crypto/modes/asm/aesni-gcm-x86_64.pl

   1 #!/usr/bin/env perl
   2 #
   3 # ====================================================================
   4 # Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
   5 # project. The module is, however, dual licensed under OpenSSL and
   6 # CRYPTOGAMS licenses depending on where you obtain it. For further
   7 # details see http://www.openssl.org/~appro/cryptogams/.
   8 # ====================================================================
   9 #
  10 #
  11 # AES-NI-CTR+GHASH stitch.
  12 #
  13 # February 2013
  14 #
  15 # OpenSSL GCM implementation is organized in such way that its
  16 # performance is rather close to the sum of its streamed components,
  17 # in the context parallelized AES-NI CTR and modulo-scheduled
  18 # PCLMULQDQ-enabled GHASH. Unfortunately, as no stitch implementation
  19 # was observed to perform significantly better than the sum of the
  20 # components on contemporary CPUs, the effort was deemed impossible to
  21 # justify. This module is based on combination of Intel submissions,
  22 # [1] and [2], with MOVBE twist suggested by Ilya Albrekht and Max
  23 # Locktyukhin of Intel Corp. who verified that it reduces shuffles
  24 # pressure with notable relative improvement on upcoming Haswell
  25 # processor. [Exact performance numbers to be added at launch.]
  26 #
  27 # [1] http://rt.openssl.org/Ticket/Display.html?id=2900&user=guest&pass=guest
  28 # [2] http://www.intel.com/content/dam/www/public/us/en/documents/software-support/enabling-high-performance-gcm.pdf
  29
  30 $flavour = shift;
  31 $output  = shift;
  32 if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
  33
  34 $win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);
  35
  36 $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
  37 ( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
  38 ( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or
  39 die "can't locate x86_64-xlate.pl";
  40
  41 if (`$ENV{CC} -Wa,-v -c -o /dev/null -x assembler /dev/null 2>&1`
  42                 =~ /GNU assembler version ([2-9]\.[0-9]+)/) {
  43         $avx = ($1>=2.19) + ($1>=2.22);
  44 }
  45
  46 if (!$avx && $win64 && ($flavour =~ /nasm/ || $ENV{ASM} =~ /nasm/) &&
  47             `nasm -v 2>&1` =~ /NASM version ([2-9]\.[0-9]+)/) {
  48         $avx = ($1>=2.09) + ($1>=2.10);
  49 }
  50
  51 if (!$avx && $win64 && ($flavour =~ /masm/ || $ENV{ASM} =~ /ml64/) &&
  52             `ml64 2>&1` =~ /Version ([0-9]+)\./) {
  53         $avx = ($1>=10) + ($1>=11);
  54 }
  55
  56 open OUT,"| \"$^X\" $xlate $flavour $output";
  57 *STDOUT=*OUT;
  58
  59 if ($avx>1) {{{
  60
  61 ($inp,$out,$len,$key,$ivp,$Xip)=("%rdi","%rsi","%rdx","%rcx","%r8","%r9");
  62
  63 ($Ii,$T1,$T2,$Hkey,
  64  $Z0,$Z1,$Z2,$Z3,$Xi) = map("%xmm$_",(0..8));
  65
  66 ($inout0,$inout1,$inout2,$inout3,$inout4,$inout5,$rndkey) = map("%xmm$_",(9..15));
  67
  68 ($counter,$rounds,$ret,$const,$in0,$end0)=("%ebx","%ebp","%r10","%r11","%r14","%r15");
  69
  70 $code=<<___;
  71 .text
  72
  73 .type   _aesni_ctr32_ghash_6x,\@abi-omnipotent
  74 .align  32
  75 _aesni_ctr32_ghash_6x:
  76         vmovdqu         0x20($const),$T2        # borrow $T2, .Lone_msb
  77         sub             \$6,$len
  78         vpxor           $Z0,$Z0,$Z0             # $Z0   = 0
  79         vmovdqu         0x00-0x80($key),$rndkey
  80         vpaddb          $T2,$T1,$inout1
  81         vpaddb          $T2,$inout1,$inout2
  82         vpaddb          $T2,$inout2,$inout3
  83         vpaddb          $T2,$inout3,$inout4
  84         vpaddb          $T2,$inout4,$inout5
  85         vpxor           $rndkey,$T1,$inout0
  86         vmovdqu         $Z0,16+8(%rsp)          # "$Z3" = 0
  87         jmp             .Loop6x
  88
  89 .align  32
  90 .Loop6x:
  91         add             \$6<<24,$counter
  92         jc              .Lhandle_ctr32          # discard $inout[1-5]?
  93         vmovdqu         0x00-0x20($Xip),$Hkey   # $Hkey^1
  94           vpaddb        $T2,$inout5,$T1         # next counter value
  95           vpxor         $rndkey,$inout1,$inout1
  96           vpxor         $rndkey,$inout2,$inout2
  97
  98 .Lresume_ctr32:
  99         vmovdqu         $T1,($ivp)              # save next counter value
 100         vpclmulqdq      \$0x10,$Hkey,$Z3,$Z1
 101           vpxor         $rndkey,$inout3,$inout3
 102           vmovups       0x10-0x80($key),$T2     # borrow $T2 for $rndkey
 103         vpclmulqdq      \$0x01,$Hkey,$Z3,$Z2
 104         xor             %r12,%r12
 105         cmp             $in0,$end0
 106
 107           vaesenc       $T2,$inout0,$inout0
 108         vmovdqu         0x30+8(%rsp),$Ii        # I[4]
 109           vpxor         $rndkey,$inout4,$inout4
 110         vpclmulqdq      \$0x00,$Hkey,$Z3,$T1
 111           vaesenc       $T2,$inout1,$inout1
 112           vpxor         $rndkey,$inout5,$inout5
 113         setnc           %r12b
 114         vpclmulqdq      \$0x11,$Hkey,$Z3,$Z3
 115           vaesenc       $T2,$inout2,$inout2
 116         vmovdqu         0x10-0x20($Xip),$Hkey   # $Hkey^2
 117         neg             %r12
 118           vaesenc       $T2,$inout3,$inout3
 119          vpxor          $Z1,$Z2,$Z2
 120         vpclmulqdq      \$0x00,$Hkey,$Ii,$Z1
 121          vpxor          $Z0,$Xi,$Xi             # modulo-scheduled
 122           vaesenc       $T2,$inout4,$inout4
 123          vpxor          $Z1,$T1,$Z0
 124         and             \$0x60,%r12
