0
 297         aes${dir}       $rndkey1,$inout1
 298         dec             $rounds
 299         aes${dir}       $rndkey1,$inout2
 300         $movkey         16($key),$rndkey1
 301         aes${dir}       $rndkey0,$inout0
 302         aes${dir}       $rndkey0,$inout1
 303         lea             32($key),$key
 304         aes${dir}       $rndkey0,$inout2
 305         $movkey         ($key),$rndkey0
 306         jnz             .L${dir}_loop3
 307
 308         aes${dir}       $rndkey1,$inout0
 309         aes${dir}       $rndkey1,$inout1
 310         aes${dir}       $rndkey1,$inout2
 311         aes${dir}last   $rndkey0,$inout0
 312         aes${dir}last   $rndkey0,$inout1
 313         aes${dir}last   $rndkey0,$inout2
 314         ret
 315 .size   _aesni_${dir}rypt3,.-_aesni_${dir}rypt3
 316 ___
 317 }
 318 # 4x interleave is implemented to improve small block performance,
 319 # most notably [and naturally] 4 block by ~30%. One can argue that one
 320 # should have implemented 5x as well, but improvement would be <20%,
 321 # so it's not worth it...
 322 sub aesni_generate4 {
 323 my $dir=shift;
 324 # As already mentioned it takes in $key and $rounds, which are *not*
 325 # preserved. $inout[0-3] is cipher/clear text...
 326 $code.=<<___;
 327 .type   _aesni_${dir}rypt4,\@abi-omnipotent
 328 .align  16
 329 _aesni_${dir}rypt4:
 330         $movkey ($key),$rndkey0
 331         shr     \$1,$rounds
 332         $movkey 16($key),$rndkey1
 333         lea     32($key),$key
 334         xorps   $rndkey0,$inout0
 335         xorps   $rndkey0,$inout1
 336         xorps   $rndkey0,$inout2
 337         xorps   $rndkey0,$inout3
 338         $movkey ($key),$rndkey0
 339
 340 .L${dir}_loop4:
 341         aes${dir}       $rndkey1,$inout0
 342         aes${dir}       $rndkey1,$inout1
 343         dec             $rounds
 344         aes${dir}       $rndkey1,$inout2
 345         aes${dir}       $rndkey1,$inout3
 346         $movkey         16($key),$rndkey1
 347         aes${dir}       $rndkey0,$inout0
 348         aes${dir}       $rndkey0,$inout1
 349         lea             32($key),$key
 350         aes${dir}       $rndkey0,$inout2
 351         aes${dir}       $rndkey0,$inout3
 352         $movkey         ($key),$rndkey0
 353         jnz             .L${dir}_loop4
 354
 355         aes${dir}       $rndkey1,$inout0
 356         aes${dir}       $rndkey1,$inout1
 357         aes${dir}       $rndkey1,$inout2
 358         aes${dir}       $rndkey1,$inout3
 359         aes${dir}last   $rndkey0,$inout0
 360         aes${dir}last   $rndkey0,$inout1
 361         aes${dir}last   $rndkey0,$inout2
 362         aes${dir}last   $rndkey0,$inout3
 363         ret
 364 .size   _aesni_${dir}rypt4,.-_aesni_${dir}rypt4
 365 ___
 366 }
 367 sub aesni_generate6 {
 368 my $dir=shift;
 369 # As already mentioned it takes in $key and $rounds, which are *not*
 370 # preserved. $inout[0-5] is cipher/clear text...
 371 $code.=<<___;
 372 .type   _aesni_${dir}rypt6,\@abi-omnipotent
 373 .align  16
 374 _aesni_${dir}rypt6:
 375         $movkey         ($key),$rndkey0
 376         shr             \$1,$rounds
 377         $movkey         16($key),$rndkey1
 378         lea             32($key),$key
 379         xorps           $rndkey0,$inout0
 380         pxor            $rndkey0,$inout1
 381         aes${dir}       $rndkey1,$inout0
 382         pxor            $rndkey0,$inout2
 383         aes${dir}       $rndkey1,$inout1
 384         pxor            $rndkey0,$inout3
 385         aes${dir}       $rndkey1,$inout2
 386         pxor            $rndkey0,$inout4
 387         aes${dir}       $rndkey1,$inout3
 388         pxor            $rndkey0,$inout5
 389         dec             $rounds
 390         aes${dir}       $rndkey1,$inout4
 391         $movkey         ($key),$rndkey0
 392         aes${dir}       $rndkey1,$inout5
 393         jmp             .L${dir}_loop6_enter
 394 .align  16
 395 .L${dir}_loop6:
 396         aes${dir}       $rndkey1,$inout0
 397         aes${dir}       $rndkey1,$inout1
 398         dec             $rounds
 399         aes${dir}       $rndkey1,$inout2
 400         aes${dir}       $rndkey1,$inout3
 401         aes${dir}       $rndkey1,$inout4
 402         aes${dir}       $rndkey1,$inout5
 403 .L${dir}_loop6_enter:                           # happens to be 16-byte aligned
 404         $movkey         16($key),$rndkey1
 405         aes${dir}       $rndkey0,$inout0
 406         aes${dir}       $rndkey0,$inout1
 407         lea             32($key),$key
 408         aes${dir}       $rndkey0,$inout2
 409         aes${dir}       $rndkey0,$inout3
 410         aes${dir}       $rndkey0,$inout4
 411         aes${dir}       $rndkey0,$inout5
 412         $movkey         ($key),$rndkey0
 413         jnz             .L${dir}_loop6
 414
 415         aes${dir}       $rndkey1,$inout0
 416         aes${dir}       $rndkey1,$inout1
 417         aes${dir}       $rndkey1,$inout2
 418         aes${dir}       $rndkey1,$inout3
 419         aes${dir}       $rndkey1,$inout4
 420         aes${dir}       $rndkey1,$inout5
 421         aes${dir}last   $rndkey0,$inout0
 422         aes${dir}last   $rndkey0,$inout1
 423         aes${dir}last   $rndkey0,$inout2
 424         aes${dir}last   $rndkey0,$inout3
 425         aes${dir}last   $rndkey0,$inout4
 426         aes${dir}last   $rndkey0,$inout5
 427         ret
 428 .size   _aesni_${dir}rypt6,.-_aesni_${dir}rypt6
 429 ___
 430 }
 431 sub aesni_generate8 {
 432 my $dir=shift;
 433 # As already mentioned it takes in $key and $rounds, which are *not*
 434 # preserved. $inout[0-7] is cipher/clear text...
 435 $code.=<<___;
 436 .type   _aesni_${dir}rypt8,\@abi-omnipotent
 437 .align  16
 438 _aesni_${dir}rypt8:
 439         $movkey         ($key),$rndkey0
 440         shr             \$1,$rounds
 441         $movkey         16($key),$rndkey1
 442         lea             32($key),$key
 443         xorps           $rndkey0,$inout0
 444         xorps           $rndkey0,$inout1
 445         aes${dir}       $rndkey1,$inout0
 446         pxor            $rndkey0,$inout2
 447         aes${dir}       $rndkey1,$inout1
 448         pxor            $rndkey0,$inout3
 449         aes${dir}       $rndkey1,$inout2
 450         pxor            $rndkey0,$inout4
 451         aes${dir}       $rndkey1,$inout3
 452         pxor            $rndkey0,$inout5
 453         dec             $rounds
 454         aes${dir}       $rndkey1,$inout4
 455         pxor            $rndkey0,$inout6
 456         aes${dir}       $rndkey1,$inout5
 457         pxor            $rndkey0,$inout7
 458         $movkey         ($key),$rndkey0
 459         aes${dir}       $rndkey1,$inout6
 460         aes${dir}       $rndkey1,$inout7
 461         $movkey         16($key),$rndkey1
 462         jmp             .L${dir}_loop8_enter
 463 .align  16
 464 .L${dir}_loop8:
 465         aes${dir}       $rndkey1,$inout0
 466         aes${dir}       $rndkey1,$inout1
 467         dec             $rounds
 468         aes${dir}       $rndkey1,$inout2
 469         aes${dir}       $rndkey1,$inout3
 470         aes${dir}       $rndkey1,$inout4
 471         aes${dir}       $rndkey1,$inout5
 472         aes${dir}       $rndkey1,$inout6
 473         aes${dir}       $rndkey1,$inout7
 474         $movkey         16($key),$rndkey1
 475 .L${dir}_loop8_enter:                           # happens to be 16-byte aligned
 476         aes${dir}       $rndkey0,$inout0
 477         aes${dir}       $rndkey0,$inout1
 478         lea             32($key),$key
 479         aes${dir}       $rndkey0,$inout2
 480         aes${dir}       $rndkey0,$inout3
 481         aes${dir}       $rndkey0,$inout4
 482         aes${dir}       $rndkey0,$inout5
 483         aes${dir}       $rndkey0,$inout6
 484         aes${dir}       $rndkey0,$inout7
 485         $movkey         ($key),$rndkey0
 486         jnz             .L${dir}_loop8
 487
 488         aes${dir}       $rndkey1,$inout0
 489         aes${dir}       $rndkey1,$inout1
 490         aes${dir}       $rndkey1,$inout2
 491         aes${dir}       $rndkey1,$inout3
 492         aes${dir}       $rndkey1,$inout4
 493         aes${dir}       $rndkey1,$inout5
 494         aes${dir}       $rndkey1,$inout6
 495         aes${dir}       $rndkey1,$inout7
 496         aes${dir}last   $rndkey0,$inout0
 497         aes${dir}last   $rndkey0,$inout1
 498         aes${dir}last   $rndkey0,$inout2
 499         aes${dir}last   $rndkey0,$inout3
 500         aes${dir}last   $rndkey0,$inout4
 501         aes${dir}last   $rndkey0,$inout5
 502         aes${dir}last   $rndkey0,$inout6
 503         aes${dir}last   $rndkey0,$inout7
 504         ret
 505 .size   _aesni_${dir}rypt8,.-_aesni_${dir}rypt8
 506 ___
 507 }
 508 &aesni_generate3("enc") if ($PREFIX eq "aesni");
 509 &aesni_generate3("dec");
 510 &aesni_generate4("enc") if ($PREFIX eq "aesni");
 511 &aesni_generate4("dec");
 512 &aesni_generate6("enc") if ($PREFIX eq "aesni");
 513 &aesni_generate6("dec");
 514 &aesni_generate8("enc") if ($PREFIX eq "aesni");
 515 &aesni_generate8("dec");
 516 \f
 517 if ($PREFIX eq "aesni") {
 518 ########################################################################
 519 # void aesni_ecb_encrypt (const void *in, void *out,
 520 #                         size_t length, const AES_KEY *key,
 521 #                         int enc);
 522 $code.=<<___;
 523 .globl  aesni_ecb_encrypt
 524 .type   aesni_ecb_encrypt,\@function,5
 525 .align  16
 526 aesni_ecb_encrypt:
 527         and     \$-16,$len
 528         jz      .Lecb_ret
 529
 530         mov     240($key),$rounds       # key->rounds
 531         $movkey ($key),$rndkey0
 532         mov     $key,$key_              # backup $key
 533         mov     $rounds,$rnds_          # backup $rounds
 534         test    %r8d,%r8d               # 5th argument
 535         jz      .Lecb_decrypt
 536 #--------------------------- ECB ENCRYPT ------------------------------#
 537         cmp     \$0x80,$len
 538         jb      .Lecb_enc_tail
 539
 540         movdqu  ($inp),$inout0
 541         movdqu  0x10($inp),$inout1
 542         movdqu  0x20($inp),$inout2
 543         movdqu  0x30($inp),$inout3
 544         movdqu  0x40($inp),$inout4
 545         movdqu  0x50($inp),$inout5
 546         movdqu  0x60($inp),$inout6
 547         movdqu  0x70($inp),$inout7
 548         lea     0x80($inp),$inp
 549         sub     \$0x80,$len
 550         jmp     .Lecb_enc_loop8_enter
 551 .align 16
 552 .Lecb_enc_loop8:
 553         movups  $inout0,($out)
 554         mov     $key_,$key              # restore $key
 555         movdqu  ($inp),$inout0
 556         mov     $rnds_,$rounds          # restore $rounds
 557         movups  $inout1,0x10($out)
 558         movdqu  0x10($inp),$inout1
 559         movups  $inout2,0x20($out)
 560         movdqu  0x20($inp),$inout2
 561         movups  $inout3,0x30($out)
 562         movdqu  0x30($inp),$inout3
 563         movups  $inout4,0x40($out)
 564         movdqu  0x40($inp),$inout4
 565         movups  $inout5,0x50($out)
 566         movdqu  0x50($inp),$inout5
 567         movups  $inout6,0x60($out)
 568         movdqu  0x60($inp),$inout6
 569         movups  $inout7,0x70($out)
 570         lea     0x80($out),$out
 571         movdqu  0x70($inp),$inout7
 572         lea     0x80($inp),$inp
 573 .Lecb_enc_loop8_enter:
 574
 575         call    _aesni_encrypt8
 576
 577         sub     \$0x80,$len
 578         jnc     .Lecb_enc_loop8
 579
 580         movups  $inout0,($out)
 581         mov     $key_,$key              # restore $key
 582         movups  $inout1,0x10($out)
 583         mov     $rnds_,$rounds          # restore $rounds
 584         movups  $inout2,0x20($out)
 585         movups  $inout3,0x30($out)
 586         movups  $inout4,0x40($out)
 587         movups  $inout5,0x50($out)
 588         movups  $inout6,0x60($out)
 589         movups  $inout7,0x70($out)
 590         lea     0x80($out),$out
 591         add     \$0x80,$len
 592         jz      .Lecb_ret
 593
 594 .Lecb_enc_tail:
 595         movups  ($inp),$inout0
 596         cmp     \$0x20,$len
 597         jb      .Lecb_enc_one
 598         movups  0x10($inp),$inout1
 599         je      .Lecb_enc_two
 600         movups  0x20($inp),$inout2
 601         cmp     \$0x40,$len
 602         jb      .Lecb_enc_three
 603         movups  0x30($inp),$inout3
 604         je      .Lecb_enc_four
 605         movups  0x40($inp),$inout4
 606         cmp     \$0x60,$len
 607         jb      .Lecb_enc_five
 608         movups  0x50($inp),$inout5
 609         je      .Lecb_enc_six
 610         movdqu  0x60($inp),$inout6
 611         call    _aesni_encrypt8
 612         movups  $inout0,($out)
 613         movups  $inout1,0x10($out)
 614         movups  $inout2,0x20($out)
 615         movups  $inout3,0x30($out)
 616         movups  $inout4,0x40($out)
 617         movups  $inout5,0x50($out)
 618         movups  $inout6,0x60($out)
 619         jmp     .Lecb_ret
 620 .align  16
 621 .Lecb_enc_one:
 622 ___
 623         &aesni_generate1("enc",$key,$rounds);
 624 $code.=<<___;
 625         movups  $inout0,($out)
 626         jmp     .Lecb_ret
 627 .align  16
 628 .Lecb_enc_two:
 629         xorps   $inout2,$inout2
 630         call    _aesni_encrypt3
 631         movups  $inout0,($out)
 632         movups  $inout1,0x10($out)
 633         jmp     .Lecb_ret
 634 .align  16
 635 .Lecb_enc_three:
 636         call    _aesni_encrypt3
 637         movups  $inout0,($out)
 638         movups  $inout1,0x10($out)
 639         movups  $inout2,0x20($out)
 640         jmp     .Lecb_ret
 641 .align  16
 642 .Lecb_enc_four:
 643         call    _aesni_encrypt4
 644         movups  $inout0,($out)
 645         movups  $inout1,0x10($out)
 646         movups  $inout2,0x20($out)
 647         movups  $inout3,0x30($out)
 648         jmp     .Lecb_ret
 649 .align  16
 650 .Lecb_enc_five:
 651         xorps   $inout5,$inout5
 652         call    _aesni_encrypt6
 653         movups  $inout0,($out)
 654         movups  $inout1,0x10($out)
 655         movups  $inout2,0x20($out)
 656         movups  $inout3,0x30($out)
 657         movups  $inout4,0x40($out)
 658         jmp     .Lecb_ret
 659 .align  16
 660 .Lecb_enc_six:
 661         call    _aesni_encrypt6
 662         movups  $inout0,($out)
 663         movups  $inout1,0x10($out)
 664         movups  $inout2,0x20($out)
 665         movups  $inout3,0x30($out)
 666         movups  $inout4,0x40($out)
 667         movups  $inout5,0x50($out)
 668         jmp     .Lecb_ret
 669 \f#--------------------------- ECB DECRYPT ------------------------------#
 670 .align  16
 671 .Lecb_decrypt:
 672         cmp     \$0x80,$len
 673         jb      .Lecb_dec_tail
