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