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

   1 #! /usr/bin/env perl
   2 # Copyright 2013-2016 The OpenSSL Project Authors. All Rights Reserved.
   3 #
   4 # Licensed under the Apache License 2.0 (the "License").  You may not use
   5 # this file except in compliance with the License.  You can obtain a copy
   6 # in the file LICENSE in the source distribution or at
   7 # https://www.openssl.org/source/license.html
   8
   9 #
  10 # ====================================================================
  11 # Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
  12 # project. The module is, however, dual licensed under OpenSSL and
  13 # CRYPTOGAMS licenses depending on where you obtain it. For further
  14 # details see http://www.openssl.org/~appro/cryptogams/.
  15 # ====================================================================
  16 #
  17 #
  18 # AES-NI-CTR+GHASH stitch.
  19 #
  20 # February 2013
  21 #
  22 # OpenSSL GCM implementation is organized in such way that its
  23 # performance is rather close to the sum of its streamed components,
  24 # in the context parallelized AES-NI CTR and modulo-scheduled
  25 # PCLMULQDQ-enabled GHASH. Unfortunately, as no stitch implementation
  26 # was observed to perform significantly better than the sum of the
  27 # components on contemporary CPUs, the effort was deemed impossible to
  28 # justify. This module is based on combination of Intel submissions,
  29 # [1] and [2], with MOVBE twist suggested by Ilya Albrekht and Max
  30 # Locktyukhin of Intel Corp. who verified that it reduces shuffles
  31 # pressure with notable relative improvement, achieving 1.0 cycle per
  32 # byte processed with 128-bit key on Haswell processor, 0.74 - on
  33 # Broadwell, 0.63 - on Skylake... [Mentioned results are raw profiled
  34 # measurements for favourable packet size, one divisible by 96.
  35 # Applications using the EVP interface will observe a few percent
  36 # worse performance.]
  37 #
  38 # Knights Landing processes 1 byte in 1.25 cycles (measured with EVP).
  39 #
  40 # [1] http://rt.openssl.org/Ticket/Display.html?id=2900&user=guest&pass=guest
  41 # [2] http://www.intel.com/content/dam/www/public/us/en/documents/software-support/enabling-high-performance-gcm.pdf
  42
  43 $flavour = shift;
  44 $output  = shift;
  45 if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
  46
  47 $win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);
  48
  49 $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
  50 ( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
  51 ( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or
  52 die "can't locate x86_64-xlate.pl";
  53
  54 if (`$ENV{CC} -Wa,-v -c -o /dev/null -x assembler /dev/null 2>&1`
  55                 =~ /GNU assembler version ([2-9]\.[0-9]+)/) {
  56         $avx = ($1>=2.20) + ($1>=2.22);
  57 }
  58
  59 if (!$avx && $win64 && ($flavour =~ /nasm/ || $ENV{ASM} =~ /nasm/) &&
  60             `nasm -v 2>&1` =~ /NASM version ([2-9]\.[0-9]+)/) {
  61         $avx = ($1>=2.09) + ($1>=2.10);
  62 }
  63
  64 if (!$avx && $win64 && ($flavour =~ /masm/ || $ENV{ASM} =~ /ml64/) &&
  65             `ml64 2>&1` =~ /Version ([0-9]+)\./) {
  66         $avx = ($1>=10) + ($1>=11);
  67 }
  68
  69 if (!$avx && `$ENV{CC} -v 2>&1` =~ /((?:^clang|LLVM) version|.*based on LLVM) ([3-9]\.[0-9]+)/) {