 125           vmovups       0x20-0x80($key),$rndkey
 126         vpclmulqdq      \$0x10,$Hkey,$Ii,$T1
 127           vaesenc       $T2,$inout5,$inout5
 128
 129         vpclmulqdq      \$0x01,$Hkey,$Ii,$T2
 130         lea             ($in0,%r12),$in0
 131           vaesenc       $rndkey,$inout0,$inout0
 132          vpxor          16+8(%rsp),$Xi,$Xi      # modulo-scheduled [vpxor $Z3,$Xi,$Xi]
 133         vpclmulqdq      \$0x11,$Hkey,$Ii,$Hkey
 134          vmovdqu        0x40+8(%rsp),$Ii        # I[3]
 135           vaesenc       $rndkey,$inout1,$inout1
 136         movbe           0x58($in0),%r13
 137           vaesenc       $rndkey,$inout2,$inout2
 138         movbe           0x50($in0),%r12
 139           vaesenc       $rndkey,$inout3,$inout3
 140         mov             %r13,0x20+8(%rsp)
 141           vaesenc       $rndkey,$inout4,$inout4
 142         mov             %r12,0x28+8(%rsp)
 143         vmovdqu         0x30-0x20($Xip),$Z1     # borrow $Z1 for $Hkey^3
 144           vaesenc       $rndkey,$inout5,$inout5
 145
 146           vmovups       0x30-0x80($key),$rndkey
 147          vpxor          $T1,$Z2,$Z2
 148         vpclmulqdq      \$0x00,$Z1,$Ii,$T1
 149           vaesenc       $rndkey,$inout0,$inout0
 150          vpxor          $T2,$Z2,$Z2
 151         vpclmulqdq      \$0x10,$Z1,$Ii,$T2
 152           vaesenc       $rndkey,$inout1,$inout1
 153          vpxor          $Hkey,$Z3,$Z3
 154         vpclmulqdq      \$0x01,$Z1,$Ii,$Hkey
 155           vaesenc       $rndkey,$inout2,$inout2
 156         vpclmulqdq      \$0x11,$Z1,$Ii,$Z1
 157          vmovdqu        0x50+8(%rsp),$Ii        # I[2]
 158           vaesenc       $rndkey,$inout3,$inout3
 159           vaesenc       $rndkey,$inout4,$inout4
 160          vpxor          $T1,$Z0,$Z0
 161         vmovdqu         0x40-0x20($Xip),$T1     # borrow $T1 for $Hkey^4
 162           vaesenc       $rndkey,$inout5,$inout5
 163
 164           vmovups       0x40-0x80($key),$rndkey
 165          vpxor          $T2,$Z2,$Z2
 166         vpclmulqdq      \$0x00,$T1,$Ii,$T2
 167           vaesenc       $rndkey,$inout0,$inout0
 168          vpxor          $Hkey,$Z2,$Z2
 169         vpclmulqdq      \$0x10,$T1,$Ii,$Hkey
 170           vaesenc       $rndkey,$inout1,$inout1
 171         movbe           0x48($in0),%r13
 172          vpxor          $Z1,$Z3,$Z3
 173         vpclmulqdq      \$0x01,$T1,$Ii,$Z1
 174           vaesenc       $rndkey,$inout2,$inout2
 175         movbe           0x40($in0),%r12
 176         vpclmulqdq      \$0x11,$T1,$Ii,$T1
 177          vmovdqu        0x60+8(%rsp),$Ii        # I[1]
 178           vaesenc       $rndkey,$inout3,$inout3
 179         mov             %r13,0x30+8(%rsp)
 180           vaesenc       $rndkey,$inout4,$inout4
 181         mov             %r12,0x38+8(%rsp)
 182          vpxor          $T2,$Z0,$Z0
 183         vmovdqu         0x60-0x20($Xip),$T2     # borrow $T2 for $Hkey^5
 184           vaesenc       $rndkey,$inout5,$inout5
 185
 186           vmovups       0x50-0x80($key),$rndkey
 187          vpxor          $Hkey,$Z2,$Z2
 188         vpclmulqdq      \$0x00,$T2,$Ii,$Hkey
 189           vaesenc       $rndkey,$inout0,$inout0
 190          vpxor          $Z1,$Z2,$Z2
 191         vpclmulqdq      \$0x10,$T2,$Ii,$Z1
 192           vaesenc       $rndkey,$inout1,$inout1
 193         movbe           0x38($in0),%r13
 194          vpxor          $T1,$Z3,$Z3
 195         vpclmulqdq      \$0x01,$T2,$Ii,$T1
 196          vpxor          0x70+8(%rsp),$Xi,$Xi    # accumulate I[0]
 197           vaesenc       $rndkey,$inout2,$inout2
 198         movbe           0x30($in0),%r12
 199         vpclmulqdq      \$0x11,$T2,$Ii,$T2
 200           vaesenc       $rndkey,$inout3,$inout3
 201         mov             %r13,0x40+8(%rsp)
 202           vaesenc       $rndkey,$inout4,$inout4
 203         mov             %r12,0x48+8(%rsp)
 204          vpxor          $Hkey,$Z0,$Z0
 205          vmovdqu        0x70-0x20($Xip),$Hkey   # $Hkey^6
 206           vaesenc       $rndkey,$inout5,$inout5
 207
 208           vmovups       0x60-0x80($key),$rndkey
 209          vpxor          $Z1,$Z2,$Z2
 210         vpclmulqdq      \$0x10,$Hkey,$Xi,$Z1
 211           vaesenc       $rndkey,$inout0,$inout0
 212          vpxor          $T1,$Z2,$Z2
 213         vpclmulqdq      \$0x01,$Hkey,$Xi,$T1
 214           vaesenc       $rndkey,$inout1,$inout1
 215         movbe           0x28($in0),%r13
 216          vpxor          $T2,$Z3,$Z3
 217         vpclmulqdq      \$0x00,$Hkey,$Xi,$T2
 218           vaesenc       $rndkey,$inout2,$inout2
 219         movbe           0x20($in0),%r12
 220         vpclmulqdq      \$0x11,$Hkey,$Xi,$Xi
 221           vaesenc       $rndkey,$inout3,$inout3
 222         mov             %r13,0x50+8(%rsp)
 223           vaesenc       $rndkey,$inout4,$inout4
 224         mov             %r12,0x58+8(%rsp)
 225         vpxor           $Z1,$Z2,$Z2
 226           vaesenc       $rndkey,$inout5,$inout5
 227         vpxor           $T1,$Z2,$Z2
 228
 229           vmovups       0x70-0x80($key),$rndkey
 230         vpslldq         \$8,$Z2,$Z1
 231         vpxor           $T2,$Z0,$Z0
 232         vmovdqu         0x10($const),$Hkey      # .Lpoly
 233
 234           vaesenc       $rndkey,$inout0,$inout0
 235         vpxor           $Xi,$Z3,$Z3
 236           vaesenc       $rndkey,$inout1,$inout1
 237         vpxor           $Z1,$Z0,$Z0
 238         movbe           0x18($in0),%r13
 239           vaesenc       $rndkey,$inout2,$inout2
 240         movbe           0x10($in0),%r12