 674
 675         movdqu  ($inp),$inout0
 676         movdqu  0x10($inp),$inout1
 677         movdqu  0x20($inp),$inout2
 678         movdqu  0x30($inp),$inout3
 679         movdqu  0x40($inp),$inout4
 680         movdqu  0x50($inp),$inout5
 681         movdqu  0x60($inp),$inout6
 682         movdqu  0x70($inp),$inout7
 683         lea     0x80($inp),$inp
 684         sub     \$0x80,$len
 685         jmp     .Lecb_dec_loop8_enter
 686 .align 16
 687 .Lecb_dec_loop8:
 688         movups  $inout0,($out)
 689         mov     $key_,$key              # restore $key
 690         movdqu  ($inp),$inout0
 691         mov     $rnds_,$rounds          # restore $rounds
 692         movups  $inout1,0x10($out)
 693         movdqu  0x10($inp),$inout1
 694         movups  $inout2,0x20($out)
 695         movdqu  0x20($inp),$inout2
 696         movups  $inout3,0x30($out)
 697         movdqu  0x30($inp),$inout3
 698         movups  $inout4,0x40($out)
 699         movdqu  0x40($inp),$inout4
 700         movups  $inout5,0x50($out)
 701         movdqu  0x50($inp),$inout5
 702         movups  $inout6,0x60($out)
 703         movdqu  0x60($inp),$inout6
 704         movups  $inout7,0x70($out)
 705         lea     0x80($out),$out
 706         movdqu  0x70($inp),$inout7
 707         lea     0x80($inp),$inp
 708 .Lecb_dec_loop8_enter:
 709
 710         call    _aesni_decrypt8
 711
 712         $movkey ($key_),$rndkey0
 713         sub     \$0x80,$len
 714         jnc     .Lecb_dec_loop8
 715
 716         movups  $inout0,($out)
 717         mov     $key_,$key              # restore $key
 718         movups  $inout1,0x10($out)
 719         mov     $rnds_,$rounds          # restore $rounds
 720         movups  $inout2,0x20($out)
 721         movups  $inout3,0x30($out)
 722         movups  $inout4,0x40($out)
 723         movups  $inout5,0x50($out)
 724         movups  $inout6,0x60($out)
 725         movups  $inout7,0x70($out)
 726         lea     0x80($out),$out
 727         add     \$0x80,$len
 728         jz      .Lecb_ret
 729
 730 .Lecb_dec_tail:
 731         movups  ($inp),$inout0
 732         cmp     \$0x20,$len
 733         jb      .Lecb_dec_one
 734         movups  0x10($inp),$inout1
 735         je      .Lecb_dec_two
 736         movups  0x20($inp),$inout2
 737         cmp     \$0x40,$len
 738         jb      .Lecb_dec_three
 739         movups  0x30($inp),$inout3
 740         je      .Lecb_dec_four
 741         movups  0x40($inp),$inout4
 742         cmp     \$0x60,$len
 743         jb      .Lecb_dec_five
 744         movups  0x50($inp),$inout5
 745         je      .Lecb_dec_six
 746         movups  0x60($inp),$inout6
 747         $movkey ($key),$rndkey0
 748         call    _aesni_decrypt8
 749         movups  $inout0,($out)
 750         movups  $inout1,0x10($out)
 751         movups  $inout2,0x20($out)
 752         movups  $inout3,0x30($out)
 753         movups  $inout4,0x40($out)
 754         movups  $inout5,0x50($out)
 755         movups  $inout6,0x60($out)
 756         jmp     .Lecb_ret
 757 .align  16
 758 .Lecb_dec_one:
 759 ___
 760         &aesni_generate1("dec",$key,$rounds);
 761 $code.=<<___;
 762         movups  $inout0,($out)
 763         jmp     .Lecb_ret
 764 .align  16
 765 .Lecb_dec_two:
 766         xorps   $inout2,$inout2
 767         call    _aesni_decrypt3
 768         movups  $inout0,($out)
 769         movups  $inout1,0x10($out)
 770         jmp     .Lecb_ret
 771 .align  16
 772 .Lecb_dec_three:
 773         call    _aesni_decrypt3
 774         movups  $inout0,($out)
 775         movups  $inout1,0x10($out)
 776         movups  $inout2,0x20($out)
 777         jmp     .Lecb_ret
 778 .align  16
 779 .Lecb_dec_four:
 780         call    _aesni_decrypt4
 781         movups  $inout0,($out)
 782         movups  $inout1,0x10($out)
 783         movups  $inout2,0x20($out)
 784         movups  $inout3,0x30($out)
 785         jmp     .Lecb_ret
 786 .align  16
 787 .Lecb_dec_five:
 788         xorps   $inout5,$inout5
 789         call    _aesni_decrypt6
 790         movups  $inout0,($out)
 791         movups  $inout1,0x10($out)
 792         movups  $inout2,0x20($out)
 793         movups  $inout3,0x30($out)
 794         movups  $inout4,0x40($out)
 795         jmp     .Lecb_ret
 796 .align  16
 797 .Lecb_dec_six:
 798         call    _aesni_decrypt6
 799         movups  $inout0,($out)
 800         movups  $inout1,0x10($out)
 801         movups  $inout2,0x20($out)
 802         movups  $inout3,0x30($out)
 803         movups  $inout4,0x40($out)
 804         movups  $inout5,0x50($out)
 805
 806 .Lecb_ret:
 807         ret
 808 .size   aesni_ecb_encrypt,.-aesni_ecb_encrypt
 809 ___
 810 \f
 811 {
 812 ######################################################################
 813 # void aesni_ccm64_[en|de]crypt_blocks (const void *in, void *out,
 814 #                         size_t blocks, const AES_KEY *key,
 815 #                         const char *ivec,char *cmac);
 816 #
 817 # Handles only complete blocks, operates on 64-bit counter and
 818 # does not update *ivec! Nor does it finalize CMAC value
 819 # (see engine/eng_aesni.c for details)
 820 #
 821 {
 822 my $cmac="%r9"; # 6th argument
 823
 824 my $increment="%xmm6";
 825 my $bswap_mask="%xmm7";
 826
 827 $code.=<<___;
 828 .globl  aesni_ccm64_encrypt_blocks
 829 .type   aesni_ccm64_encrypt_blocks,\@function,6
 830 .align  16
 831 aesni_ccm64_encrypt_blocks:
 832 ___
 833 $code.=<<___ if ($win64);
 834         lea     -0x58(%rsp),%rsp
 835         movaps  %xmm6,(%rsp)
 836         movaps  %xmm7,0x10(%rsp)
 837         movaps  %xmm8,0x20(%rsp)
 838         movaps  %xmm9,0x30(%rsp)
 839 .Lccm64_enc_body:
 840 ___
 841 $code.=<<___;
 842         mov     240($key),$rounds               # key->rounds
 843         movdqu  ($ivp),$iv
 844         movdqa  .Lincrement64(%rip),$increment
 845         movdqa  .Lbswap_mask(%rip),$bswap_mask
 846
 847         shr     \$1,$rounds
 848         lea     0($key),$key_
 849         movdqu  ($cmac),$inout1
 850         movdqa  $iv,$inout0
 851         mov     $rounds,$rnds_
 852         pshufb  $bswap_mask,$iv
 853         jmp     .Lccm64_enc_outer
 854 .align  16
 855 .Lccm64_enc_outer:
 856         $movkey ($key_),$rndkey0
 857         mov     $rnds_,$rounds
 858         movups  ($inp),$in0                     # load inp
 859
 860         xorps   $rndkey0,$inout0                # counter
 861         $movkey 16($key_),$rndkey1
 862         xorps   $in0,$rndkey0
 863         lea     32($key_),$key
 864         xorps   $rndkey0,$inout1                # cmac^=inp
 865         $movkey ($key),$rndkey0
 866
 867 .Lccm64_enc2_loop:
 868         aesenc  $rndkey1,$inout0
 869         dec     $rounds
 870         aesenc  $rndkey1,$inout1
 871         $movkey 16($key),$rndkey1
 872         aesenc  $rndkey0,$inout0
 873         lea     32($key),$key
 874         aesenc  $rndkey0,$inout1
 875         $movkey 0($key),$rndkey0
 876         jnz     .Lccm64_enc2_loop
 877         aesenc  $rndkey1,$inout0
 878         aesenc  $rndkey1,$inout1
 879         paddq   $increment,$iv
 880         aesenclast      $rndkey0,$inout0
 881         aesenclast      $rndkey0,$inout1
 882
 883         dec     $len
 884         lea     16($inp),$inp
 885         xorps   $inout0,$in0                    # inp ^= E(iv)
 886         movdqa  $iv,$inout0
 887         movups  $in0,($out)                     # save output
 888         lea     16($out),$out
 889         pshufb  $bswap_mask,$inout0
 890         jnz     .Lccm64_enc_outer
 891
 892         movups  $inout1,($cmac)
 893 ___
 894 $code.=<<___ if ($win64);
 895         movaps  (%rsp),%xmm6
 896         movaps  0x10(%rsp),%xmm7
 897         movaps  0x20(%rsp),%xmm8
 898         movaps  0x30(%rsp),%xmm9
 899         lea     0x58(%rsp),%rsp
 900 .Lccm64_enc_ret:
 901 ___
 902 $code.=<<___;
 903         ret
 904 .size   aesni_ccm64_encrypt_blocks,.-aesni_ccm64_encrypt_blocks
 905 ___
 906 ######################################################################
 907 $code.=<<___;
 908 .globl  aesni_ccm64_decrypt_blocks
 909 .type   aesni_ccm64_decrypt_blocks,\@function,6
 910 .align  16
 911 aesni_ccm64_decrypt_blocks:
 912 ___
 913 $code.=<<___ if ($win64);
 914         lea     -0x58(%rsp),%rsp
 915         movaps  %xmm6,(%rsp)
 916         movaps  %xmm7,0x10(%rsp)
 917         movaps  %xmm8,0x20(%rsp)
 918         movaps  %xmm9,0x30(%rsp)
 919 .Lccm64_dec_body:
 920 ___
 921 $code.=<<___;
 922         mov     240($key),$rounds               # key->rounds
 923         movups  ($ivp),$iv
 924         movdqu  ($cmac),$inout1
 925         movdqa  .Lincrement64(%rip),$increment
 926         movdqa  .Lbswap_mask(%rip),$bswap_mask
 927
 928         movaps  $iv,$inout0
 929         mov     $rounds,$rnds_
 930         mov     $key,$key_
 931         pshufb  $bswap_mask,$iv
 932 ___
 933         &aesni_generate1("enc",$key,$rounds);
 934 $code.=<<___;
 935         movups  ($inp),$in0                     # load inp
 936         paddq   $increment,$iv
 937         lea     16($inp),$inp
 938         jmp     .Lccm64_dec_outer
 939 .align  16
 940 .Lccm64_dec_outer:
 941         xorps   $inout0,$in0                    # inp ^= E(iv)
 942         movdqa  $iv,$inout0
 943         mov     $rnds_,$rounds
 944         movups  $in0,($out)                     # save output
 945         lea     16($out),$out
 946         pshufb  $bswap_mask,$inout0
 947
 948         sub     \$1,$len
 949         jz      .Lccm64_dec_break
 950
 951         $movkey ($key_),$rndkey0
 952         shr     \$1,$rounds
 953         $movkey 16($key_),$rndkey1
 954         xorps   $rndkey0,$in0
 955         lea     32($key_),$key
 956         xorps   $rndkey0,$inout0
 957         xorps   $in0,$inout1                    # cmac^=out
 958         $movkey ($key),$rndkey0
 959
 960 .Lccm64_dec2_loop:
 961         aesenc  $rndkey1,$inout0
 962         dec     $rounds
 963         aesenc  $rndkey1,$inout1
 964         $movkey 16($key),$rndkey1
 965         aesenc  $rndkey0,$inout0
 966         lea     32($key),$key
 967         aesenc  $rndkey0,$inout1
 968         $movkey 0($key),$rndkey0
 969         jnz     .Lccm64_dec2_loop
 970         movups  ($inp),$in0                     # load inp
 971         paddq   $increment,$iv
 972         aesenc  $rndkey1,$inout0
 973         aesenc  $rndkey1,$inout1
 974         lea     16($inp),$inp
 975         aesenclast      $rndkey0,$inout0
 976         aesenclast      $rndkey0,$inout1
 977         jmp     .Lccm64_dec_outer
 978
 979 .align  16
 980 .Lccm64_dec_break:
 981         #xorps  $in0,$inout1                    # cmac^=out
 982 ___
 983         &aesni_generate1("enc",$key_,$rounds,$inout1,$in0);
 984 $code.=<<___;
 985         movups  $inout1,($cmac)
 986 ___
 987 $code.=<<___ if ($win64);
 988         movaps  (%rsp),%xmm6
 989         movaps  0x10(%rsp),%xmm7
 990         movaps  0x20(%rsp),%xmm8
 991         movaps  0x30(%rsp),%xmm9
 992         lea     0x58(%rsp),%rsp
 993 .Lccm64_dec_ret:
 994 ___
 995 $code.=<<___;
 996         ret
 997 .size   aesni_ccm64_decrypt_blocks,.-aesni_ccm64_decrypt_blocks
 998 ___
 999 }\f
1000 ######################################################################
1001 # void aesni_ctr32_encrypt_blocks (const void *in, void *out,
1002 #                         size_t blocks, const AES_KEY *key,
1003 #                         const char *ivec);
1004 #
1005 # Handles only complete blocks, operates on 32-bit counter and
1006 # does not update *ivec! (see engine/eng_aesni.c for details)
1007 #
1008 {
1009 my $reserved = $win64?0:-0x28;
1010 my ($in0,$in1,$in2,$in3)=map("%xmm$_",(8..11));
1011 my ($iv0,$iv1,$ivec)=("%xmm12","%xmm13","%xmm14");
1012 my $bswap_mask="%xmm15";
1013
1014 $code.=<<___;
1015 .globl  aesni_ctr32_encrypt_blocks
1016 .type   aesni_ctr32_encrypt_blocks,\@function,5
1017 .align  16
1018 aesni_ctr32_encrypt_blocks:
1019 ___
1020 $code.=<<___ if ($win64);
1021         lea     -0xc8(%rsp),%rsp
1022         movaps  %xmm6,0x20(%rsp)
1023         movaps  %xmm7,0x30(%rsp)
1024         movaps  %xmm8,0x40(%rsp)
1025         movaps  %xmm9,0x50(%rsp)
1026         movaps  %xmm10,0x60(%rsp)
1027         movaps  %xmm11,0x70(%rsp)
1028         movaps  %xmm12,0x80(%rsp)
1029         movaps  %xmm13,0x90(%rsp)
1030         movaps  %xmm14,0xa0(%rsp)
1031         movaps  %xmm15,0xb0(%rsp)
1032 .Lctr32_body:
1033 ___
1034 $code.=<<___;
1035         cmp     \$1,$len
1036         je      .Lctr32_one_shortcut
1037
1038         movdqu  ($ivp),$ivec
1039         movdqa  .Lbswap_mask(%rip),$bswap_mask
1040         xor     $rounds,$rounds
1041         pextrd  \$3,$ivec,$rnds_                # pull 32-bit counter
1042         pinsrd  \$3,$rounds,$ivec               # wipe 32-bit counter
1043
1044         mov     240($key),$rounds               # key->rounds
1045         bswap   $rnds_
1046         pxor    $iv0,$iv0                       # vector of 3 32-bit counters
1047         pxor    $iv1,$iv1                       # vector of 3 32-bit counters
1048         pinsrd  \$0,$rnds_,$iv0
1049         lea     3($rnds_),$key_
1050         pinsrd  \$0,$key_,$iv1
1051         inc     $rnds_
1052         pinsrd  \$1,$rnds_,$iv0
1053         inc     $key_
1054         pinsrd  \$1,$key_,$iv1
1055         inc     $rnds_
1056         pinsrd  \$2,$rnds_,$iv0
1057         inc     $key_
1058         pinsrd  \$2,$key_,$iv1
1059         movdqa  $iv0,$reserved(%rsp)
1060         pshufb  $bswap_mask,$iv0
1061         movdqa  $iv1,`$reserved+0x10`(%rsp)
1062         pshufb  $bswap_mask,$iv1
1063
1064         pshufd  \$`3<<6`,$iv0,$inout0           # place counter to upper dword
1065         pshufd  \$`2<<6`,$iv0,$inout1
1066         pshufd  \$`1<<6`,$iv0,$inout2
1067         cmp     \$6,$len
1068         jb      .Lctr32_tail
1069         shr     \$1,$rounds
1070         mov     $key,$key_                      # backup $key
1071         mov     $rounds,$rnds_                  # backup $rounds
1072         sub     \$6,$len
1073         jmp     .Lctr32_loop6
1074
1075 .align  16
1076 .Lctr32_loop6:
1077         pshufd  \$`3<<6`,$iv1,$inout3
1078         por     $ivec,$inout0                   # merge counter-less ivec
1079          $movkey        ($key_),$rndkey0
1080         pshufd  \$`2<<6`,$iv1,$inout4
1081         por     $ivec,$inout1
1082          $movkey        16($key_),$rndkey1
1083         pshufd  \$`1<<6`,$iv1,$inout5
1084         por     $ivec,$inout2
1085         por     $ivec,$inout3
1086          xorps          $rndkey0,$inout0
1087         por     $ivec,$inout4
1088         por     $ivec,$inout5
1089
1090         # inline _aesni_encrypt6 and interleave last rounds
1091         # with own code...