  70         $avx = ($2>=3.0) + ($2>3.0);
  71 }
  72
  73 open OUT,"| \"$^X\" \"$xlate\" $flavour \"$output\"";
  74 *STDOUT=*OUT;
  75
  76 if ($avx>1) {{{
  77
  78 ($inp,$out,$len,$key,$ivp,$Xip)=("%rdi","%rsi","%rdx","%rcx","%r8","%r9");
  79
  80 ($Ii,$T1,$T2,$Hkey,
  81  $Z0,$Z1,$Z2,$Z3,$Xi) = map("%xmm$_",(0..8));
  82
  83 ($inout0,$inout1,$inout2,$inout3,$inout4,$inout5,$rndkey) = map("%xmm$_",(9..15));
  84
  85 ($counter,$rounds,$ret,$const,$in0,$end0)=("%ebx","%ebp","%r10","%r11","%r14","%r15");
  86
  87 $code=<<___;
  88 .text
  89
  90 .type   _aesni_ctr32_ghash_6x,\@abi-omnipotent
  91 .align  32
  92 _aesni_ctr32_ghash_6x:
  93         vmovdqu         0x20($const),$T2        # borrow $T2, .Lone_msb
  94         sub             \$6,$len
  95         vpxor           $Z0,$Z0,$Z0             # $Z0   = 0
  96         vmovdqu         0x00-0x80($key),$rndkey
  97         vpaddb          $T2,$T1,$inout1
  98         vpaddb          $T2,$inout1,$inout2
  99         vpaddb          $T2,$inout2,$inout3
 100         vpaddb          $T2,$inout3,$inout4
 101         vpaddb          $T2,$inout4,$inout5
 102         vpxor           $rndkey,$T1,$inout0
 103         vmovdqu         $Z0,16+8(%rsp)          # "$Z3" = 0
 104         jmp             .Loop6x
 105
 106 .align  32
 107 .Loop6x:
 108         add             \$`6<<24`,$counter
 109         jc              .Lhandle_ctr32          # discard $inout[1-5]?
 110         vmovdqu         0x00-0x20($Xip),$Hkey   # $Hkey^1
 111           vpaddb        $T2,$inout5,$T1         # next counter value
 112           vpxor         $rndkey,$inout1,$inout1
 113           vpxor         $rndkey,$inout2,$inout2
 114
 115 .Lresume_ctr32:
 116         vmovdqu         $T1,($ivp)              # save next counter value
 117         vpclmulqdq      \$0x10,$Hkey,$Z3,$Z1
 118           vpxor         $rndkey,$inout3,$inout3
 119           vmovups       0x10-0x80($key),$T2     # borrow $T2 for $rndkey
 120         vpclmulqdq      \$0x01,$Hkey,$Z3,$Z2
 121         xor             %r12,%r12
 122         cmp             $in0,$end0
 123
 124           vaesenc       $T2,$inout0,$inout0
 125         vmovdqu         0x30+8(%rsp),$Ii        # I[4]
 126           vpxor         $rndkey,$inout4,$inout4
 127         vpclmulqdq      \$0x00,$Hkey,$Z3,$T1
 128           vaesenc       $T2,$inout1,$inout1
 129           vpxor         $rndkey,$inout5,$inout5
 130         setnc           %r12b
 131         vpclmulqdq      \$0x11,$Hkey,$Z3,$Z3
 132           vaesenc       $T2,$inout2,$inout2
 133         vmovdqu         0x10-0x20($Xip),$Hkey   # $Hkey^2
 134         neg             %r12
 135           vaesenc       $T2,$inout3,$inout3
 136          vpxor          $Z1,$Z2,$Z2
 137         vpclmulqdq      \$0x00,$Hkey,$Ii,$Z1
 138          vpxor          $Z0,$Xi,$Xi             # modulo-scheduled
 139           vaesenc       $T2,$inout4,$inout4
 140          vpxor          $Z1,$T1,$Z0
 141         and             \$0x60,%r12
 142           vmovups       0x20-0x80($key),$rndkey
 143         vpclmulqdq      \$0x10,$Hkey,$Ii,$T1
 144           vaesenc       $T2,$inout5,$inout5
 145
 146         vpclmulqdq      \$0x01,$Hkey,$Ii,$T2
 147         lea             ($in0,%r12),$in0
 148           vaesenc       $rndkey,$inout0,$inout0
 149          vpxor          16+8(%rsp),$Xi,$Xi      # modulo-scheduled [vpxor $Z3,$Xi,$Xi]