 241         vpalignr        \$8,$Z0,$Z0,$Ii         # 1st phase
 242         vpclmulqdq      \$0x10,$Hkey,$Z0,$Z0
 243         mov             %r13,0x60+8(%rsp)
 244           vaesenc       $rndkey,$inout3,$inout3
 245         mov             %r12,0x68+8(%rsp)
 246           vaesenc       $rndkey,$inout4,$inout4
 247           vmovups       0x80-0x80($key),$T1     # borrow $T1 for $rndkey
 248           vaesenc       $rndkey,$inout5,$inout5
 249
 250           vaesenc       $T1,$inout0,$inout0
 251           vmovups       0x90-0x80($key),$rndkey
 252           vaesenc       $T1,$inout1,$inout1
 253         vpsrldq         \$8,$Z2,$Z2
 254           vaesenc       $T1,$inout2,$inout2
 255         vpxor           $Z2,$Z3,$Z3
 256           vaesenc       $T1,$inout3,$inout3
 257         vpxor           $Ii,$Z0,$Z0
 258         movbe           0x08($in0),%r13
 259           vaesenc       $T1,$inout4,$inout4
 260         movbe           0x00($in0),%r12
 261           vaesenc       $T1,$inout5,$inout5
 262           vmovups       0xa0-0x80($key),$T1
 263           cmp           \$11,$rounds
 264           jb            .Lenc_tail              # 128-bit key
 265
 266           vaesenc       $rndkey,$inout0,$inout0
 267           vaesenc       $rndkey,$inout1,$inout1
 268           vaesenc       $rndkey,$inout2,$inout2
 269           vaesenc       $rndkey,$inout3,$inout3
 270           vaesenc       $rndkey,$inout4,$inout4
 271           vaesenc       $rndkey,$inout5,$inout5
 272
 273           vaesenc       $T1,$inout0,$inout0
 274           vaesenc       $T1,$inout1,$inout1
 275           vaesenc       $T1,$inout2,$inout2
 276           vaesenc       $T1,$inout3,$inout3
 277           vaesenc       $T1,$inout4,$inout4
 278           vmovups       0xb0-0x80($key),$rndkey
 279           vaesenc       $T1,$inout5,$inout5
 280           vmovups       0xc0-0x80($key),$T1
 281           je            .Lenc_tail              # 192-bit key
 282
 283           vaesenc       $rndkey,$inout0,$inout0
 284           vaesenc       $rndkey,$inout1,$inout1
 285           vaesenc       $rndkey,$inout2,$inout2
 286           vaesenc       $rndkey,$inout3,$inout3
 287           vaesenc       $rndkey,$inout4,$inout4
 288           vaesenc       $rndkey,$inout5,$inout5
 289
 290           vaesenc       $T1,$inout0,$inout0
 291           vaesenc       $T1,$inout1,$inout1
 292           vaesenc       $T1,$inout2,$inout2
 293           vaesenc       $T1,$inout3,$inout3
 294           vaesenc       $T1,$inout4,$inout4
 295           vmovups       0xd0-0x80($key),$rndkey
 296           vaesenc       $T1,$inout5,$inout5
 297           vmovups       0xe0-0x80($key),$T1
 298           jmp           .Lenc_tail              # 256-bit key
 299
 300 .align  32
 301 .Lhandle_ctr32:
 302         vmovdqu         ($const),$Ii            # borrow $Ii for .Lbswap_mask
 303           vpshufb       $Ii,$T1,$Z2             # byte-swap counter
 304           vmovdqu       0x30($const),$Z1        # borrow $Z1, .Ltwo_lsb
 305           vpaddd        0x40($const),$Z2,$inout1        # .Lone_lsb
 306           vpaddd        $Z1,$Z2,$inout2
 307         vmovdqu         0x00-0x20($Xip),$Hkey   # $Hkey^1
 308           vpaddd        $Z1,$inout1,$inout3
 309           vpshufb       $Ii,$inout1,$inout1
 310           vpaddd        $Z1,$inout2,$inout4
 311           vpshufb       $Ii,$inout2,$inout2
 312           vpxor         $rndkey,$inout1,$inout1
 313           vpaddd        $Z1,$inout3,$inout5
 314           vpshufb       $Ii,$inout3,$inout3
 315           vpxor         $rndkey,$inout2,$inout2
 316           vpaddd        $Z1,$inout4,$T1         # byte-swapped next counter value
 317           vpshufb       $Ii,$inout4,$inout4
 318           vpshufb       $Ii,$inout5,$inout5
 319           vpshufb       $Ii,$T1,$T1             # next counter value
 320         jmp             .Lresume_ctr32
 321
 322 .align  32
 323 .Lenc_tail:
 324           vaesenc       $rndkey,$inout0,$inout0
 325         vmovdqu         $Z3,16+8(%rsp)          # postpone vpxor $Z3,$Xi,$Xi
 326         vpalignr        \$8,$Z0,$Z0,$Xi         # 2nd phase
 327           vaesenc       $rndkey,$inout1,$inout1
 328         vpclmulqdq      \$0x10,$Hkey,$Z0,$Z0
 329           vpxor         0x00($inp),$T1,$T2
 330           vaesenc       $rndkey,$inout2,$inout2
 331           vpxor         0x10($inp),$T1,$Ii
 332           vaesenc       $rndkey,$inout3,$inout3
 333           vpxor         0x20($inp),$T1,$Z1
 334           vaesenc       $rndkey,$inout4,$inout4
 335           vpxor         0x30($inp),$T1,$Z2
 336           vaesenc       $rndkey,$inout5,$inout5
 337           vpxor         0x40($inp),$T1,$Z3
 338           vpxor         0x50($inp),$T1,$Hkey
 339           vmovdqu       ($ivp),$T1              # load next counter value
 340
 341           vaesenclast   $T2,$inout0,$inout0
 342           vmovdqu       0x20($const),$T2        # borrow $T2, .Lone_msb
 343           vaesenclast   $Ii,$inout1,$inout1
 344          vpaddb         $T2,$T1,$Ii
 345         mov             %r13,0x70+8(%rsp)
 346         lea             0x60($inp),$inp
 347           vaesenclast   $Z1,$inout2,$inout2
 348          vpaddb         $T2,$Ii,$Z1
 349         mov             %r12,0x78+8(%rsp)
 350         lea             0x60($out),$out
 351           vmovdqu       0x00-0x80($key),$rndkey
 352           vaesenclast   $Z2,$inout3,$inout3
 353          vpaddb         $T2,$Z1,$Z2
 354           vaesenclast   $Z3, $inout4,$inout4