1092
1093         pxor            $rndkey0,$inout1
1094         aesenc          $rndkey1,$inout0
1095         lea             32($key_),$key
1096         pxor            $rndkey0,$inout2
1097         aesenc          $rndkey1,$inout1
1098          movdqa         .Lincrement32(%rip),$iv1
1099         pxor            $rndkey0,$inout3
1100         aesenc          $rndkey1,$inout2
1101          movdqa         $reserved(%rsp),$iv0
1102         pxor            $rndkey0,$inout4
1103         aesenc          $rndkey1,$inout3
1104         pxor            $rndkey0,$inout5
1105         $movkey         ($key),$rndkey0
1106         dec             $rounds
1107         aesenc          $rndkey1,$inout4
1108         aesenc          $rndkey1,$inout5
1109         jmp             .Lctr32_enc_loop6_enter
1110 .align  16
1111 .Lctr32_enc_loop6:
1112         aesenc          $rndkey1,$inout0
1113         aesenc          $rndkey1,$inout1
1114         dec             $rounds
1115         aesenc          $rndkey1,$inout2
1116         aesenc          $rndkey1,$inout3
1117         aesenc          $rndkey1,$inout4
1118         aesenc          $rndkey1,$inout5
1119 .Lctr32_enc_loop6_enter:
1120         $movkey         16($key),$rndkey1
1121         aesenc          $rndkey0,$inout0
1122         aesenc          $rndkey0,$inout1
1123         lea             32($key),$key
1124         aesenc          $rndkey0,$inout2
1125         aesenc          $rndkey0,$inout3
1126         aesenc          $rndkey0,$inout4
1127         aesenc          $rndkey0,$inout5
1128         $movkey         ($key),$rndkey0
1129         jnz             .Lctr32_enc_loop6
1130
1131         aesenc          $rndkey1,$inout0
1132          paddd          $iv1,$iv0               # increment counter vector
1133         aesenc          $rndkey1,$inout1
1134          paddd          `$reserved+0x10`(%rsp),$iv1
1135         aesenc          $rndkey1,$inout2
1136          movdqa         $iv0,$reserved(%rsp)    # save counter vector
1137         aesenc          $rndkey1,$inout3
1138          movdqa         $iv1,`$reserved+0x10`(%rsp)
1139         aesenc          $rndkey1,$inout4
1140          pshufb         $bswap_mask,$iv0        # byte swap
1141         aesenc          $rndkey1,$inout5
1142          pshufb         $bswap_mask,$iv1
1143
1144         aesenclast      $rndkey0,$inout0
1145          movups         ($inp),$in0             # load input
1146         aesenclast      $rndkey0,$inout1
1147          movups         0x10($inp),$in1
1148         aesenclast      $rndkey0,$inout2
1149          movups         0x20($inp),$in2
1150         aesenclast      $rndkey0,$inout3
1151          movups         0x30($inp),$in3
1152         aesenclast      $rndkey0,$inout4
1153          movups         0x40($inp),$rndkey1
1154         aesenclast      $rndkey0,$inout5
1155          movups         0x50($inp),$rndkey0
1156          lea    0x60($inp),$inp
1157
1158         xorps   $inout0,$in0                    # xor
1159          pshufd \$`3<<6`,$iv0,$inout0
1160         xorps   $inout1,$in1
1161          pshufd \$`2<<6`,$iv0,$inout1
1162         movups  $in0,($out)                     # store output
1163         xorps   $inout2,$in2
1164          pshufd \$`1<<6`,$iv0,$inout2
1165         movups  $in1,0x10($out)
1166         xorps   $inout3,$in3
1167         movups  $in2,0x20($out)
1168         xorps   $inout4,$rndkey1
1169         movups  $in3,0x30($out)
1170         xorps   $inout5,$rndkey0
1171         movups  $rndkey1,0x40($out)
1172         movups  $rndkey0,0x50($out)
1173         lea     0x60($out),$out
1174         mov     $rnds_,$rounds
1175         sub     \$6,$len
1176         jnc     .Lctr32_loop6
1177
1178         add     \$6,$len
1179         jz      .Lctr32_done
1180         mov     $key_,$key                      # restore $key
1181         lea     1($rounds,$rounds),$rounds      # restore original value
1182
1183 .Lctr32_tail:
1184         por     $ivec,$inout0
1185         movups  ($inp),$in0
1186         cmp     \$2,$len
1187         jb      .Lctr32_one
1188
1189         por     $ivec,$inout1
1190         movups  0x10($inp),$in1
1191         je      .Lctr32_two
1192
1193         pshufd  \$`3<<6`,$iv1,$inout3
1194         por     $ivec,$inout2
1195         movups  0x20($inp),$in2
1196         cmp     \$4,$len
1197         jb      .Lctr32_three
1198
1199         pshufd  \$`2<<6`,$iv1,$inout4
1200         por     $ivec,$inout3
1201         movups  0x30($inp),$in3
1202         je      .Lctr32_four
1203
1204         por     $ivec,$inout4
1205         xorps   $inout5,$inout5
1206
1207         call    _aesni_encrypt6
1208
1209         movups  0x40($inp),$rndkey1
1210         xorps   $inout0,$in0
1211         xorps   $inout1,$in1
1212         movups  $in0,($out)
1213         xorps   $inout2,$in2
1214         movups  $in1,0x10($out)
1215         xorps   $inout3,$in3
1216         movups  $in2,0x20($out)
1217         xorps   $inout4,$rndkey1
1218         movups  $in3,0x30($out)
1219         movups  $rndkey1,0x40($out)
1220         jmp     .Lctr32_done
1221
1222 .align  16
1223 .Lctr32_one_shortcut:
1224         movups  ($ivp),$inout0
1225         movups  ($inp),$in0
1226         mov     240($key),$rounds               # key->rounds
1227 .Lctr32_one:
1228 ___
1229         &aesni_generate1("enc",$key,$rounds);
1230 $code.=<<___;
1231         xorps   $inout0,$in0
1232         movups  $in0,($out)
1233         jmp     .Lctr32_done
1234
1235 .align  16
1236 .Lctr32_two:
1237         xorps   $inout2,$inout2
1238         call    _aesni_encrypt3
1239         xorps   $inout0,$in0
1240         xorps   $inout1,$in1
1241         movups  $in0,($out)
1242         movups  $in1,0x10($out)
1243         jmp     .Lctr32_done
1244
1245 .align  16
1246 .Lctr32_three:
1247         call    _aesni_encrypt3
1248         xorps   $inout0,$in0
1249         xorps   $inout1,$in1
1250         movups  $in0,($out)
1251         xorps   $inout2,$in2
1252         movups  $in1,0x10($out)
1253         movups  $in2,0x20($out)
1254         jmp     .Lctr32_done
1255
1256 .align  16
1257 .Lctr32_four:
1258         call    _aesni_encrypt4
1259         xorps   $inout0,$in0
1260         xorps   $inout1,$in1
1261         movups  $in0,($out)
1262         xorps   $inout2,$in2
1263         movups  $in1,0x10($out)
1264         xorps   $inout3,$in3
1265         movups  $in2,0x20($out)
1266         movups  $in3,0x30($out)
1267
1268 .Lctr32_done:
1269 ___
1270 $code.=<<___ if ($win64);
1271         movaps  0x20(%rsp),%xmm6
1272         movaps  0x30(%rsp),%xmm7
1273         movaps  0x40(%rsp),%xmm8
1274         movaps  0x50(%rsp),%xmm9
1275         movaps  0x60(%rsp),%xmm10
1276         movaps  0x70(%rsp),%xmm11
1277         movaps  0x80(%rsp),%xmm12
1278         movaps  0x90(%rsp),%xmm13
1279         movaps  0xa0(%rsp),%xmm14
1280         movaps  0xb0(%rsp),%xmm15
1281         lea     0xc8(%rsp),%rsp
1282 .Lctr32_ret:
1283 ___
1284 $code.=<<___;
1285         ret
1286 .size   aesni_ctr32_encrypt_blocks,.-aesni_ctr32_encrypt_blocks
1287 ___
1288 }
1289 \f
1290 ######################################################################
1291 # void aesni_xts_[en|de]crypt(const char *inp,char *out,size_t len,
1292 #       const AES_KEY *key1, const AES_KEY *key2
1293 #       const unsigned char iv[16]);
1294 #
1295 {
1296 my @tweak=map("%xmm$_",(10..15));
1297 my ($twmask,$twres,$twtmp)=("%xmm8","%xmm9",@tweak[4]);
1298 my ($key2,$ivp,$len_)=("%r8","%r9","%r9");
1299 my $frame_size = 0x68 + ($win64?160:0);
1300
1301 $code.=<<___;
1302 .globl  aesni_xts_encrypt
1303 .type   aesni_xts_encrypt,\@function,6
1304 .align  16
1305 aesni_xts_encrypt:
1306         lea     -$frame_size(%rsp),%rsp
1307 ___
1308 $code.=<<___ if ($win64);
1309         movaps  %xmm6,0x60(%rsp)
1310         movaps  %xmm7,0x70(%rsp)
1311         movaps  %xmm8,0x80(%rsp)
1312         movaps  %xmm9,0x90(%rsp)
1313         movaps  %xmm10,0xa0(%rsp)
1314         movaps  %xmm11,0xb0(%rsp)
1315         movaps  %xmm12,0xc0(%rsp)
1316         movaps  %xmm13,0xd0(%rsp)
1317         movaps  %xmm14,0xe0(%rsp)
1318         movaps  %xmm15,0xf0(%rsp)
1319 .Lxts_enc_body:
1320 ___
1321 $code.=<<___;
1322         movups  ($ivp),@tweak[5]                # load clear-text tweak
1323         mov     240(%r8),$rounds                # key2->rounds
1324         mov     240($key),$rnds_                # key1->rounds
1325 ___
1326         # generate the tweak
1327         &aesni_generate1("enc",$key2,$rounds,@tweak[5]);
1328 $code.=<<___;
1329         mov     $key,$key_                      # backup $key
1330         mov     $rnds_,$rounds                  # backup $rounds
1331         mov     $len,$len_                      # backup $len
1332         and     \$-16,$len
1333
1334         movdqa  .Lxts_magic(%rip),$twmask
1335         pxor    $twtmp,$twtmp
1336         pcmpgtd @tweak[5],$twtmp                # broadcast upper bits
1337 ___
1338     for ($i=0;$i<4;$i++) {
1339     $code.=<<___;
1340         pshufd  \$0x13,$twtmp,$twres
1341         pxor    $twtmp,$twtmp
1342         movdqa  @tweak[5],@tweak[$i]
1343         paddq   @tweak[5],@tweak[5]             # psllq 1,$tweak
1344         pand    $twmask,$twres                  # isolate carry and residue
1345         pcmpgtd @tweak[5],$twtmp                # broadcat upper bits
1346         pxor    $twres,@tweak[5]
1347 ___
1348     }
1349 $code.=<<___;
1350         sub     \$16*6,$len
1351         jc      .Lxts_enc_short
1352
1353         shr     \$1,$rounds
1354         sub     \$1,$rounds
1355         mov     $rounds,$rnds_
1356         jmp     .Lxts_enc_grandloop
1357
1358 .align  16
1359 .Lxts_enc_grandloop:
1360         pshufd  \$0x13,$twtmp,$twres
1361         movdqa  @tweak[5],@tweak[4]
1362         paddq   @tweak[5],@tweak[5]             # psllq 1,$tweak
1363         movdqu  `16*0`($inp),$inout0            # load input
1364         pand    $twmask,$twres                  # isolate carry and residue
1365         movdqu  `16*1`($inp),$inout1
1366         pxor    $twres,@tweak[5]
1367
1368         movdqu  `16*2`($inp),$inout2
1369         pxor    @tweak[0],$inout0               # input^=tweak
1370         movdqu  `16*3`($inp),$inout3
1371         pxor    @tweak[1],$inout1
1372         movdqu  `16*4`($inp),$inout4
1373         pxor    @tweak[2],$inout2
1374         movdqu  `16*5`($inp),$inout5
1375         lea     `16*6`($inp),$inp
1376         pxor    @tweak[3],$inout3
1377         $movkey         ($key_),$rndkey0
1378         pxor    @tweak[4],$inout4
1379         pxor    @tweak[5],$inout5
1380
1381         # inline _aesni_encrypt6 and interleave first and last rounds
1382         # with own code...