 150         vpclmulqdq      \$0x11,$Hkey,$Ii,$Hkey
 151          vmovdqu        0x40+8(%rsp),$Ii        # I[3]
 152           vaesenc       $rndkey,$inout1,$inout1
 153         movbe           0x58($in0),%r13
 154           vaesenc       $rndkey,$inout2,$inout2
 155         movbe           0x50($in0),%r12
 156           vaesenc       $rndkey,$inout3,$inout3
 157         mov             %r13,0x20+8(%rsp)
 158           vaesenc       $rndkey,$inout4,$inout4
 159         mov             %r12,0x28+8(%rsp)
 160         vmovdqu         0x30-0x20($Xip),$Z1     # borrow $Z1 for $Hkey^3
 161           vaesenc       $rndkey,$inout5,$inout5
 162
 163           vmovups       0x30-0x80($key),$rndkey
 164          vpxor          $T1,$Z2,$Z2
 165         vpclmulqdq      \$0x00,$Z1,$Ii,$T1
 166           vaesenc       $rndkey,$inout0,$inout0
 167          vpxor          $T2,$Z2,$Z2
 168         vpclmulqdq      \$0x10,$Z1,$Ii,$T2
 169           vaesenc       $rndkey,$inout1,$inout1
 170          vpxor          $Hkey,$Z3,$Z3
 171         vpclmulqdq      \$0x01,$Z1,$Ii,$Hkey
 172           vaesenc       $rndkey,$inout2,$inout2
 173         vpclmulqdq      \$0x11,$Z1,$Ii,$Z1
 174          vmovdqu        0x50+8(%rsp),$Ii        # I[2]
 175           vaesenc       $rndkey,$inout3,$inout3
 176           vaesenc       $rndkey,$inout4,$inout4
 177          vpxor          $T1,$Z0,$Z0
 178         vmovdqu         0x40-0x20($Xip),$T1     # borrow $T1 for $Hkey^4
 179           vaesenc       $rndkey,$inout5,$inout5
 180
 181           vmovups       0x40-0x80($key),$rndkey
 182          vpxor          $T2,$Z2,$Z2
 183         vpclmulqdq      \$0x00,$T1,$Ii,$T2
 184           vaesenc       $rndkey,$inout0,$inout0
 185          vpxor          $Hkey,$Z2,$Z2
 186         vpclmulqdq      \$0x10,$T1,$Ii,$Hkey
 187           vaesenc       $rndkey,$inout1,$inout1
 188         movbe           0x48($in0),%r13
 189          vpxor          $Z1,$Z3,$Z3
 190         vpclmulqdq      \$0x01,$T1,$Ii,$Z1
 191           vaesenc       $rndkey,$inout2,$inout2
 192         movbe           0x40($in0),%r12
 193         vpclmulqdq      \$0x11,$T1,$Ii,$T1
 194          vmovdqu        0x60+8(%rsp),$Ii        # I[1]
 195           vaesenc       $rndkey,$inout3,$inout3
 196         mov             %r13,0x30+8(%rsp)
 197           vaesenc       $rndkey,$inout4,$inout4
 198         mov             %r12,0x38+8(%rsp)
 199          vpxor          $T2,$Z0,$Z0
 200         vmovdqu         0x60-0x20($Xip),$T2     # borrow $T2 for $Hkey^5
 201           vaesenc       $rndkey,$inout5,$inout5
 202
 203           vmovups       0x50-0x80($key),$rndkey
 204          vpxor          $Hkey,$Z2,$Z2
 205         vpclmulqdq      \$0x00,$T2,$Ii,$Hkey
 206           vaesenc       $rndkey,$inout0,$inout0
 207          vpxor          $Z1,$Z2,$Z2
 208         vpclmulqdq      \$0x10,$T2,$Ii,$Z1
 209           vaesenc       $rndkey,$inout1,$inout1
 210         movbe           0x38($in0),%r13
 211          vpxor          $T1,$Z3,$Z3
 212         vpclmulqdq      \$0x01,$T2,$Ii,$T1
 213          vpxor          0x70+8(%rsp),$Xi,$Xi    # accumulate I[0]
 214           vaesenc       $rndkey,$inout2,$inout2
 215         movbe           0x30($in0),%r12
 216         vpclmulqdq      \$0x11,$T2,$Ii,$T2
 217           vaesenc       $rndkey,$inout3,$inout3
 218         mov             %r13,0x40+8(%rsp)
 219           vaesenc       $rndkey,$inout4,$inout4