 355          vpaddb         $T2,$Z2,$Z3
 356           vaesenclast   $Hkey,$inout5,$inout5
 357          vpaddb         $T2,$Z3,$Hkey
 358
 359         add             \$0x60,$ret
 360         sub             \$0x6,$len
 361         jc              .L6x_done
 362
 363           vmovups       $inout0,-0x60($out)     # save output
 364          vpxor          $rndkey,$T1,$inout0
 365           vmovups       $inout1,-0x50($out)
 366          vmovdqa        $Ii,$inout1             # 0 latency
 367           vmovups       $inout2,-0x40($out)
 368          vmovdqa        $Z1,$inout2             # 0 latency
 369           vmovups       $inout3,-0x30($out)
 370          vmovdqa        $Z2,$inout3             # 0 latency
 371           vmovups       $inout4,-0x20($out)
 372          vmovdqa        $Z3,$inout4             # 0 latency
 373           vmovups       $inout5,-0x10($out)
 374          vmovdqa        $Hkey,$inout5           # 0 latency
 375         vmovdqu         0x20+8(%rsp),$Z3        # I[5]
 376         jmp             .Loop6x
 377
 378 .L6x_done:
 379         vpxor           16+8(%rsp),$Xi,$Xi      # modulo-scheduled
 380         vpxor           $Z0,$Xi,$Xi             # modulo-scheduled
 381
 382         ret
 383 .size   _aesni_ctr32_ghash_6x,.-_aesni_ctr32_ghash_6x
 384 ___
 385 ######################################################################
 386 #
 387 # size_t aesni_gcm_[en|de]crypt(const void *inp, void *out, size_t len,
 388 #               const AES_KEY *key, unsigned char iv[16],
 389 #               struct { u128 Xi,H,Htbl[9]; } *Xip);
 390 $code.=<<___;
 391 .globl  aesni_gcm_decrypt
 392 .type   aesni_gcm_decrypt,\@function,6
 393 .align  32
 394 aesni_gcm_decrypt:
 395         xor     $ret,$ret
 396         cmp     \$0x60,$len                     # minimal accepted length
 397         jb      .Lgcm_dec_abort
 398
 399         lea     (%rsp),%rax                     # save stack pointer
 400         push    %rbx
 401         push    %rbp
 402         push    %r12
 403         push    %r13
 404         push    %r14
 405         push    %r15
 406 ___
 407 $code.=<<___ if ($win64);
 408         lea     -0xa8(%rsp),%rsp
 409         movaps  %xmm6,-0xd8(%rax)
 410         movaps  %xmm7,-0xc8(%rax)
 411         movaps  %xmm8,-0xb8(%rax)
 412         movaps  %xmm9,-0xa8(%rax)
 413         movaps  %xmm10,-0x98(%rax)
 414         movaps  %xmm11,-0x88(%rax)
 415         movaps  %xmm12,-0x78(%rax)
 416         movaps  %xmm13,-0x68(%rax)
 417         movaps  %xmm14,-0x58(%rax)
 418         movaps  %xmm15,-0x48(%rax)
 419 .Lgcm_dec_body:
 420 ___
 421 $code.=<<___;
 422         vzeroupper
 423
 424         vmovdqu         ($ivp),$T1              # input counter value
 425         sub             \$128,%rsp
 426         mov             12($ivp),$counter
 427         lea             .Lbswap_mask(%rip),$const
 428         vmovdqu         ($Xip),$Xi              # load Xi
 429         and             \$-64,%rsp              # ensure stack alignment
 430         vmovdqu         ($const),$Ii            # borrow $Ii for .Lbswap_mask
 431         lea             0x80($key),$key         # size optimization
 432         lea             0x20+0x20($Xip),$Xip    # size optimization
 433         mov             0xf0-0x80($key),$rounds
 434         vpshufb         $Ii,$Xi,$Xi
 435
 436         vmovdqu         0x50($inp),$Z3          # I[5]
 437         lea             ($inp),$in0
 438         vmovdqu         0x40($inp),$Z0
 439         lea             -0xc0($inp,$len),$end0
 440         vmovdqu         0x30($inp),$Z1
 441         shr             \$4,$len
 442         xor             $ret,$ret
 443         vmovdqu         0x20($inp),$Z2
 444          vpshufb        $Ii,$Z3,$Z3             # passed to _aesni_ctr32_ghash_6x
 445         vmovdqu         0x10($inp),$T2
 446          vpshufb        $Ii,$Z0,$Z0
 447         vmovdqu         ($inp),$Hkey
 448          vpshufb        $Ii,$Z1,$Z1
 449         vmovdqu         $Z0,0x30(%rsp)
 450          vpshufb        $Ii,$Z2,$Z2
 451         vmovdqu         $Z1,0x40(%rsp)
 452          vpshufb        $Ii,$T2,$T2
 453         vmovdqu         $Z2,0x50(%rsp)
 454          vpshufb        $Ii,$Hkey,$Hkey
 455         vmovdqu         $T2,0x60(%rsp)
 456         vmovdqu         $Hkey,0x70(%rsp)
 457
 458         call            _aesni_ctr32_ghash_6x
 459
 460         vmovups         $inout0,-0x60($out)     # save output
 461         vmovups         $inout1,-0x50($out)
 462         vmovups         $inout2,-0x40($out)
 463         vmovups         $inout3,-0x30($out)
 464         vmovups         $inout4,-0x20($out)
 465         vmovups         $inout5,-0x10($out)
 466
 467         vpshufb         ($const),$Xi,$Xi        # .Lbswap_mask
 468         vmovdqu         $Xi,-0x40($Xip)         # output Xi
 469
 470         vzeroupper
 471 ___
 472 $code.=<<___ if ($win64);
 473         movaps  -0xd8(%rax),%xmm6
 474         movaps  -0xd8(%rax),%xmm7
 475         movaps  -0xb8(%rax),%xmm8
 476         movaps  -0xa8(%rax),%xmm9
 477         movaps  -0x98(%rax),%xmm10
 478         movaps  -0x88(%rax),%xmm11
 479         movaps  -0x78(%rax),%xmm12
 480         movaps  -0x68(%rax),%xmm13
 481         movaps  -0x58(%rax),%xmm14
 482         movaps  -0x48(%rax),%xmm15
 483 ___
 484 $code.=<<___;
 485         mov     -48(%rax),%r15
 486         mov     -40(%rax),%r14
 487         mov     -32(%rax),%r13
 488         mov     -24(%rax),%r12
 489         mov     -16(%rax),%rbp
 490         mov     -8(%rax),%rbx