1383         $movkey         16($key_),$rndkey1
1384         pxor            $rndkey0,$inout0
1385         pxor            $rndkey0,$inout1
1386          movdqa @tweak[0],`16*0`(%rsp)          # put aside tweaks
1387         aesenc          $rndkey1,$inout0
1388         lea             32($key_),$key
1389         pxor            $rndkey0,$inout2
1390          movdqa @tweak[1],`16*1`(%rsp)
1391         aesenc          $rndkey1,$inout1
1392         pxor            $rndkey0,$inout3
1393          movdqa @tweak[2],`16*2`(%rsp)
1394         aesenc          $rndkey1,$inout2
1395         pxor            $rndkey0,$inout4
1396          movdqa @tweak[3],`16*3`(%rsp)
1397         aesenc          $rndkey1,$inout3
1398         pxor            $rndkey0,$inout5
1399         $movkey         ($key),$rndkey0
1400         dec             $rounds
1401          movdqa @tweak[4],`16*4`(%rsp)
1402         aesenc          $rndkey1,$inout4
1403          movdqa @tweak[5],`16*5`(%rsp)
1404         aesenc          $rndkey1,$inout5
1405         pxor    $twtmp,$twtmp
1406         pcmpgtd @tweak[5],$twtmp
1407         jmp             .Lxts_enc_loop6_enter
1408
1409 .align  16
1410 .Lxts_enc_loop6:
1411         aesenc          $rndkey1,$inout0
1412         aesenc          $rndkey1,$inout1
1413         dec             $rounds
1414         aesenc          $rndkey1,$inout2
1415         aesenc          $rndkey1,$inout3
1416         aesenc          $rndkey1,$inout4
1417         aesenc          $rndkey1,$inout5
1418 .Lxts_enc_loop6_enter:
1419         $movkey         16($key),$rndkey1
1420         aesenc          $rndkey0,$inout0
1421         aesenc          $rndkey0,$inout1
1422         lea             32($key),$key
1423         aesenc          $rndkey0,$inout2
1424         aesenc          $rndkey0,$inout3
1425         aesenc          $rndkey0,$inout4
1426         aesenc          $rndkey0,$inout5
1427         $movkey         ($key),$rndkey0
1428         jnz             .Lxts_enc_loop6
1429
1430         pshufd  \$0x13,$twtmp,$twres
1431         pxor    $twtmp,$twtmp
1432         paddq   @tweak[5],@tweak[5]             # psllq 1,$tweak
1433          aesenc         $rndkey1,$inout0
1434         pand    $twmask,$twres                  # isolate carry and residue
1435          aesenc         $rndkey1,$inout1
1436         pcmpgtd @tweak[5],$twtmp                # broadcast upper bits
1437          aesenc         $rndkey1,$inout2
1438         pxor    $twres,@tweak[5]
1439          aesenc         $rndkey1,$inout3
1440          aesenc         $rndkey1,$inout4
1441          aesenc         $rndkey1,$inout5
1442          $movkey        16($key),$rndkey1
1443
1444         pshufd  \$0x13,$twtmp,$twres
1445         pxor    $twtmp,$twtmp
1446         movdqa  @tweak[5],@tweak[0]
1447         paddq   @tweak[5],@tweak[5]             # psllq 1,$tweak
1448          aesenc         $rndkey0,$inout0
1449         pand    $twmask,$twres                  # isolate carry and residue
1450          aesenc         $rndkey0,$inout1
1451         pcmpgtd @tweak[5],$twtmp                # broadcat upper bits
1452          aesenc         $rndkey0,$inout2
1453         pxor    $twres,@tweak[5]
1454          aesenc         $rndkey0,$inout3
1455          aesenc         $rndkey0,$inout4
1456          aesenc         $rndkey0,$inout5
1457          $movkey        32($key),$rndkey0
1458
1459         pshufd  \$0x13,$twtmp,$twres
1460         pxor    $twtmp,$twtmp
1461         movdqa  @tweak[5],@tweak[1]
1462         paddq   @tweak[5],@tweak[5]             # psllq 1,$tweak
1463          aesenc         $rndkey1,$inout0
1464         pand    $twmask,$twres                  # isolate carry and residue
1465          aesenc         $rndkey1,$inout1
1466         pcmpgtd @tweak[5],$twtmp                # broadcat upper bits
1467          aesenc         $rndkey1,$inout2
1468         pxor    $twres,@tweak[5]
1469          aesenc         $rndkey1,$inout3
1470          aesenc         $rndkey1,$inout4
1471          aesenc         $rndkey1,$inout5
1472
1473         pshufd  \$0x13,$twtmp,$twres
1474         pxor    $twtmp,$twtmp
1475         movdqa  @tweak[5],@tweak[2]
1476         paddq   @tweak[5],@tweak[5]             # psllq 1,$tweak
1477          aesenclast     $rndkey0,$inout0
1478         pand    $twmask,$twres                  # isolate carry and residue
1479          aesenclast     $rndkey0,$inout1
1480         pcmpgtd @tweak[5],$twtmp                # broadcat upper bits
1481          aesenclast     $rndkey0,$inout2
1482         pxor    $twres,@tweak[5]
1483          aesenclast     $rndkey0,$inout3
1484          aesenclast     $rndkey0,$inout4
1485          aesenclast     $rndkey0,$inout5
1486
1487         pshufd  \$0x13,$twtmp,$twres
1488         pxor    $twtmp,$twtmp
1489         movdqa  @tweak[5],@tweak[3]
1490         paddq   @tweak[5],@tweak[5]             # psllq 1,$tweak
1491          xorps  `16*0`(%rsp),$inout0            # output^=tweak
1492         pand    $twmask,$twres                  # isolate carry and residue
1493          xorps  `16*1`(%rsp),$inout1
1494         pcmpgtd @tweak[5],$twtmp                # broadcat upper bits
1495         pxor    $twres,@tweak[5]
1496
1497         xorps   `16*2`(%rsp),$inout2
1498         movups  $inout0,`16*0`($out)            # write output
1499         xorps   `16*3`(%rsp),$inout3
1500         movups  $inout1,`16*1`($out)
1501         xorps   `16*4`(%rsp),$inout4
1502         movups  $inout2,`16*2`($out)
1503         xorps   `16*5`(%rsp),$inout5
1504         movups  $inout3,`16*3`($out)
1505         mov     $rnds_,$rounds                  # restore $rounds
1506         movups  $inout4,`16*4`($out)
1507         movups  $inout5,`16*5`($out)
1508         lea     `16*6`($out),$out
1509         sub     \$16*6,$len
1510         jnc     .Lxts_enc_grandloop
1511
1512         lea     3($rounds,$rounds),$rounds      # restore original value
1513         mov     $key_,$key                      # restore $key
1514         mov     $rounds,$rnds_                  # backup $rounds
1515
1516 .Lxts_enc_short:
1517         add     \$16*6,$len
1518         jz      .Lxts_enc_done
1519
1520         cmp     \$0x20,$len
1521         jb      .Lxts_enc_one
1522         je      .Lxts_enc_two
1523
1524         cmp     \$0x40,$len
1525         jb      .Lxts_enc_three
1526         je      .Lxts_enc_four
1527
1528         pshufd  \$0x13,$twtmp,$twres
1529         movdqa  @tweak[5],@tweak[4]
1530         paddq   @tweak[5],@tweak[5]             # psllq 1,$tweak
1531          movdqu ($inp),$inout0
1532         pand    $twmask,$twres                  # isolate carry and residue
1533          movdqu 16*1($inp),$inout1
1534         pxor    $twres,@tweak[5]
1535
1536         movdqu  16*2($inp),$inout2
1537         pxor    @tweak[0],$inout0
1538         movdqu  16*3($inp),$inout3
1539         pxor    @tweak[1],$inout1
1540         movdqu  16*4($inp),$inout4
1541         lea     16*5($inp),$inp
1542         pxor    @tweak[2],$inout2
1543         pxor    @tweak[3],$inout3
1544         pxor    @tweak[4],$inout4
1545
1546         call    _aesni_encrypt6
1547
1548         xorps   @tweak[0],$inout0
1549         movdqa  @tweak[5],@tweak[0]
1550         xorps   @tweak[1],$inout1
1551         xorps   @tweak[2],$inout2
1552         movdqu  $inout0,($out)
1553         xorps   @tweak[3],$inout3
1554         movdqu  $inout1,16*1($out)
1555         xorps   @tweak[4],$inout4
1556         movdqu  $inout2,16*2($out)
1557         movdqu  $inout3,16*3($out)
1558         movdqu  $inout4,16*4($out)
1559         lea     16*5($out),$out
1560         jmp     .Lxts_enc_done
1561
1562 .align  16
1563 .Lxts_enc_one:
1564         movups  ($inp),$inout0
1565         lea     16*1($inp),$inp
1566         xorps   @tweak[0],$inout0
1567 ___
1568         &aesni_generate1("enc",$key,$rounds);
1569 $code.=<<___;
1570         xorps   @tweak[0],$inout0
1571         movdqa  @tweak[1],@tweak[0]
1572         movups  $inout0,($out)
1573         lea     16*1($out),$out
1574         jmp     .Lxts_enc_done
1575
1576 .align  16
1577 .Lxts_enc_two:
1578         movups  ($inp),$inout0
1579         movups  16($inp),$inout1
1580         lea     32($inp),$inp
1581         xorps   @tweak[0],$inout0
1582         xorps   @tweak[1],$inout1
1583
1584         call    _aesni_encrypt3
1585
1586         xorps   @tweak[0],$inout0
1587         movdqa  @tweak[2],@tweak[0]
1588         xorps   @tweak[1],$inout1
1589         movups  $inout0,($out)
1590         movups  $inout1,16*1($out)
1591         lea     16*2($out),$out
1592         jmp     .Lxts_enc_done
1593
1594 .align  16
1595 .Lxts_enc_three:
1596         movups  ($inp),$inout0
1597         movups  16*1($inp),$inout1
1598         movups  16*2($inp),$inout2
1599         lea     16*3($inp),$inp
1600         xorps   @tweak[0],$inout0
1601         xorps   @tweak[1],$inout1
1602         xorps   @tweak[2],$inout2
1603
1604         call    _aesni_encrypt3
1605
1606         xorps   @tweak[0],$inout0
1607         movdqa  @tweak[3],@tweak[0]
1608         xorps   @tweak[1],$inout1
1609         xorps   @tweak[2],$inout2
1610         movups  $inout0,($out)
1611         movups  $inout1,16*1($out)
1612         movups  $inout2,16*2($out)
1613         lea     16*3($out),$out
1614         jmp     .Lxts_enc_done
1615
1616 .align  16
1617 .Lxts_enc_four:
1618         movups  ($inp),$inout0
1619         movups  16*1($inp),$inout1
1620         movups  16*2($inp),$inout2
1621         xorps   @tweak[0],$inout0
1622         movups  16*3($inp),$inout3
1623         lea     16*4($inp),$inp
1624         xorps   @tweak[1],$inout1
1625         xorps   @tweak[2],$inout2
1626         xorps   @tweak[3],$inout3
1627
1628         call    _aesni_encrypt4
1629
1630         xorps   @tweak[0],$inout0
1631         movdqa  @tweak[5],@tweak[0]
1632         xorps   @tweak[1],$inout1
1633         xorps   @tweak[2],$inout2
1634         movups  $inout0,($out)
1635         xorps   @tweak[3],$inout3
1636         movups  $inout1,16*1($out)
1637         movups  $inout2,16*2($out)
1638         movups  $inout3,16*3($out)
1639         lea     16*4($out),$out
1640         jmp     .Lxts_enc_done
1641
1642 .align  16
1643 .Lxts_enc_done:
1644         and     \$15,$len_
1645         jz      .Lxts_enc_ret
1646         mov     $len_,$len
1647
1648 .Lxts_enc_steal:
1649         movzb   ($inp),%eax                     # borrow $rounds ...