 491         lea     (%rax),%rsp             # restore %rsp
 492 .Lgcm_dec_abort:
 493         mov     $ret,%rax               # return value
 494         ret
 495 .size   aesni_gcm_decrypt,.-aesni_gcm_decrypt
 496 ___
 497
 498 $code.=<<___;
 499 .type   _aesni_ctr32_6x,\@abi-omnipotent
 500 .align  32
 501 _aesni_ctr32_6x:
 502         vmovdqu         0x00-0x80($key),$Z0     # borrow $Z0 for $rndkey
 503         vmovdqu         0x20($const),$T2        # borrow $T2, .Lone_msb
 504         lea             -1($rounds),%r13
 505         vmovups         0x10-0x80($key),$rndkey
 506         lea             0x20-0x80($key),%r12
 507         vpxor           $Z0,$T1,$inout0
 508         add             \$6<<24,$counter
 509         jc              .Lhandle_ctr32_2
 510         vpaddb          $T2,$T1,$inout1
 511         vpaddb          $T2,$inout1,$inout2
 512         vpxor           $Z0,$inout1,$inout1
 513         vpaddb          $T2,$inout2,$inout3
 514         vpxor           $Z0,$inout2,$inout2
 515         vpaddb          $T2,$inout3,$inout4
 516         vpxor           $Z0,$inout3,$inout3
 517         vpaddb          $T2,$inout4,$inout5
 518         vpxor           $Z0,$inout4,$inout4
 519         vpaddb          $T2,$inout5,$T1
 520         vpxor           $Z0,$inout5,$inout5
 521         jmp             .Loop_ctr32
 522
 523 .align  16
 524 .Loop_ctr32:
 525         vaesenc         $rndkey,$inout0,$inout0
 526         vaesenc         $rndkey,$inout1,$inout1
 527         vaesenc         $rndkey,$inout2,$inout2
 528         vaesenc         $rndkey,$inout3,$inout3
 529         vaesenc         $rndkey,$inout4,$inout4
 530         vaesenc         $rndkey,$inout5,$inout5
 531         vmovups         (%r12),$rndkey
 532         lea             0x10(%r12),%r12
 533         dec             %r13d
 534         jnz             .Loop_ctr32
 535
 536         vmovdqu         (%r12),$Hkey            # last round key
 537         vaesenc         $rndkey,$inout0,$inout0
 538         vpxor           0x00($inp),$Hkey,$Z0
 539         vaesenc         $rndkey,$inout1,$inout1
 540         vpxor           0x10($inp),$Hkey,$Z1
 541         vaesenc         $rndkey,$inout2,$inout2
 542         vpxor           0x20($inp),$Hkey,$Z2
 543         vaesenc         $rndkey,$inout3,$inout3
 544         vpxor           0x30($inp),$Hkey,$Xi
 545         vaesenc         $rndkey,$inout4,$inout4
 546         vpxor           0x40($inp),$Hkey,$T2
 547         vaesenc         $rndkey,$inout5,$inout5
 548         vpxor           0x50($inp),$Hkey,$Hkey
 549         lea             0x60($inp),$inp
 550
 551         vaesenclast     $Z0,$inout0,$inout0
 552         vaesenclast     $Z1,$inout1,$inout1
 553         vaesenclast     $Z2,$inout2,$inout2
 554         vaesenclast     $Xi,$inout3,$inout3
 555         vaesenclast     $T2,$inout4,$inout4
 556         vaesenclast     $Hkey,$inout5,$inout5
 557         vmovups         $inout0,0x00($out)
 558         vmovups         $inout1,0x10($out)
 559         vmovups         $inout2,0x20($out)
 560         vmovups         $inout3,0x30($out)
 561         vmovups         $inout4,0x40($out)
 562         vmovups         $inout5,0x50($out)
 563         lea             0x60($out),$out
 564
 565         ret
 566 .align  32
 567 .Lhandle_ctr32_2:
 568         vpshufb         $Ii,$T1,$Z2             # byte-swap counter
 569         vmovdqu         0x30($const),$Z1        # borrow $Z1, .Ltwo_lsb
 570         vpaddd          0x40($const),$Z2,$inout1        # .Lone_lsb
 571         vpaddd          $Z1,$Z2,$inout2
 572         vpaddd          $Z1,$inout1,$inout3
 573         vpshufb         $Ii,$inout1,$inout1
 574         vpaddd          $Z1,$inout2,$inout4
 575         vpshufb         $Ii,$inout2,$inout2
 576         vpxor           $Z0,$inout1,$inout1
 577         vpaddd          $Z1,$inout3,$inout5
 578         vpshufb         $Ii,$inout3,$inout3
 579         vpxor           $Z0,$inout2,$inout2
 580         vpaddd          $Z1,$inout4,$T1         # byte-swapped next counter value
 581         vpshufb         $Ii,$inout4,$inout4
 582         vpxor           $Z0,$inout3,$inout3
 583         vpshufb         $Ii,$inout5,$inout5
 584         vpxor           $Z0,$inout4,$inout4
 585         vpshufb         $Ii,$T1,$T1             # next counter value
 586         vpxor           $Z0,$inout5,$inout5
 587         jmp     .Loop_ctr32
 588 .size   _aesni_ctr32_6x,.-_aesni_ctr32_6x
 589
 590 .globl  aesni_gcm_encrypt
 591 .type   aesni_gcm_encrypt,\@function,6
 592 .align  32
 593 aesni_gcm_encrypt:
 594         xor     $ret,$ret
 595         cmp     \$0x60*3,$len                   # minimal accepted length
 596         jb      .Lgcm_enc_abort
 597
 598         lea     (%rsp),%rax                     # save stack pointer
 599         push    %rbx
 600         push    %rbp
 601         push    %r12
 602         push    %r13
 603         push    %r14
 604         push    %r15
 605 ___
 606 $code.=<<___ if ($win64);
 607         lea     -0xa8(%rsp),%rsp
 608         movaps  %xmm6,-0xd8(%rax)
 609         movaps  %xmm7,-0xc8(%rax)
 610         movaps  %xmm8,-0xb8(%rax)
 611         movaps  %xmm9,-0xa8(%rax)
 612         movaps  %xmm10,-0x98(%rax)
 613         movaps  %xmm11,-0x88(%rax)
 614         movaps  %xmm12,-0x78(%rax)
 615         movaps  %xmm13,-0x68(%rax)
 616         movaps  %xmm14,-0x58(%rax)
 617         movaps  %xmm15,-0x48(%rax)
 618 .Lgcm_enc_body:
 619 ___
 620 $code.=<<___;
 621         vzeroupper
 622
 623         vmovdqu         ($ivp),$T1              # input counter value
 624         sub             \$128,%rsp
 625         mov             12($ivp),$counter
 626         lea             .Lbswap_mask(%rip),$const
 627         lea             0x80($key),$key         # size optimization
 628         vmovdqu         ($const),$Ii            # borrow $Ii for .Lbswap_mask
 629         and             \$-64,%rsp              # ensure stack alignment
 630         mov             0xf0-0x80($key),$rounds
 631
 632         lea             ($out),$in0
 633         lea             -0xc0($out,$len),$end0
 634         shr             \$4,$len
 635
 636         call            _aesni_ctr32_6x
 637         vpshufb         $Ii,$inout0,$Xi         # save bswapped output on stack
 638         vpshufb         $Ii,$inout1,$T2
 639         vmovdqu         $Xi,0x70(%rsp)
 640         vpshufb         $Ii,$inout2,$Z0
 641         vmovdqu         $T2,0x60(%rsp)
 642         vpshufb         $Ii,$inout3,$Z1
 643         vmovdqu         $Z0,0x50(%rsp)
 644         vpshufb         $Ii,$inout4,$Z2
 645         vmovdqu         $Z1,0x40(%rsp)
 646         vpshufb         $Ii,$inout5,$Z3         # passed to _aesni_ctr32_ghash_6x
 647         vmovdqu         $Z2,0x30(%rsp)
 648
 649         call            _aesni_ctr32_6x
 650
 651         vmovdqu         ($Xip),$Xi              # load Xi
 652         lea             0x20+0x20($Xip),$Xip    # size optimization
 653         sub             \$12,$len
 654         mov             \$0x60*2,$ret
 655         vpshufb         $Ii,$Xi,$Xi
 656
 657         call            _aesni_ctr32_ghash_6x
 658         vmovdqu         0x20(%rsp),$Z3          # I[5]
 659          vmovdqu        ($const),$Ii            # borrow $Ii for .Lbswap_mask
 660         vmovdqu         0x00-0x20($Xip),$Hkey   # $Hkey^1
 661         vpunpckhqdq     $Z3,$Z3,$T1
 662         vmovdqu         0x20-0x20($Xip),$rndkey # borrow $rndkey for $HK
 663          vmovups        $inout0,-0x60($out)     # save output
 664          vpshufb        $Ii,$inout0,$inout0     # but keep bswapped copy
 665         vpxor           $Z3,$T1,$T1
 666          vmovups        $inout1,-0x50($out)
 667          vpshufb        $Ii,$inout1,$inout1
 668          vmovups        $inout2,-0x40($out)
 669          vpshufb        $Ii,$inout2,$inout2
 670          vmovups        $inout3,-0x30($out)
 671          vpshufb        $Ii,$inout3,$inout3
 672          vmovups        $inout4,-0x20($out)
 673          vpshufb        $Ii,$inout4,$inout4
 674          vmovups        $inout5,-0x10($out)
 675          vpshufb        $Ii,$inout5,$inout5
 676          vmovdqu        $inout0,0x10(%rsp)      # free $inout0
 677 ___
 678 { my ($HK,$T3)=($rndkey,$inout0);
 679
 680 $code.=<<___;
 681          vmovdqu        0x30(%rsp),$Z2          # I[4]
 682          vmovdqu        0x10-0x20($Xip),$Ii     # borrow $Ii for $Hkey^2
 683          vpunpckhqdq    $Z2,$Z2,$T2
 684         vpclmulqdq      \$0x00,$Hkey,$Z3,$Z1
 685          vpxor          $Z2,$T2,$T2
 686         vpclmulqdq      \$0x11,$Hkey,$Z3,$Z3
 687         vpclmulqdq      \$0x00,$HK,$T1,$T1
 688
 689          vmovdqu        0x40(%rsp),$T3          # I[3]
 690         vpclmulqdq      \$0x00,$Ii,$Z2,$Z0
 691          vmovdqu        0x30-0x20($Xip),$Hkey   # $Hkey^3
 692         vpxor           $Z1,$Z0,$Z0
 693          vpunpckhqdq    $T3,$T3,$Z1
 694         vpclmulqdq      \$0x11,$Ii,$Z2,$Z2
 695          vpxor          $T3,$Z1,$Z1
 696         vpxor           $Z3,$Z2,$Z2
 697         vpclmulqdq      \$0x10,$HK,$T2,$T2
 698          vmovdqu        0x50-0x20($Xip),$HK
 699         vpxor           $T1,$T2,$T2
 700
 701          vmovdqu        0x50(%rsp),$T1          # I[2]
 702         vpclmulqdq      \$0x00,$Hkey,$T3,$Z3
 703          vmovdqu        0x40-0x20($Xip),$Ii     # borrow $Ii for $Hkey^4
 704         vpxor           $Z0,$Z3,$Z3
 705          vpunpckhqdq    $T1,$T1,$Z0
 706         vpclmulqdq      \$0x11,$Hkey,$T3,$T3
 707          vpxor          $T1,$Z0,$Z0
 708         vpxor           $Z2,$T3,$T3
 709         vpclmulqdq      \$0x00,$HK,$Z1,$Z1
 710         vpxor           $T2,$Z1,$Z1
 711
 712          vmovdqu        0x60(%rsp),$T2          # I[1]
 713         vpclmulqdq      \$0x00,$Ii,$T1,$Z2
 714          vmovdqu        0x60-0x20($Xip),$Hkey   # $Hkey^5
 715         vpxor           $Z3,$Z2,$Z2
 716          vpunpckhqdq    $T2,$T2,$Z3
 717         vpclmulqdq      \$0x11,$Ii,$T1,$T1
 718          vpxor          $T2,$Z3,$Z3
 719         vpxor           $T3,$T1,$T1
 720         vpclmulqdq      \$0x10,$HK,$Z0,$Z0
 721          vmovdqu        0x80-0x20($Xip),$HK
 722         vpxor           $Z1,$Z0,$Z0
 723
 724          vpxor          0x70(%rsp),$Xi,$Xi      # accumulate I[0]
 725         vpclmulqdq      \$0x00,$Hkey,$T2,$Z1
 726          vmovdqu        0x70-0x20($Xip),$Ii     # borrow $Ii for $Hkey^6
 727          vpunpckhqdq    $Xi,$Xi,$T3
 728         vpxor           $Z2,$Z1,$Z1
 729         vpclmulqdq      \$0x11,$Hkey,$T2,$T2
 730          vpxor          $Xi,$T3,$T3
 731         vpxor           $T1,$T2,$T2
 732         vpclmulqdq      \$0x00,$HK,$Z3,$Z3
 733         vpxor           $Z0,$Z3,$Z0
 734
 735         vpclmulqdq      \$0x00,$Ii,$Xi,$Z2
 736          vmovdqu        0x00-0x20($Xip),$Hkey   # $Hkey^1
 737          vpunpckhqdq    $inout5,$inout5,$T1
 738         vpclmulqdq      \$0x11,$Ii,$Xi,$Xi
 739          vpxor          $inout5,$T1,$T1
 740         vpxor           $Z1,$Z2,$Z1