1650         movzb   -16($out),%ecx                  # ... and $key
1651         lea     1($inp),$inp
1652         mov     %al,-16($out)
1653         mov     %cl,0($out)
1654         lea     1($out),$out
1655         sub     \$1,$len
1656         jnz     .Lxts_enc_steal
1657
1658         sub     $len_,$out                      # rewind $out
1659         mov     $key_,$key                      # restore $key
1660         mov     $rnds_,$rounds                  # restore $rounds
1661
1662         movups  -16($out),$inout0
1663         xorps   @tweak[0],$inout0
1664 ___
1665         &aesni_generate1("enc",$key,$rounds);
1666 $code.=<<___;
1667         xorps   @tweak[0],$inout0
1668         movups  $inout0,-16($out)
1669
1670 .Lxts_enc_ret:
1671 ___
1672 $code.=<<___ if ($win64);
1673         movaps  0x60(%rsp),%xmm6
1674         movaps  0x70(%rsp),%xmm7
1675         movaps  0x80(%rsp),%xmm8
1676         movaps  0x90(%rsp),%xmm9
1677         movaps  0xa0(%rsp),%xmm10
1678         movaps  0xb0(%rsp),%xmm11
1679         movaps  0xc0(%rsp),%xmm12
1680         movaps  0xd0(%rsp),%xmm13
1681         movaps  0xe0(%rsp),%xmm14
1682         movaps  0xf0(%rsp),%xmm15
1683 ___
1684 $code.=<<___;
1685         lea     $frame_size(%rsp),%rsp
1686 .Lxts_enc_epilogue:
1687         ret
1688 .size   aesni_xts_encrypt,.-aesni_xts_encrypt
1689 ___
1690
1691 $code.=<<___;
1692 .globl  aesni_xts_decrypt
1693 .type   aesni_xts_decrypt,\@function,6
1694 .align  16
1695 aesni_xts_decrypt:
1696         lea     -$frame_size(%rsp),%rsp
1697 ___
1698 $code.=<<___ if ($win64);
1699         movaps  %xmm6,0x60(%rsp)
1700         movaps  %xmm7,0x70(%rsp)
1701         movaps  %xmm8,0x80(%rsp)
1702         movaps  %xmm9,0x90(%rsp)
1703         movaps  %xmm10,0xa0(%rsp)
1704         movaps  %xmm11,0xb0(%rsp)
1705         movaps  %xmm12,0xc0(%rsp)
1706         movaps  %xmm13,0xd0(%rsp)
1707         movaps  %xmm14,0xe0(%rsp)
1708         movaps  %xmm15,0xf0(%rsp)
1709 .Lxts_dec_body:
1710 ___
1711 $code.=<<___;
1712         movups  ($ivp),@tweak[5]                # load clear-text tweak
1713         mov     240($key2),$rounds              # key2->rounds
1714         mov     240($key),$rnds_                # key1->rounds
1715 ___
1716         # generate the tweak
1717         &aesni_generate1("enc",$key2,$rounds,@tweak[5]);
1718 $code.=<<___;
1719         xor     %eax,%eax                       # if ($len%16) len-=16;
1720         test    \$15,$len
1721         setnz   %al
1722         shl     \$4,%rax
1723         sub     %rax,$len
1724
1725         mov     $key,$key_                      # backup $key
1726         mov     $rnds_,$rounds                  # backup $rounds
1727         mov     $len,$len_                      # backup $len
1728         and     \$-16,$len
1729
1730         movdqa  .Lxts_magic(%rip),$twmask
1731         pxor    $twtmp,$twtmp
1732         pcmpgtd @tweak[5],$twtmp                # broadcast upper bits
1733 ___
1734     for ($i=0;$i<4;$i++) {
1735     $code.=<<___;
1736         pshufd  \$0x13,$twtmp,$twres
1737         pxor    $twtmp,$twtmp
1738         movdqa  @tweak[5],@tweak[$i]
1739         paddq   @tweak[5],@tweak[5]             # psllq 1,$tweak
1740         pand    $twmask,$twres                  # isolate carry and residue
1741         pcmpgtd @tweak[5],$twtmp                # broadcat upper bits
1742         pxor    $twres,@tweak[5]
1743 ___
1744     }
1745 $code.=<<___;
1746         sub     \$16*6,$len
1747         jc      .Lxts_dec_short
1748
1749         shr     \$1,$rounds
1750         sub     \$1,$rounds
1751         mov     $rounds,$rnds_
1752         jmp     .Lxts_dec_grandloop
1753
1754 .align  16
1755 .Lxts_dec_grandloop:
1756         pshufd  \$0x13,$twtmp,$twres
1757         movdqa  @tweak[5],@tweak[4]
1758         paddq   @tweak[5],@tweak[5]             # psllq 1,$tweak
1759         movdqu  `16*0`($inp),$inout0            # load input
1760         pand    $twmask,$twres                  # isolate carry and residue
1761         movdqu  `16*1`($inp),$inout1
1762         pxor    $twres,@tweak[5]
1763
1764         movdqu  `16*2`($inp),$inout2
1765         pxor    @tweak[0],$inout0               # input^=tweak
1766         movdqu  `16*3`($inp),$inout3
1767         pxor    @tweak[1],$inout1
1768         movdqu  `16*4`($inp),$inout4
1769         pxor    @tweak[2],$inout2
1770         movdqu  `16*5`($inp),$inout5
1771         lea     `16*6`($inp),$inp
1772         pxor    @tweak[3],$inout3
1773         $movkey         ($key_),$rndkey0
1774         pxor    @tweak[4],$inout4
1775         pxor    @tweak[5],$inout5
1776
1777         # inline _aesni_decrypt6 and interleave first and last rounds
1778         # with own code...
1779         $movkey         16($key_),$rndkey1
1780         pxor            $rndkey0,$inout0
1781         pxor            $rndkey0,$inout1
1782          movdqa @tweak[0],`16*0`(%rsp)          # put aside tweaks
1783         aesdec          $rndkey1,$inout0
1784         lea             32($key_),$key
1785         pxor            $rndkey0,$inout2
1786          movdqa @tweak[1],`16*1`(%rsp)
1787         aesdec          $rndkey1,$inout1
1788         pxor            $rndkey0,$inout3
1789          movdqa @tweak[2],`16*2`(%rsp)
1790         aesdec          $rndkey1,$inout2
1791         pxor            $rndkey0,$inout4
1792          movdqa @tweak[3],`16*3`(%rsp)
1793         aesdec          $rndkey1,$inout3
1794         pxor            $rndkey0,$inout5
1795         $movkey         ($key),$rndkey0
1796         dec             $rounds
1797          movdqa @tweak[4],`16*4`(%rsp)
1798         aesdec          $rndkey1,$inout4
1799          movdqa @tweak[5],`16*5`(%rsp)
1800         aesdec          $rndkey1,$inout5
1801         pxor    $twtmp,$twtmp
1802         pcmpgtd @tweak[5],$twtmp
1803         jmp             .Lxts_dec_loop6_enter
1804
1805 .align  16
1806 .Lxts_dec_loop6:
1807         aesdec          $rndkey1,$inout0
1808         aesdec          $rndkey1,$inout1
1809         dec             $rounds
1810         aesdec          $rndkey1,$inout2
1811         aesdec          $rndkey1,$inout3
1812         aesdec          $rndkey1,$inout4
1813         aesdec          $rndkey1,$inout5
1814 .Lxts_dec_loop6_enter:
1815         $movkey         16($key),$rndkey1
1816         aesdec          $rndkey0,$inout0
1817         aesdec          $rndkey0,$inout1
1818         lea             32($key),$key
1819         aesdec          $rndkey0,$inout2
1820         aesdec          $rndkey0,$inout3
1821         aesdec          $rndkey0,$inout4
1822         aesdec          $rndkey0,$inout5
1823         $movkey         ($key),$rndkey0
1824         jnz             .Lxts_dec_loop6
1825
1826         pshufd  \$0x13,$twtmp,$twres
1827         pxor    $twtmp,$twtmp
1828         paddq   @tweak[5],@tweak[5]             # psllq 1,$tweak
1829          aesdec         $rndkey1,$inout0
1830         pand    $twmask,$twres                  # isolate carry and residue
1831          aesdec         $rndkey1,$inout1
1832         pcmpgtd @tweak[5],$twtmp                # broadcast upper bits
1833          aesdec         $rndkey1,$inout2
1834         pxor    $twres,@tweak[5]
1835          aesdec         $rndkey1,$inout3
1836          aesdec         $rndkey1,$inout4
1837          aesdec         $rndkey1,$inout5
1838          $movkey        16($key),$rndkey1
1839
1840         pshufd  \$0x13,$twtmp,$twres
1841         pxor    $twtmp,$twtmp
1842         movdqa  @tweak[5],@tweak[0]
1843         paddq   @tweak[5],@tweak[5]             # psllq 1,$tweak
1844          aesdec         $rndkey0,$inout0
1845         pand    $twmask,$twres                  # isolate carry and residue
1846          aesdec         $rndkey0,$inout1
1847         pcmpgtd @tweak[5],$twtmp                # broadcat upper bits
1848          aesdec         $rndkey0,$inout2
1849         pxor    $twres,@tweak[5]
1850          aesdec         $rndkey0,$inout3
1851          aesdec         $rndkey0,$inout4
1852          aesdec         $rndkey0,$inout5
1853          $movkey        32($key),$rndkey0
1854
1855         pshufd  \$0x13,$twtmp,$twres
1856         pxor    $twtmp,$twtmp
1857         movdqa  @tweak[5],@tweak[1]
1858         paddq   @tweak[5],@tweak[5]             # psllq 1,$tweak
1859          aesdec         $rndkey1,$inout0
1860         pand    $twmask,$twres                  # isolate carry and residue
1861          aesdec         $rndkey1,$inout1
1862         pcmpgtd @tweak[5],$twtmp                # broadcat upper bits
1863          aesdec         $rndkey1,$inout2
1864         pxor    $twres,@tweak[5]
1865          aesdec         $rndkey1,$inout3
1866          aesdec         $rndkey1,$inout4
1867          aesdec         $rndkey1,$inout5
1868
1869         pshufd  \$0x13,$twtmp,$twres
1870         pxor    $twtmp,$twtmp
1871         movdqa  @tweak[5],@tweak[2]
1872         paddq   @tweak[5],@tweak[5]             # psllq 1,$tweak
1873          aesdeclast     $rndkey0,$inout0
1874         pand    $twmask,$twres                  # isolate carry and residue
1875          aesdeclast     $rndkey0,$inout1
1876         pcmpgtd @tweak[5],$twtmp                # broadcat upper bits
1877          aesdeclast     $rndkey0,$inout2
1878         pxor    $twres,@tweak[5]
1879          aesdeclast     $rndkey0,$inout3
1880          aesdeclast     $rndkey0,$inout4
1881          aesdeclast     $rndkey0,$inout5
1882
1883         pshufd  \$0x13,$twtmp,$twres
1884         pxor    $twtmp,$twtmp
1885         movdqa  @tweak[5],@tweak[3]
1886         paddq   @tweak[5],@tweak[5]             # psllq 1,$tweak
1887          xorps  `16*0`(%rsp),$inout0            # output^=tweak
1888         pand    $twmask,$twres                  # isolate carry and residue
1889          xorps  `16*1`(%rsp),$inout1
1890         pcmpgtd @tweak[5],$twtmp                # broadcat upper bits
1891         pxor    $twres,@tweak[5]
1892
1893         xorps   `16*2`(%rsp),$inout2
1894         movups  $inout0,`16*0`($out)            # write output
1895         xorps   `16*3`(%rsp),$inout3
1896         movups  $inout1,`16*1`($out)
1897         xorps   `16*4`(%rsp),$inout4
1898         movups  $inout2,`16*2`($out)
1899         xorps   `16*5`(%rsp),$inout5
1900         movups  $inout3,`16*3`($out)
1901         mov     $rnds_,$rounds                  # restore $rounds
1902         movups  $inout4,`16*4`($out)
1903         movups  $inout5,`16*5`($out)
1904         lea     `16*6`($out),$out
1905         sub     \$16*6,$len
1906         jnc     .Lxts_dec_grandloop
1907
1908         lea     3($rounds,$rounds),$rounds      # restore original value
1909         mov     $key_,$key                      # restore $key
1910         mov     $rounds,$rnds_                  # backup $rounds
1911
1912 .Lxts_dec_short:
1913         add     \$16*6,$len
1914         jz      .Lxts_dec_done
1915
1916         cmp     \$0x20,$len
1917         jb      .Lxts_dec_one
1918         je      .Lxts_dec_two
1919
1920         cmp     \$0x40,$len
1921         jb      .Lxts_dec_three
1922         je      .Lxts_dec_four
1923
1924         pshufd  \$0x13,$twtmp,$twres
1925         movdqa  @tweak[5],@tweak[4]
1926         paddq   @tweak[5],@tweak[5]             # psllq 1,$tweak
1927          movdqu ($inp),$inout0
1928         pand    $twmask,$twres                  # isolate carry and residue
1929          movdqu 16*1($inp),$inout1
1930         pxor    $twres,@tweak[5]
1931
1932         movdqu  16*2($inp),$inout2
1933         pxor    @tweak[0],$inout0
1934         movdqu  16*3($inp),$inout3
1935         pxor    @tweak[1],$inout1
1936         movdqu  16*4($inp),$inout4
1937         lea     16*5($inp),$inp
1938         pxor    @tweak[2],$inout2
1939         pxor    @tweak[3],$inout3
1940         pxor    @tweak[4],$inout4
1941
1942         call    _aesni_decrypt6
1943
1944         xorps   @tweak[0],$inout0
1945         xorps   @tweak[1],$inout1
1946         xorps   @tweak[2],$inout2
1947         movdqu  $inout0,($out)
1948         xorps   @tweak[3],$inout3
1949         movdqu  $inout1,16*1($out)
1950         xorps   @tweak[4],$inout4
1951         movdqu  $inout2,16*2($out)
1952          pxor           $twtmp,$twtmp
1953         movdqu  $inout3,16*3($out)
1954          pcmpgtd        @tweak[5],$twtmp
1955         movdqu  $inout4,16*4($out)
1956         lea     16*5($out),$out
1957          pshufd         \$0x13,$twtmp,@tweak[1] # $twres
1958         and     \$15,$len_
1959         jz      .Lxts_dec_ret
1960
1961         movdqa  @tweak[5],@tweak[0]
1962         paddq   @tweak[5],@tweak[5]             # psllq 1,$tweak
1963         pand    $twmask,@tweak[1]               # isolate carry and residue
1964         pxor    @tweak[5],@tweak[1]
1965         jmp     .Lxts_dec_done2
1966
1967 .align  16
1968 .Lxts_dec_one:
1969         movups  ($inp),$inout0
1970         lea     16*1($inp),$inp
1971         xorps   @tweak[0],$inout0
1972 ___
1973         &aesni_generate1("dec",$key,$rounds);
1974 $code.=<<___;
1975         xorps   @tweak[0],$inout0
1976         movdqa  @tweak[1],@tweak[0]
1977         movups  $inout0,($out)
1978         movdqa  @tweak[2],@tweak[1]
1979         lea     16*1($out),$out
1980         jmp     .Lxts_dec_done
1981
1982 .align  16
1983 .Lxts_dec_two:
1984         movups  ($inp),$inout0
1985         movups  16($inp),$inout1
1986         lea     32($inp),$inp
1987         xorps   @tweak[0],$inout0
1988         xorps   @tweak[1],$inout1
1989
1990         call    _aesni_decrypt3
1991
1992         xorps   @tweak[0],$inout0
1993         movdqa  @tweak[2],@tweak[0]
1994         xorps   @tweak[1],$inout1
1995         movdqa  @tweak[3],@tweak[1]
1996         movups  $inout0,($out)
1997         movups  $inout1,16*1($out)
1998         lea     16*2($out),$out
1999         jmp     .Lxts_dec_done
2000
2001 .align  16
2002 .Lxts_dec_three:
2003         movups  ($inp),$inout0
2004         movups  16*1($inp),$inout1
2005         movups  16*2($inp),$inout2
2006         lea     16*3($inp),$inp
2007         xorps   @tweak[0],$inout0
2008         xorps   @tweak[1],$inout1
2009         xorps   @tweak[2],$inout2
2010
2011         call    _aesni_decrypt3
2012
2013         xorps   @tweak[0],$inout0
2014         movdqa  @tweak[3],@tweak[0]
2015         xorps   @tweak[1],$inout1
2016         movdqa  @tweak[5],@tweak[1]
2017         xorps   @tweak[2],$inout2
2018         movups  $inout0,($out)
2019         movups  $inout1,16*1($out)
2020         movups  $inout2,16*2($out)
2021         lea     16*3($out),$out
2022         jmp     .Lxts_dec_done
2023
2024 .align  16
2025 .Lxts_dec_four:
2026         pshufd  \$0x13,$twtmp,$twres
2027         movdqa  @tweak[5],@tweak[4]
2028         paddq   @tweak[5],@tweak[5]             # psllq 1,$tweak
2029          movups ($inp),$inout0
2030         pand    $twmask,$twres                  # isolate carry and residue
2031          movups 16*1($inp),$inout1
2032         pxor    $twres,@tweak[5]
2033
2034         movups  16*2($inp),$inout2
2035         xorps   @tweak[0],$inout0
2036         movups  16*3($inp),$inout3
2037         lea     16*4($inp),$inp
2038         xorps   @tweak[1],$inout1
2039         xorps   @tweak[2],$inout2
2040         xorps   @tweak[3],$inout3
2041
2042         call    _aesni_decrypt4
2043
2044         xorps   @tweak[0],$inout0
2045         movdqa  @tweak[4],@tweak[0]
2046         xorps   @tweak[1],$inout1
2047         movdqa  @tweak[5],@tweak[1]
2048         xorps   @tweak[2],$inout2
2049         movups  $inout0,($out)
2050         xorps   @tweak[3],$inout3
2051         movups  $inout1,16*1($out)
2052         movups  $inout2,16*2($out)
2053         movups  $inout3,16*3($out)
2054         lea     16*4($out),$out
2055         jmp     .Lxts_dec_done
2056
2057 .align  16
2058 .Lxts_dec_done:
2059         and     \$15,$len_
2060         jz      .Lxts_dec_ret
2061 .Lxts_dec_done2:
2062         mov     $len_,$len
2063         mov     $key_,$key                      # restore $key
2064         mov     $rnds_,$rounds                  # restore $rounds
2065
2066         movups  ($inp),$inout0
2067         xorps   @tweak[1],$inout0
2068 ___
2069         &aesni_generate1("dec",$key,$rounds);
2070 $code.=<<___;
2071         xorps   @tweak[1],$inout0
2072         movups  $inout0,($out)
2073
2074 .Lxts_dec_steal:
2075         movzb   16($inp),%eax                   # borrow $rounds ...