 741         vpclmulqdq      \$0x10,$HK,$T3,$T3
 742          vmovdqu        0x20-0x20($Xip),$HK
 743         vpxor           $T2,$Xi,$Z3
 744         vpxor           $Z0,$T3,$Z2
 745
 746          vmovdqu        0x10-0x20($Xip),$Ii     # borrow $Ii for $Hkey^2
 747           vpxor         $Z1,$Z3,$T3             # aggregated Karatsuba post-processing
 748         vpclmulqdq      \$0x00,$Hkey,$inout5,$Z0
 749           vpxor         $T3,$Z2,$Z2
 750          vpunpckhqdq    $inout4,$inout4,$T2
 751         vpclmulqdq      \$0x11,$Hkey,$inout5,$inout5
 752          vpxor          $inout4,$T2,$T2
 753           vpslldq       \$8,$Z2,$T3
 754         vpclmulqdq      \$0x00,$HK,$T1,$T1
 755           vpxor         $T3,$Z1,$Xi
 756           vpsrldq       \$8,$Z2,$Z2
 757           vpxor         $Z2,$Z3,$Z3
 758
 759         vpclmulqdq      \$0x00,$Ii,$inout4,$Z1
 760          vmovdqu        0x30-0x20($Xip),$Hkey   # $Hkey^3
 761         vpxor           $Z0,$Z1,$Z1
 762          vpunpckhqdq    $inout3,$inout3,$T3
 763         vpclmulqdq      \$0x11,$Ii,$inout4,$inout4
 764          vpxor          $inout3,$T3,$T3
 765         vpxor           $inout5,$inout4,$inout4
 766           vpalignr      \$8,$Xi,$Xi,$inout5     # 1st phase
 767         vpclmulqdq      \$0x10,$HK,$T2,$T2
 768          vmovdqu        0x50-0x20($Xip),$HK
 769         vpxor           $T1,$T2,$T2
 770
 771         vpclmulqdq      \$0x00,$Hkey,$inout3,$Z0
 772          vmovdqu        0x40-0x20($Xip),$Ii     # borrow $Ii for $Hkey^4
 773         vpxor           $Z1,$Z0,$Z0
 774          vpunpckhqdq    $inout2,$inout2,$T1
 775         vpclmulqdq      \$0x11,$Hkey,$inout3,$inout3
 776          vpxor          $inout2,$T1,$T1
 777         vpxor           $inout4,$inout3,$inout3
 778           vxorps        0x10(%rsp),$Z3,$Z3      # accumulate $inout0
 779         vpclmulqdq      \$0x00,$HK,$T3,$T3
 780         vpxor           $T2,$T3,$T3
 781
 782           vpclmulqdq    \$0x10,0x10($const),$Xi,$Xi
 783           vxorps        $inout5,$Xi,$Xi
 784
 785         vpclmulqdq      \$0x00,$Ii,$inout2,$Z1
 786          vmovdqu        0x60-0x20($Xip),$Hkey   # $Hkey^5
 787         vpxor           $Z0,$Z1,$Z1
 788          vpunpckhqdq    $inout1,$inout1,$T2
 789         vpclmulqdq      \$0x11,$Ii,$inout2,$inout2
 790          vpxor          $inout1,$T2,$T2
 791           vpalignr      \$8,$Xi,$Xi,$inout5     # 2nd phase
 792         vpxor           $inout3,$inout2,$inout2
 793         vpclmulqdq      \$0x10,$HK,$T1,$T1
 794          vmovdqu        0x80-0x20($Xip),$HK
 795         vpxor           $T3,$T1,$T1
 796
 797           vxorps        $Z3,$inout5,$inout5
 798           vpclmulqdq    \$0x10,0x10($const),$Xi,$Xi
 799           vxorps        $inout5,$Xi,$Xi
 800
 801         vpclmulqdq      \$0x00,$Hkey,$inout1,$Z0
 802          vmovdqu        0x70-0x20($Xip),$Ii     # borrow $Ii for $Hkey^6
 803         vpxor           $Z1,$Z0,$Z0
 804          vpunpckhqdq    $Xi,$Xi,$T3
 805         vpclmulqdq      \$0x11,$Hkey,$inout1,$inout1
 806          vpxor          $Xi,$T3,$T3
 807         vpxor           $inout2,$inout1,$inout1
 808         vpclmulqdq      \$0x00,$HK,$T2,$T2
 809         vpxor           $T1,$T2,$T2
 810
 811         vpclmulqdq      \$0x00,$Ii,$Xi,$Z1
 812         vpclmulqdq      \$0x11,$Ii,$Xi,$Z3
 813         vpxor           $Z0,$Z1,$Z1
 814         vpclmulqdq      \$0x10,$HK,$T3,$Z2
 815         vpxor           $inout1,$Z3,$Z3
 816         vpxor           $T2,$Z2,$Z2
 817
 818         vpxor           $Z1,$Z3,$Z0             # aggregated Karatsuba post-processing
 819         vpxor           $Z0,$Z2,$Z2
 820         vpslldq         \$8,$Z2,$T1
 821         vmovdqu         0x10($const),$Hkey      # .Lpoly
 822         vpsrldq         \$8,$Z2,$Z2
 823         vpxor           $T1,$Z1,$Xi
 824         vpxor           $Z2,$Z3,$Z3
 825
 826         vpalignr        \$8,$Xi,$Xi,$T2         # 1st phase
 827         vpclmulqdq      \$0x10,$Hkey,$Xi,$Xi
 828         vpxor           $T2,$Xi,$Xi
 829
 830         vpalignr        \$8,$Xi,$Xi,$T2         # 2nd phase
 831         vpclmulqdq      \$0x10,$Hkey,$Xi,$Xi
 832         vpxor           $Z3,$T2,$T2
 833         vpxor           $T2,$Xi,$Xi
 834 ___
 835 }
 836 $code.=<<___;
 837         vpshufb         ($const),$Xi,$Xi        # .Lbswap_mask
 838         vmovdqu         $Xi,-0x40($Xip)         # output Xi
 839
 840         vzeroupper
 841 ___
 842 $code.=<<___ if ($win64);
 843         movaps  -0xd8(%rax),%xmm6
 844         movaps  -0xc8(%rax),%xmm7
 845         movaps  -0xb8(%rax),%xmm8
 846         movaps  -0xa8(%rax),%xmm9
 847         movaps  -0x98(%rax),%xmm10
 848         movaps  -0x88(%rax),%xmm11
 849         movaps  -0x78(%rax),%xmm12
 850         movaps  -0x68(%rax),%xmm13
 851         movaps  -0x58(%rax),%xmm14
 852         movaps  -0x48(%rax),%xmm15
 853 ___
 854 $code.=<<___;
 855         mov     -48(%rax),%r15
 856         mov     -40(%rax),%r14
 857         mov     -32(%rax),%r13
 858         mov     -24(%rax),%r12
 859         mov     -16(%rax),%rbp
 860         mov     -8(%rax),%rbx
 861         lea     (%rax),%rsp             # restore %rsp
 862 .Lgcm_enc_abort:
 863         mov     $ret,%rax               # return value
 864         ret
 865 .size   aesni_gcm_encrypt,.-aesni_gcm_encrypt
 866 ___
 867
 868 $code.=<<___;
 869 .align  64
 870 .Lbswap_mask:
 871         .byte   15,14,13,12,11,10,9,8,7,6,5,4,3,2,1,0
 872 .Lpoly:
 873         .byte   0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0xc2