2076         movzb   ($out),%ecx                     # ... and $key
2077         lea     1($inp),$inp
2078         mov     %al,($out)
2079         mov     %cl,16($out)
2080         lea     1($out),$out
2081         sub     \$1,$len
2082         jnz     .Lxts_dec_steal
2083
2084         sub     $len_,$out                      # rewind $out
2085         mov     $key_,$key                      # restore $key
2086         mov     $rnds_,$rounds                  # restore $rounds
2087
2088         movups  ($out),$inout0
2089         xorps   @tweak[0],$inout0
2090 ___
2091         &aesni_generate1("dec",$key,$rounds);
2092 $code.=<<___;
2093         xorps   @tweak[0],$inout0
2094         movups  $inout0,($out)
2095
2096 .Lxts_dec_ret:
2097 ___
2098 $code.=<<___ if ($win64);
2099         movaps  0x60(%rsp),%xmm6
2100         movaps  0x70(%rsp),%xmm7
2101         movaps  0x80(%rsp),%xmm8
2102         movaps  0x90(%rsp),%xmm9
2103         movaps  0xa0(%rsp),%xmm10
2104         movaps  0xb0(%rsp),%xmm11
2105         movaps  0xc0(%rsp),%xmm12
2106         movaps  0xd0(%rsp),%xmm13
2107         movaps  0xe0(%rsp),%xmm14
2108         movaps  0xf0(%rsp),%xmm15
2109 ___
2110 $code.=<<___;
2111         lea     $frame_size(%rsp),%rsp
2112 .Lxts_dec_epilogue:
2113         ret
2114 .size   aesni_xts_decrypt,.-aesni_xts_decrypt
2115 ___
2116 } }}
2117 \f
2118 ########################################################################
2119 # void $PREFIX_cbc_encrypt (const void *inp, void *out,
2120 #                           size_t length, const AES_KEY *key,
2121 #                           unsigned char *ivp,const int enc);
2122 {
2123 my $reserved = $win64?0x40:-0x18;       # used in decrypt
2124 $code.=<<___;
2125 .globl  ${PREFIX}_cbc_encrypt
2126 .type   ${PREFIX}_cbc_encrypt,\@function,6
2127 .align  16
2128 ${PREFIX}_cbc_encrypt:
2129         test    $len,$len               # check length
2130         jz      .Lcbc_ret
2131
2132         mov     240($key),$rnds_        # key->rounds
2133         mov     $key,$key_              # backup $key
2134         test    %r9d,%r9d               # 6th argument
2135         jz      .Lcbc_decrypt
2136 #--------------------------- CBC ENCRYPT ------------------------------#
2137         movups  ($ivp),$inout0          # load iv as initial state
2138         mov     $rnds_,$rounds
2139         cmp     \$16,$len
2140         jb      .Lcbc_enc_tail
2141         sub     \$16,$len
2142         jmp     .Lcbc_enc_loop
2143 .align  16
2144 .Lcbc_enc_loop:
2145         movups  ($inp),$inout1          # load input
2146         lea     16($inp),$inp
2147         #xorps  $inout1,$inout0
2148 ___
2149         &aesni_generate1("enc",$key,$rounds,$inout0,$inout1);
2150 $code.=<<___;
2151         mov     $rnds_,$rounds          # restore $rounds
2152         mov     $key_,$key              # restore $key
2153         movups  $inout0,0($out)         # store output
2154         lea     16($out),$out
2155         sub     \$16,$len
2156         jnc     .Lcbc_enc_loop
2157         add     \$16,$len
2158         jnz     .Lcbc_enc_tail
2159         movups  $inout0,($ivp)
2160         jmp     .Lcbc_ret
2161
2162 .Lcbc_enc_tail:
2163         mov     $len,%rcx       # zaps $key
2164         xchg    $inp,$out       # $inp is %rsi and $out is %rdi now
2165         .long   0x9066A4F3      # rep movsb
2166         mov     \$16,%ecx       # zero tail
2167         sub     $len,%rcx
2168         xor     %eax,%eax
2169         .long   0x9066AAF3      # rep stosb
2170         lea     -16(%rdi),%rdi  # rewind $out by 1 block
2171         mov     $rnds_,$rounds  # restore $rounds
2172         mov     %rdi,%rsi       # $inp and $out are the same
2173         mov     $key_,$key      # restore $key
2174         xor     $len,$len       # len=16
2175         jmp     .Lcbc_enc_loop  # one more spin
2176 \f#--------------------------- CBC DECRYPT ------------------------------#
2177 .align  16
2178 .Lcbc_decrypt:
2179 ___
2180 $code.=<<___ if ($win64);
2181         lea     -0x58(%rsp),%rsp
2182         movaps  %xmm6,(%rsp)
2183         movaps  %xmm7,0x10(%rsp)
2184         movaps  %xmm8,0x20(%rsp)
2185         movaps  %xmm9,0x30(%rsp)
2186 .Lcbc_decrypt_body:
2187 ___
2188 $code.=<<___;
2189         movups  ($ivp),$iv
2190         mov     $rnds_,$rounds
2191         cmp     \$0x70,$len
2192         jbe     .Lcbc_dec_tail
2193         shr     \$1,$rnds_
2194         sub     \$0x70,$len
2195         mov     $rnds_,$rounds
2196         movaps  $iv,$reserved(%rsp)
2197         jmp     .Lcbc_dec_loop8_enter
2198 .align  16
2199 .Lcbc_dec_loop8:
2200         movaps  $rndkey0,$reserved(%rsp)        # save IV
2201         movups  $inout7,($out)
2202         lea     0x10($out),$out
2203 .Lcbc_dec_loop8_enter:
2204         $movkey         ($key),$rndkey0
2205         movups  ($inp),$inout0                  # load input
2206         movups  0x10($inp),$inout1
2207         $movkey         16($key),$rndkey1
2208
2209         lea             32($key),$key
2210         movdqu  0x20($inp),$inout2
2211         xorps           $rndkey0,$inout0
2212         movdqu  0x30($inp),$inout3
2213         xorps           $rndkey0,$inout1
2214         movdqu  0x40($inp),$inout4
2215         aesdec          $rndkey1,$inout0
2216         pxor            $rndkey0,$inout2
2217         movdqu  0x50($inp),$inout5
2218         aesdec          $rndkey1,$inout1
2219         pxor            $rndkey0,$inout3
2220         movdqu  0x60($inp),$inout6
2221         aesdec          $rndkey1,$inout2
2222         pxor            $rndkey0,$inout4
2223         movdqu  0x70($inp),$inout7
2224         aesdec          $rndkey1,$inout3
2225         pxor            $rndkey0,$inout5
2226         dec             $rounds
2227         aesdec          $rndkey1,$inout4
2228         pxor            $rndkey0,$inout6
2229         aesdec          $rndkey1,$inout5
2230         pxor            $rndkey0,$inout7
2231         $movkey         ($key),$rndkey0
2232         aesdec          $rndkey1,$inout6
2233         aesdec          $rndkey1,$inout7
2234         $movkey         16($key),$rndkey1
2235
2236         call            .Ldec_loop8_enter
2237
2238         movups  ($inp),$rndkey1         # re-load input
2239         movups  0x10($inp),$rndkey0
2240         xorps   $reserved(%rsp),$inout0 # ^= IV
2241         xorps   $rndkey1,$inout1
2242         movups  0x20($inp),$rndkey1
2243         xorps   $rndkey0,$inout2
2244         movups  0x30($inp),$rndkey0
2245         xorps   $rndkey1,$inout3
2246         movups  0x40($inp),$rndkey1
2247         xorps   $rndkey0,$inout4
2248         movups  0x50($inp),$rndkey0
2249         xorps   $rndkey1,$inout5
2250         movups  0x60($inp),$rndkey1
2251         xorps   $rndkey0,$inout6
2252         movups  0x70($inp),$rndkey0     # IV
2253         xorps   $rndkey1,$inout7
2254         movups  $inout0,($out)
2255         movups  $inout1,0x10($out)
2256         movups  $inout2,0x20($out)
2257         movups  $inout3,0x30($out)
2258         mov     $rnds_,$rounds          # restore $rounds
2259         movups  $inout4,0x40($out)
2260         mov     $key_,$key              # restore $key
2261         movups  $inout5,0x50($out)
2262         lea     0x80($inp),$inp
2263         movups  $inout6,0x60($out)
2264         lea     0x70($out),$out
2265         sub     \$0x80,$len
2266         ja      .Lcbc_dec_loop8
2267
2268         movaps  $inout7,$inout0
2269         movaps  $rndkey0,$iv
2270         add     \$0x70,$len
2271         jle     .Lcbc_dec_tail_collected
2272         movups  $inout0,($out)
2273         lea     1($rnds_,$rnds_),$rounds
2274         lea     0x10($out),$out
2275 .Lcbc_dec_tail:
2276         movups  ($inp),$inout0
2277         movaps  $inout0,$in0
2278         cmp     \$0x10,$len
2279         jbe     .Lcbc_dec_one
2280
2281         movups  0x10($inp),$inout1
2282         movaps  $inout1,$in1
2283         cmp     \$0x20,$len
2284         jbe     .Lcbc_dec_two
2285
2286         movups  0x20($inp),$inout2
2287         movaps  $inout2,$in2
2288         cmp     \$0x30,$len
2289         jbe     .Lcbc_dec_three
2290
2291         movups  0x30($inp),$inout3
2292         cmp     \$0x40,$len
2293         jbe     .Lcbc_dec_four
2294
2295         movups  0x40($inp),$inout4
2296         cmp     \$0x50,$len
2297         jbe     .Lcbc_dec_five
2298
2299         movups  0x50($inp),$inout5
2300         cmp     \$0x60,$len
2301         jbe     .Lcbc_dec_six
2302
2303         movups  0x60($inp),$inout6
2304         movaps  $iv,$reserved(%rsp)     # save IV
2305         call    _aesni_decrypt8
2306         movups  ($inp),$rndkey1
2307         movups  0x10($inp),$rndkey0
2308         xorps   $reserved(%rsp),$inout0 # ^= IV
2309         xorps   $rndkey1,$inout1
2310         movups  0x20($inp),$rndkey1
2311         xorps   $rndkey0,$inout2
2312         movups  0x30($inp),$rndkey0
2313         xorps   $rndkey1,$inout3
2314         movups  0x40($inp),$rndkey1
2315         xorps   $rndkey0,$inout4
2316         movups  0x50($inp),$rndkey0
2317         xorps   $rndkey1,$inout5
2318         movups  0x60($inp),$iv          # IV
2319         xorps   $rndkey0,$inout6
2320         movups  $inout0,($out)
2321         movups  $inout1,0x10($out)
2322         movups  $inout2,0x20($out)
2323         movups  $inout3,0x30($out)
2324         movups  $inout4,0x40($out)
2325         movups  $inout5,0x50($out)
2326         lea     0x60($out),$out
2327         movaps  $inout6,$inout0
2328         sub     \$0x70,$len
2329         jmp     .Lcbc_dec_tail_collected
2330 .align  16
2331 .Lcbc_dec_one:
2332 ___
2333         &aesni_generate1("dec",$key,$rounds);
2334 $code.=<<___;
2335         xorps   $iv,$inout0
2336         movaps  $in0,$iv
2337         sub     \$0x10,$len
2338         jmp     .Lcbc_dec_tail_collected
2339 .align  16
2340 .Lcbc_dec_two:
2341         xorps   $inout2,$inout2
2342         call    _aesni_decrypt3
2343         xorps   $iv,$inout0
2344         xorps   $in0,$inout1
2345         movups  $inout0,($out)
2346         movaps  $in1,$iv
2347         movaps  $inout1,$inout0
2348         lea     0x10($out),$out
2349         sub     \$0x20,$len
2350         jmp     .Lcbc_dec_tail_collected
2351 .align  16
2352 .Lcbc_dec_three:
2353         call    _aesni_decrypt3
2354         xorps   $iv,$inout0
2355         xorps   $in0,$inout1
2356         movups  $inout0,($out)
2357         xorps   $in1,$inout2
2358         movups  $inout1,0x10($out)
2359         movaps  $in2,$iv
2360         movaps  $inout2,$inout0
2361         lea     0x20($out),$out
2362         sub     \$0x30,$len
2363         jmp     .Lcbc_dec_tail_collected
2364 .align  16
2365 .Lcbc_dec_four:
2366         call    _aesni_decrypt4
2367         xorps   $iv,$inout0
2368         movups  0x30($inp),$iv
2369         xorps   $in0,$inout1
2370         movups  $inout0,($out)
2371         xorps   $in1,$inout2
2372         movups  $inout1,0x10($out)
2373         xorps   $in2,$inout3
2374         movups  $inout2,0x20($out)
2375         movaps  $inout3,$inout0
2376         lea     0x30($out),$out
2377         sub     \$0x40,$len
2378         jmp     .Lcbc_dec_tail_collected
2379 .align  16
2380 .Lcbc_dec_five:
2381         xorps   $inout5,$inout5
2382         call    _aesni_decrypt6
2383         movups  0x10($inp),$rndkey1
2384         movups  0x20($inp),$rndkey0
2385         xorps   $iv,$inout0
2386         xorps   $in0,$inout1
2387         xorps   $rndkey1,$inout2
2388         movups  0x30($inp),$rndkey1
2389         xorps   $rndkey0,$inout3
2390         movups  0x40($inp),$iv
2391         xorps   $rndkey1,$inout4
2392         movups  $inout0,($out)
2393         movups  $inout1,0x10($out)
2394         movups  $inout2,0x20($out)
2395         movups  $inout3,0x30($out)
2396         lea     0x40($out),$out
2397         movaps  $inout4,$inout0
2398         sub     \$0x50,$len
2399         jmp     .Lcbc_dec_tail_collected
2400 .align  16
2401 .Lcbc_dec_six:
2402         call    _aesni_decrypt6
2403         movups  0x10($inp),$rndkey1
2404         movups  0x20($inp),$rndkey0
2405         xorps   $iv,$inout0
2406         xorps   $in0,$inout1
2407         xorps   $rndkey1,$inout2
2408         movups  0x30($inp),$rndkey1
2409         xorps   $rndkey0,$inout3
2410         movups  0x40($inp),$rndkey0
2411         xorps   $rndkey1,$inout4
2412         movups  0x50($inp),$iv
2413         xorps   $rndkey0,$inout5
2414         movups  $inout0,($out)
2415         movups  $inout1,0x10($out)
2416         movups  $inout2,0x20($out)
2417         movups  $inout3,0x30($out)
2418         movups  $inout4,0x40($out)
2419         lea     0x50($out),$out
2420         movaps  $inout5,$inout0
2421         sub     \$0x60,$len
2422         jmp     .Lcbc_dec_tail_collected
2423 .align  16
2424 .Lcbc_dec_tail_collected:
2425         and     \$15,$len
2426         movups  $iv,($ivp)
2427         jnz     .Lcbc_dec_tail_partial
2428         movups  $inout0,($out)
2429         jmp     .Lcbc_dec_ret
2430 .align  16
2431 .Lcbc_dec_tail_partial:
2432         movaps  $inout0,$reserved(%rsp)
2433         mov     \$16,%rcx
2434         mov     $out,%rdi
2435         sub     $len,%rcx
2436         lea     $reserved(%rsp),%rsi
2437         .long   0x9066A4F3      # rep movsb
2438
2439 .Lcbc_dec_ret:
2440 ___
2441 $code.=<<___ if ($win64);
2442         movaps  (%rsp),%xmm6
2443         movaps  0x10(%rsp),%xmm7
2444         movaps  0x20(%rsp),%xmm8
2445         movaps  0x30(%rsp),%xmm9
2446         lea     0x58(%rsp),%rsp
2447 ___
2448 $code.=<<___;
2449 .Lcbc_ret:
2450         ret
2451 .size   ${PREFIX}_cbc_encrypt,.-${PREFIX}_cbc_encrypt
2452 ___
2453 } \f
2454 # int $PREFIX_set_[en|de]crypt_key (const unsigned char *userKey,
2455 #                               int bits, AES_KEY *key)
2456 { my ($inp,$bits,$key) = @_4args;
2457   $bits =~ s/%r/%e/;
2458
2459 $code.=<<___;
2460 .globl  ${PREFIX}_set_decrypt_key
2461 .type   ${PREFIX}_set_decrypt_key,\@abi-omnipotent
2462 .align  16
2463 ${PREFIX}_set_decrypt_key:
2464         .byte   0x48,0x83,0xEC,0x08     # sub rsp,8
2465         call    __aesni_set_encrypt_key
2466         shl     \$4,$bits               # rounds-1 after _aesni_set_encrypt_key
2467         test    %eax,%eax
2468         jnz     .Ldec_key_ret
2469         lea     16($key,$bits),$inp     # points at the end of key schedule
2470
2471         $movkey ($key),%xmm0            # just swap
2472         $movkey ($inp),%xmm1
2473         $movkey %xmm0,($inp)
2474         $movkey %xmm1,($key)
2475         lea     16($key),$key
2476         lea     -16($inp),$inp
2477
2478 .Ldec_key_inverse:
2479         $movkey ($key),%xmm0            # swap and inverse
2480         $movkey ($inp),%xmm1
2481         aesimc  %xmm0,%xmm0
2482         aesimc  %xmm1,%xmm1
2483         lea     16($key),$key
2484         lea     -16($inp),$inp
2485         $movkey %xmm0,16($inp)
2486         $movkey %xmm1,-16($key)
2487         cmp     $key,$inp
2488         ja      .Ldec_key_inverse
2489
2490         $movkey ($key),%xmm0            # inverse middle
2491         aesimc  %xmm0,%xmm0
2492         $movkey %xmm0,($inp)
2493 .Ldec_key_ret:
2494         add     \$8,%rsp
2495         ret
2496 .LSEH_end_set_decrypt_key:
2497 .size   ${PREFIX}_set_decrypt_key,.-${PREFIX}_set_decrypt_key
2498 ___
2499 \f
2500 # This is based on submission by
2501 #
2502 #       Huang Ying <ying.huang@intel.com>
2503 #       Vinodh Gopal <vinodh.gopal@intel.com>
2504 #       Kahraman Akdemir
2505 #
2506 # Agressively optimized in respect to aeskeygenassist's critical path
2507 # and is contained in %xmm0-5 to meet Win64 ABI requirement.
2508 #
2509 $code.=<<___;
2510 .globl  ${PREFIX}_set_encrypt_key
2511 .type   ${PREFIX}_set_encrypt_key,\@abi-omnipotent
2512 .align  16
2513 ${PREFIX}_set_encrypt_key:
2514 __aesni_set_encrypt_key:
2515         .byte   0x48,0x83,0xEC,0x08     # sub rsp,8
2516         mov     \$-1,%rax
2517         test    $inp,$inp
2518         jz      .Lenc_key_ret
2519         test    $key,$key
2520         jz      .Lenc_key_ret
2521
2522         movups  ($inp),%xmm0            # pull first 128 bits of *userKey
2523         xorps   %xmm4,%xmm4             # low dword of xmm4 is assumed 0
2524         lea     16($key),%rax
2525         cmp     \$256,$bits
2526         je      .L14rounds
2527         cmp     \$192,$bits
2528         je      .L12rounds
2529         cmp     \$128,$bits
2530         jne     .Lbad_keybits
2531
2532 .L10rounds:
2533         mov     \$9,$bits                       # 10 rounds for 128-bit key
2534         $movkey %xmm0,($key)                    # round 0
2535         aeskeygenassist \$0x1,%xmm0,%xmm1       # round 1
2536         call            .Lkey_expansion_128_cold
2537         aeskeygenassist \$0x2,%xmm0,%xmm1       # round 2
2538         call            .Lkey_expansion_128
2539         aeskeygenassist \$0x4,%xmm0,%xmm1       # round 3
2540         call            .Lkey_expansion_128
2541         aeskeygenassist \$0x8,%xmm0,%xmm1       # round 4
2542         call            .Lkey_expansion_128
2543         aeskeygenassist \$0x10,%xmm0,%xmm1      # round 5
2544         call            .Lkey_expansion_128
2545         aeskeygenassist \$0x20,%xmm0,%xmm1      # round 6
2546         call            .Lkey_expansion_128
2547         aeskeygenassist \$0x40,%xmm0,%xmm1      # round 7
2548         call            .Lkey_expansion_128
2549         aeskeygenassist \$0x80,%xmm0,%xmm1      # round 8
2550         call            .Lkey_expansion_128
2551         aeskeygenassist \$0x1b,%xmm0,%xmm1      # round 9
2552         call            .Lkey_expansion_128
2553         aeskeygenassist \$0x36,%xmm0,%xmm1      # round 10
2554         call            .Lkey_expansion_128
2555         $movkey %xmm0,(%rax)
2556         mov     $bits,80(%rax)  # 240(%rdx)
2557         xor     %eax,%eax
2558         jmp     .Lenc_key_ret
2559
2560 .align  16
2561 .L12rounds:
2562         movq    16($inp),%xmm2                  # remaining 1/3 of *userKey
2563         mov     \$11,$bits                      # 12 rounds for 192
2564         $movkey %xmm0,($key)                    # round 0
2565         aeskeygenassist \$0x1,%xmm2,%xmm1       # round 1,2
2566         call            .Lkey_expansion_192a_cold
2567         aeskeygenassist \$0x2,%xmm2,%xmm1       # round 2,3
2568         call            .Lkey_expansion_192b
2569         aeskeygenassist \$0x4,%xmm2,%xmm1       # round 4,5
2570         call            .Lkey_expansion_192a
2571         aeskeygenassist \$0x8,%xmm2,%xmm1       # round 5,6
2572         call            .Lkey_expansion_192b
2573         aeskeygenassist \$0x10,%xmm2,%xmm1      # round 7,8
2574         call            .Lkey_expansion_192a
2575         aeskeygenassist \$0x20,%xmm2,%xmm1      # round 8,9
2576         call            .Lkey_expansion_192b
2577         aeskeygenassist \$0x40,%xmm2,%xmm1      # round 10,11
2578         call            .Lkey_expansion_192a
2579         aeskeygenassist \$0x80,%xmm2,%xmm1      # round 11,12
2580         call            .Lkey_expansion_192b
2581         $movkey %xmm0,(%rax)
2582         mov     $bits,48(%rax)  # 240(%rdx)
2583         xor     %rax, %rax
2584         jmp     .Lenc_key_ret
2585
2586 .align  16
2587 .L14rounds:
2588         movups  16($inp),%xmm2                  # remaning half of *userKey
2589         mov     \$13,$bits                      # 14 rounds for 256
2590         lea     16(%rax),%rax
2591         $movkey %xmm0,($key)                    # round 0
2592         $movkey %xmm2,16($key)                  # round 1
2593         aeskeygenassist \$0x1,%xmm2,%xmm1       # round 2
2594         call            .Lkey_expansion_256a_cold
2595         aeskeygenassist \$0x1,%xmm0,%xmm1       # round 3
2596         call            .Lkey_expansion_256b
2597         aeskeygenassist \$0x2,%xmm2,%xmm1       # round 4
2598         call            .Lkey_expansion_256a
2599         aeskeygenassist \$0x2,%xmm0,%xmm1       # round 5
2600         call            .Lkey_expansion_256b
2601         aeskeygenassist \$0x4,%xmm2,%xmm1       # round 6
2602         call            .Lkey_expansion_256a
2603         aeskeygenassist \$0x4,%xmm0,%xmm1       # round 7
2604         call            .Lkey_expansion_256b
2605         aeskeygenassist \$0x8,%xmm2,%xmm1       # round 8
2606         call            .Lkey_expansion_256a
2607         aeskeygenassist \$0x8,%xmm0,%xmm1       # round 9
2608         call            .Lkey_expansion_256b
2609         aeskeygenassist \$0x10,%xmm2,%xmm1      # round 10
2610         call            .Lkey_expansion_256a
2611         aeskeygenassist \$0x10,%xmm0,%xmm1      # round 11
2612         call            .Lkey_expansion_256b
2613         aeskeygenassist \$0x20,%xmm2,%xmm1      # round 12
2614         call            .Lkey_expansion_256a
2615         aeskeygenassist \$0x20,%xmm0,%xmm1      # round 13
2616         call            .Lkey_expansion_256b
2617         aeskeygenassist \$0x40,%xmm2,%xmm1      # round 14
2618         call            .Lkey_expansion_256a
2619         $movkey %xmm0,(%rax)
2620         mov     $bits,16(%rax)  # 240(%rdx)
2621         xor     %rax,%rax
2622         jmp     .Lenc_key_ret
2623
2624 .align  16
2625 .Lbad_keybits:
2626         mov     \$-2,%rax
2627 .Lenc_key_ret:
2628         add     \$8,%rsp
2629         ret
2630 .LSEH_end_set_encrypt_key:
2631 \f
2632 .align  16
2633 .Lkey_expansion_128:
2634         $movkey %xmm0,(%rax)
2635         lea     16(%rax),%rax
2636 .Lkey_expansion_128_cold:
2637         shufps  \$0b00010000,%xmm0,%xmm4
2638         xorps   %xmm4, %xmm0
2639         shufps  \$0b10001100,%xmm0,%xmm4
2640         xorps   %xmm4, %xmm0
2641         shufps  \$0b11111111,%xmm1,%xmm1        # critical path
2642         xorps   %xmm1,%xmm0
2643         ret
2644
2645 .align 16
2646 .Lkey_expansion_192a:
2647         $movkey %xmm0,(%rax)
2648         lea     16(%rax),%rax
2649 .Lkey_expansion_192a_cold:
2650         movaps  %xmm2, %xmm5
2651 .Lkey_expansion_192b_warm:
2652         shufps  \$0b00010000,%xmm0,%xmm4
2653         movdqa  %xmm2,%xmm3
2654         xorps   %xmm4,%xmm0
2655         shufps  \$0b10001100,%xmm0,%xmm4
2656         pslldq  \$4,%xmm3
2657         xorps   %xmm4,%xmm0
2658         pshufd  \$0b01010101,%xmm1,%xmm1        # critical path
2659         pxor    %xmm3,%xmm2
2660         pxor    %xmm1,%xmm0
2661         pshufd  \$0b11111111,%xmm0,%xmm3
2662         pxor    %xmm3,%xmm2
2663         ret
2664
2665 .align 16
2666 .Lkey_expansion_192b:
2667         movaps  %xmm0,%xmm3
2668         shufps  \$0b01000100,%xmm0,%xmm5
2669         $movkey %xmm5,(%rax)
2670         shufps  \$0b01001110,%xmm2,%xmm3
2671         $movkey %xmm3,16(%rax)
2672         lea     32(%rax),%rax
2673         jmp     .Lkey_expansion_192b_warm
2674
2675 .align  16
2676 .Lkey_expansion_256a:
2677         $movkey %xmm2,(%rax)
2678         lea     16(%rax),%rax
2679 .Lkey_expansion_256a_cold:
2680         shufps  \$0b00010000,%xmm0,%xmm4
2681         xorps   %xmm4,%xmm0
2682         shufps  \$0b10001100,%xmm0,%xmm4
2683         xorps   %xmm4,%xmm0
2684         shufps  \$0b11111111,%xmm1,%xmm1        # critical path
2685         xorps   %xmm1,%xmm0
2686         ret
2687
2688 .align 16
2689 .Lkey_expansion_256b:
2690         $movkey %xmm0,(%rax)
2691         lea     16(%rax),%rax
2692
2693         shufps  \$0b00010000,%xmm2,%xmm4
2694         xorps   %xmm4,%xmm2
2695         shufps  \$0b10001100,%xmm2,%xmm4
2696         xorps   %xmm4,%xmm2
2697         shufps  \$0b10101010,%xmm1,%xmm1        # critical path
2698         xorps   %xmm1,%xmm2
2699         ret
2700 .size   ${PREFIX}_set_encrypt_key,.-${PREFIX}_set_encrypt_key
2701 .size   __aesni_set_encrypt_key,.-__aesni_set_encrypt_key
2702 ___
2703 }
2704 \f
2705 $code.=<<___;
2706 .align  64
2707 .Lbswap_mask:
2708         .byte   15,14,13,12,11,10,9,8,7,6,5,4,3,2,1,0
2709 .Lincrement32:
2710         .long   6,6,6,0
2711 .Lincrement64:
2712         .long   1,0,0,0
2713 .Lxts_magic:
2714         .long   0x87,0,1,0
2715
2716 .asciz  "AES for Intel AES-NI, CRYPTOGAMS by <appro\@openssl.org>"
2717 .align  64
2718 ___
2719
2720 # EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
2721 #               CONTEXT *context,DISPATCHER_CONTEXT *disp)
2722 if ($win64) {
2723 $rec="%rcx";
2724 $frame="%rdx";
2725 $context="%r8";
2726 $disp="%r9";
2727
2728 $code.=<<___;
2729 .extern __imp_RtlVirtualUnwind
2730 ___
2731 $code.=<<___ if ($PREFIX eq "aesni");