 874 .Lone_msb:
 875         .byte   0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1
 876 .Ltwo_lsb:
 877         .byte   2,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0
 878 .Lone_lsb:
 879         .byte   1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0
 880 .asciz  "AES-NI GCM module for x86_64, CRYPTOGAMS by <appro\@openssl.org>"
 881 .align  64
 882 ___
 883 if ($win64) {
 884 $rec="%rcx";
 885 $frame="%rdx";
 886 $context="%r8";
 887 $disp="%r9";
 888
 889 $code.=<<___
 890 .extern __imp_RtlVirtualUnwind
 891 .type   gcm_se_handler,\@abi-omnipotent
 892 .align  16
 893 gcm_se_handler:
 894         push    %rsi
 895         push    %rdi
 896         push    %rbx
 897         push    %rbp
 898         push    %r12
 899         push    %r13
 900         push    %r14
 901         push    %r15
 902         pushfq
 903         sub     \$64,%rsp
 904
 905         mov     120($context),%rax      # pull context->Rax
 906         mov     248($context),%rbx      # pull context->Rip
 907
 908         mov     8($disp),%rsi           # disp->ImageBase
 909         mov     56($disp),%r11          # disp->HandlerData
 910
 911         mov     0(%r11),%r10d           # HandlerData[0]
 912         lea     (%rsi,%r10),%r10        # prologue label
 913         cmp     %r10,%rbx               # context->Rip<prologue label
 914         jb      .Lcommon_seh_tail
 915
 916         mov     152($context),%rax      # pull context->Rsp
 917
 918         mov     4(%r11),%r10d           # HandlerData[1]
 919         lea     (%rsi,%r10),%r10        # epilogue label
 920         cmp     %r10,%rbx               # context->Rip>=epilogue label
 921         jae     .Lcommon_seh_tail
 922
 923         mov     120($context),%rax      # pull context->Rax
 924
 925         mov     -48(%rax),%r15
 926         mov     -40(%rax),%r14
 927         mov     -32(%rax),%r13
 928         mov     -24(%rax),%r12
 929         mov     -16(%rax),%rbp
 930         mov     -8(%rax),%rbx
 931         mov     %r15,240($context)
 932         mov     %r14,232($context)
 933         mov     %r13,224($context)
 934         mov     %r12,216($context)
 935         mov     %rbp,160($context)
 936         mov     %rbx,144($context)
 937
 938         lea     -0xd8(%rax),%rsi        # %xmm save area
 939         lea     512($context),%rdi      # & context.Xmm6
 940         mov     \$20,%ecx               # 10*sizeof(%xmm0)/sizeof(%rax)
 941         .long   0xa548f3fc              # cld; rep movsq
 942
 943 .Lcommon_seh_tail:
 944         mov     8(%rax),%rdi
 945         mov     16(%rax),%rsi
 946         mov     %rax,152($context)      # restore context->Rsp
 947         mov     %rsi,168($context)      # restore context->Rsi
 948         mov     %rdi,176($context)      # restore context->Rdi
 949
 950         mov     40($disp),%rdi          # disp->ContextRecord
 951         mov     $context,%rsi           # context
 952         mov     \$154,%ecx              # sizeof(CONTEXT)
 953         .long   0xa548f3fc              # cld; rep movsq
 954
 955         mov     $disp,%rsi
 956         xor     %rcx,%rcx               # arg1, UNW_FLAG_NHANDLER
 957         mov     8(%rsi),%rdx            # arg2, disp->ImageBase
 958         mov     0(%rsi),%r8             # arg3, disp->ControlPc
 959         mov     16(%rsi),%r9            # arg4, disp->FunctionEntry
 960         mov     40(%rsi),%r10           # disp->ContextRecord
 961         lea     56(%rsi),%r11           # &disp->HandlerData
 962         lea     24(%rsi),%r12           # &disp->EstablisherFrame
 963         mov     %r10,32(%rsp)           # arg5
 964         mov     %r11,40(%rsp)           # arg6
 965         mov     %r12,48(%rsp)           # arg7
 966         mov     %rcx,56(%rsp)           # arg8, (NULL)
 967         call    *__imp_RtlVirtualUnwind(%rip)
 968
 969         mov     \$1,%eax                # ExceptionContinueSearch
 970         add     \$64,%rsp
 971         popfq
 972         pop     %r15
 973         pop     %r14
 974         pop     %r13
 975         pop     %r12
 976         pop     %rbp
 977         pop     %rbx
 978         pop     %rdi
 979         pop     %rsi
 980         ret
 981 .size   gcm_se_handler,.-gcm_se_handler
 982
 983 .section        .pdata
 984 .align  4
 985         .rva    .LSEH_begin_aesni_gcm_decrypt
 986         .rva    .LSEH_end_aesni_gcm_decrypt
 987         .rva    .LSEH_gcm_dec_info
 988
 989         .rva    .LSEH_begin_aesni_gcm_encrypt
 990         .rva    .LSEH_end_aesni_gcm_encrypt
 991         .rva    .LSEH_gcm_enc_info
 992 .section        .xdata
 993 .align  8
 994 .LSEH_gcm_dec_info:
 995         .byte   9,0,0,0
 996         .rva    gcm_se_handler
 997         .rva    .Lgcm_dec_body,.Lgcm_dec_abort
 998 .LSEH_gcm_enc_info:
 999         .byte   9,0,0,0
1000         .rva    gcm_se_handler
1001         .rva    .Lgcm_enc_body,.Lgcm_enc_abort
1002 ___
1003 }
1004 }}} else {{{
1005 $code=<<___;    # assembler is too old
1006 .text
1007
1008 .globl  aesni_gcm_encrypt
1009 .type   aesni_gcm_encrypt,\@abi-omnipotent
1010 aesni_gcm_encrypt:
1011         xor     %eax,%eax
1012         ret
1013 .size   aesni_gcm_encrypt,.-aesni_gcm_encrypt
1014
1015 .globl  aesni_gcm_decrypt
1016 .type   aesni_gcm_decrypt,\@abi-omnipotent
1017 aesni_gcm_decrypt:
1018         xor     %eax,%eax
1019         ret
1020 .size   aesni_gcm_decrypt,.-aesni_gcm_decrypt
1021 ___
1022 }}}
1023
1024 $code =~ s/\`([^\`]*)\`/eval($1)/gem;
1025
1026 print $code;
1027
1028 close STDOUT;