2732 .type   ecb_se_handler,\@abi-omnipotent
2733 .align  16
2734 ecb_se_handler:
2735         push    %rsi
2736         push    %rdi
2737         push    %rbx
2738         push    %rbp
2739         push    %r12
2740         push    %r13
2741         push    %r14
2742         push    %r15
2743         pushfq
2744         sub     \$64,%rsp
2745
2746         mov     152($context),%rax      # pull context->Rsp
2747
2748         jmp     .Lcommon_seh_tail
2749 .size   ecb_se_handler,.-ecb_se_handler
2750
2751 .type   ccm64_se_handler,\@abi-omnipotent
2752 .align  16
2753 ccm64_se_handler:
2754         push    %rsi
2755         push    %rdi
2756         push    %rbx
2757         push    %rbp
2758         push    %r12
2759         push    %r13
2760         push    %r14
2761         push    %r15
2762         pushfq
2763         sub     \$64,%rsp
2764
2765         mov     120($context),%rax      # pull context->Rax
2766         mov     248($context),%rbx      # pull context->Rip
2767
2768         mov     8($disp),%rsi           # disp->ImageBase
2769         mov     56($disp),%r11          # disp->HandlerData
2770
2771         mov     0(%r11),%r10d           # HandlerData[0]
2772         lea     (%rsi,%r10),%r10        # prologue label
2773         cmp     %r10,%rbx               # context->Rip<prologue label
2774         jb      .Lcommon_seh_tail
2775
2776         mov     152($context),%rax      # pull context->Rsp
2777
2778         mov     4(%r11),%r10d           # HandlerData[1]
2779         lea     (%rsi,%r10),%r10        # epilogue label
2780         cmp     %r10,%rbx               # context->Rip>=epilogue label
2781         jae     .Lcommon_seh_tail
2782
2783         lea     0(%rax),%rsi            # %xmm save area
2784         lea     512($context),%rdi      # &context.Xmm6
2785         mov     \$8,%ecx                # 4*sizeof(%xmm0)/sizeof(%rax)
2786         .long   0xa548f3fc              # cld; rep movsq
2787         lea     0x58(%rax),%rax         # adjust stack pointer
2788
2789         jmp     .Lcommon_seh_tail
2790 .size   ccm64_se_handler,.-ccm64_se_handler
2791
2792 .type   ctr32_se_handler,\@abi-omnipotent
2793 .align  16
2794 ctr32_se_handler:
2795         push    %rsi
2796         push    %rdi
2797         push    %rbx
2798         push    %rbp
2799         push    %r12
2800         push    %r13
2801         push    %r14
2802         push    %r15
2803         pushfq
2804         sub     \$64,%rsp
2805
2806         mov     120($context),%rax      # pull context->Rax
2807         mov     248($context),%rbx      # pull context->Rip
2808
2809         lea     .Lctr32_body(%rip),%r10
2810         cmp     %r10,%rbx               # context->Rip<"prologue" label
2811         jb      .Lcommon_seh_tail
2812
2813         mov     152($context),%rax      # pull context->Rsp
2814
2815         lea     .Lctr32_ret(%rip),%r10
2816         cmp     %r10,%rbx
2817         jae     .Lcommon_seh_tail
2818
2819         lea     0x20(%rax),%rsi         # %xmm save area
2820         lea     512($context),%rdi      # &context.Xmm6
2821         mov     \$20,%ecx               # 10*sizeof(%xmm0)/sizeof(%rax)
2822         .long   0xa548f3fc              # cld; rep movsq
2823         lea     0xc8(%rax),%rax         # adjust stack pointer
2824
2825         jmp     .Lcommon_seh_tail
2826 .size   ctr32_se_handler,.-ctr32_se_handler
2827
2828 .type   xts_se_handler,\@abi-omnipotent
2829 .align  16
2830 xts_se_handler:
2831         push    %rsi
2832         push    %rdi
2833         push    %rbx
2834         push    %rbp
2835         push    %r12
2836         push    %r13
2837         push    %r14
2838         push    %r15
2839         pushfq
2840         sub     \$64,%rsp
2841
2842         mov     120($context),%rax      # pull context->Rax
2843         mov     248($context),%rbx      # pull context->Rip
2844
2845         mov     8($disp),%rsi           # disp->ImageBase
2846         mov     56($disp),%r11          # disp->HandlerData
2847
2848         mov     0(%r11),%r10d           # HandlerData[0]
2849         lea     (%rsi,%r10),%r10        # prologue lable
2850         cmp     %r10,%rbx               # context->Rip<prologue label
2851         jb      .Lcommon_seh_tail
2852
2853         mov     152($context),%rax      # pull context->Rsp
2854
2855         mov     4(%r11),%r10d           # HandlerData[1]
2856         lea     (%rsi,%r10),%r10        # epilogue label
2857         cmp     %r10,%rbx               # context->Rip>=epilogue label
2858         jae     .Lcommon_seh_tail
2859
2860         lea     0x60(%rax),%rsi         # %xmm save area
2861         lea     512($context),%rdi      # & context.Xmm6
2862         mov     \$20,%ecx               # 10*sizeof(%xmm0)/sizeof(%rax)
2863         .long   0xa548f3fc              # cld; rep movsq
2864         lea     0x68+160(%rax),%rax     # adjust stack pointer
2865
2866         jmp     .Lcommon_seh_tail
2867 .size   xts_se_handler,.-xts_se_handler
2868 ___
2869 $code.=<<___;
2870 .type   cbc_se_handler,\@abi-omnipotent
2871 .align  16
2872 cbc_se_handler:
2873         push    %rsi
2874         push    %rdi
2875         push    %rbx
2876         push    %rbp
2877         push    %r12
2878         push    %r13
2879         push    %r14
2880         push    %r15
2881         pushfq
2882         sub     \$64,%rsp
2883
2884         mov     152($context),%rax      # pull context->Rsp
2885         mov     248($context),%rbx      # pull context->Rip
2886
2887         lea     .Lcbc_decrypt(%rip),%r10
2888         cmp     %r10,%rbx               # context->Rip<"prologue" label
2889         jb      .Lcommon_seh_tail
2890
2891         lea     .Lcbc_decrypt_body(%rip),%r10
2892         cmp     %r10,%rbx               # context->Rip<cbc_decrypt_body
2893         jb      .Lrestore_cbc_rax
2894
2895         lea     .Lcbc_ret(%rip),%r10
2896         cmp     %r10,%rbx               # context->Rip>="epilogue" label
2897         jae     .Lcommon_seh_tail
2898
2899         lea     0(%rax),%rsi            # top of stack
2900         lea     512($context),%rdi      # &context.Xmm6
2901         mov     \$8,%ecx                # 4*sizeof(%xmm0)/sizeof(%rax)
2902         .long   0xa548f3fc              # cld; rep movsq
2903         lea     0x58(%rax),%rax         # adjust stack pointer
2904         jmp     .Lcommon_seh_tail
2905
2906 .Lrestore_cbc_rax:
2907         mov     120($context),%rax
2908
2909 .Lcommon_seh_tail:
2910         mov     8(%rax),%rdi
2911         mov     16(%rax),%rsi
2912         mov     %rax,152($context)      # restore context->Rsp
2913         mov     %rsi,168($context)      # restore context->Rsi
2914         mov     %rdi,176($context)      # restore context->Rdi
2915
2916         mov     40($disp),%rdi          # disp->ContextRecord
2917         mov     $context,%rsi           # context
2918         mov     \$154,%ecx              # sizeof(CONTEXT)
2919         .long   0xa548f3fc              # cld; rep movsq
2920
2921         mov     $disp,%rsi
2922         xor     %rcx,%rcx               # arg1, UNW_FLAG_NHANDLER
2923         mov     8(%rsi),%rdx            # arg2, disp->ImageBase
2924         mov     0(%rsi),%r8             # arg3, disp->ControlPc
2925         mov     16(%rsi),%r9            # arg4, disp->FunctionEntry
2926         mov     40(%rsi),%r10           # disp->ContextRecord
2927         lea     56(%rsi),%r11           # &disp->HandlerData
2928         lea     24(%rsi),%r12           # &disp->EstablisherFrame
2929         mov     %r10,32(%rsp)           # arg5
2930         mov     %r11,40(%rsp)           # arg6
2931         mov     %r12,48(%rsp)           # arg7
2932         mov     %rcx,56(%rsp)           # arg8, (NULL)
2933         call    *__imp_RtlVirtualUnwind(%rip)
2934
2935         mov     \$1,%eax                # ExceptionContinueSearch
2936         add     \$64,%rsp
2937         popfq
2938         pop     %r15
2939         pop     %r14
2940         pop     %r13
2941         pop     %r12
2942         pop     %rbp
2943         pop     %rbx
2944         pop     %rdi
2945         pop     %rsi
2946         ret
2947 .size   cbc_se_handler,.-cbc_se_handler
2948
2949 .section        .pdata
2950 .align  4
2951 ___
2952 $code.=<<___ if ($PREFIX eq "aesni");
2953         .rva    .LSEH_begin_aesni_ecb_encrypt
2954         .rva    .LSEH_end_aesni_ecb_encrypt
2955         .rva    .LSEH_info_ecb
2956
2957         .rva    .LSEH_begin_aesni_ccm64_encrypt_blocks
2958         .rva    .LSEH_end_aesni_ccm64_encrypt_blocks
2959         .rva    .LSEH_info_ccm64_enc
2960
2961         .rva    .LSEH_begin_aesni_ccm64_decrypt_blocks
2962         .rva    .LSEH_end_aesni_ccm64_decrypt_blocks
2963         .rva    .LSEH_info_ccm64_dec
2964
2965         .rva    .LSEH_begin_aesni_ctr32_encrypt_blocks
2966         .rva    .LSEH_end_aesni_ctr32_encrypt_blocks
2967         .rva    .LSEH_info_ctr32
2968
2969         .rva    .LSEH_begin_aesni_xts_encrypt
2970         .rva    .LSEH_end_aesni_xts_encrypt
2971         .rva    .LSEH_info_xts_enc
2972
2973         .rva    .LSEH_begin_aesni_xts_decrypt
2974         .rva    .LSEH_end_aesni_xts_decrypt
2975         .rva    .LSEH_info_xts_dec
2976 ___
2977 $code.=<<___;
2978         .rva    .LSEH_begin_${PREFIX}_cbc_encrypt
2979         .rva    .LSEH_end_${PREFIX}_cbc_encrypt
2980         .rva    .LSEH_info_cbc
2981
2982         .rva    ${PREFIX}_set_decrypt_key
2983         .rva    .LSEH_end_set_decrypt_key
2984         .rva    .LSEH_info_key
2985
2986         .rva    ${PREFIX}_set_encrypt_key
2987         .rva    .LSEH_end_set_encrypt_key
2988         .rva    .LSEH_info_key
2989 .section        .xdata
2990 .align  8
2991 ___
2992 $code.=<<___ if ($PREFIX eq "aesni");
2993 .LSEH_info_ecb:
2994         .byte   9,0,0,0
2995         .rva    ecb_se_handler
2996 .LSEH_info_ccm64_enc:
2997         .byte   9,0,0,0
2998         .rva    ccm64_se_handler
2999         .rva    .Lccm64_enc_body,.Lccm64_enc_ret        # HandlerData[]
3000 .LSEH_info_ccm64_dec:
3001         .byte   9,0,0,0
3002         .rva    ccm64_se_handler
3003         .rva    .Lccm64_dec_body,.Lccm64_dec_ret        # HandlerData[]
3004 .LSEH_info_ctr32:
3005         .byte   9,0,0,0
3006         .rva    ctr32_se_handler
3007 .LSEH_info_xts_enc:
3008         .byte   9,0,0,0
3009         .rva    xts_se_handler
3010         .rva    .Lxts_enc_body,.Lxts_enc_epilogue       # HandlerData[]
3011 .LSEH_info_xts_dec:
3012         .byte   9,0,0,0
3013         .rva    xts_se_handler
3014         .rva    .Lxts_dec_body,.Lxts_dec_epilogue       # HandlerData[]
3015 ___
3016 $code.=<<___;
3017 .LSEH_info_cbc:
3018         .byte   9,0,0,0
3019         .rva    cbc_se_handler
3020 .LSEH_info_key:
3021         .byte   0x01,0x04,0x01,0x00
3022         .byte   0x04,0x02,0x00,0x00     # sub rsp,8
3023 ___
3024 }
3025
3026 sub rex {
3027   local *opcode=shift;
3028   my ($dst,$src)=@_;
3029   my $rex=0;
3030
3031     $rex|=0x04                  if($dst>=8);
3032     $rex|=0x01                  if($src>=8);
3033     push @opcode,$rex|0x40      if($rex);
3034 }
3035
3036 sub aesni {
3037   my $line=shift;
3038   my @opcode=(0x66);
3039
3040     if ($line=~/(aeskeygenassist)\s+\$([x0-9a-f]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) {
3041         rex(\@opcode,$4,$3);
3042         push @opcode,0x0f,0x3a,0xdf;
3043         push @opcode,0xc0|($3&7)|(($4&7)<<3);   # ModR/M
3044         my $c=$2;
3045         push @opcode,$c=~/^0/?oct($c):$c;
3046         return ".byte\t".join(',',@opcode);
3047     }
3048     elsif ($line=~/(aes[a-z]+)\s+%xmm([0-9]+),\s*%xmm([0-9]+)/) {
3049         my %opcodelet = (
3050                 "aesimc" => 0xdb,
3051                 "aesenc" => 0xdc,       "aesenclast" => 0xdd,
3052                 "aesdec" => 0xde,       "aesdeclast" => 0xdf
3053         );
3054         return undef if (!defined($opcodelet{$1}));
3055         rex(\@opcode,$3,$2);
3056         push @opcode,0x0f,0x38,$opcodelet{$1};
3057         push @opcode,0xc0|($2&7)|(($3&7)<<3);   # ModR/M
3058         return ".byte\t".join(',',@opcode);
3059     }
3060     return $line;
3061 }
3062
3063 $code =~ s/\`([^\`]*)\`/eval($1)/gem;
3064 $code =~ s/\b(aes.*%xmm[0-9]+).*$/aesni($1)/gem;
3065
3066 print $code;
3067
3068 close STDOUT;