 220         mov             %r12,0x48+8(%rsp)
 221          vpxor          $Hkey,$Z0,$Z0
 222          vmovdqu        0x70-0x20($Xip),$Hkey   # $Hkey^6
 223           vaesenc       $rndkey,$inout5,$inout5
 224
 225           vmovups       0x60-0x80($key),$rndkey
 226          vpxor          $Z1,$Z2,$Z2
 227         vpclmulqdq      \$0x10,$Hkey,$Xi,$Z1
 228           vaesenc       $rndkey,$inout0,$inout0
 229          vpxor          $T1,$Z2,$Z2
 230         vpclmulqdq      \$0x01,$Hkey,$Xi,$T1
 231           vaesenc       $rndkey,$inout1,$inout1
 232         movbe           0x28($in0),%r13
 233          vpxor          $T2,$Z3,$Z3
 234         vpclmulqdq      \$0x00,$Hkey,$Xi,$T2
 235           vaesenc       $rndkey,$inout2,$inout2
 236         movbe           0x20($in0),%r12
 237         vpclmulqdq      \$0x11,$Hkey,$Xi,$Xi
 238           vaesenc       $rndkey,$inout3,$inout3
 239         mov             %r13,0x50+8(%rsp)
 240           vaesenc       $rndkey,$inout4,$inout4
 241         mov             %r12,0x58+8(%rsp)
 242         vpxor           $Z1,$Z2,$Z2
 243           vaesenc       $rndkey,$inout5,$inout5
 244         vpxor           $T1,$Z2,$Z2
 245
 246           vmovups       0x70-0x80($key),$rndkey
 247         vpslldq         \$8,$Z2,$Z1
 248         vpxor           $T2,$Z0,$Z0
 249         vmovdqu         0x10($const),$Hkey      # .Lpoly
 250
 251           vaesenc       $rndkey,$inout0,$inout0
 252         vpxor           $Xi,$Z3,$Z3
 253           vaesenc       $rndkey,$inout1,$inout1
 254         vpxor           $Z1,$Z0,$Z0
 255         movbe           0x18($in0),%r13
 256           vaesenc       $rndkey,$inout2,$inout2
 257         movbe           0x10($in0),%r12
 258         vpalignr        \$8,$Z0,$Z0,$Ii         # 1st phase
 259         vpclmulqdq      \$0x10,$Hkey,$Z0,$Z0
 260         mov             %r13,0x60+8(%rsp)
 261           vaesenc       $rndkey,$inout3,$inout3
 262         mov             %r12,0x68+8(%rsp)
 263           vaesenc       $rndkey,$inout4,$inout4
 264           vmovups       0x80-0x80($key),$T1     # borrow $T1 for $rndkey
 265           vaesenc       $rndkey,$inout5,$inout5
 266
 267           vaesenc       $T1,$inout0,$inout0
 268           vmovups       0x90-0x80($key),$rndkey
 269           vaesenc       $T1,$inout1,$inout1
 270         vpsrldq         \$8,$Z2,$Z2
 271           vaesenc       $T1,$inout2,$inout2
 272         vpxor           $Z2,$Z3,$Z3
 273           vaesenc       $T1,$inout3,$inout3
 274         vpxor           $Ii,$Z0,$Z0
 275         movbe           0x08($in0),%r13
 276           vaesenc       $T1,$inout4,$inout4
 277         movbe           0x00($in0),%r12
 278           vaesenc       $T1,$inout5,$inout5
 279           vmovups       0xa0-0x80($key),$T1
 280           cmp           \$11,$rounds
 281           jb            .Lenc_tail              # 128-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,$inout4
 295           vmovups       0xb0-0x80($key),$rndkey
 296           vaesenc       $T1,$inout5,$inout5
 297           vmovups       0xc0-0x80($key),$T1
 298           je            .Lenc_tail              # 192-bit key
 299
 300           vaesenc       $rndkey,$inout0,$inout0
 301           vaesenc       $rndkey,$inout1,$inout1
 302           vaesenc       $rndkey,$inout2,$inout2
 303           vaesenc       $rndkey,$inout3,$inout3
 304           vaesenc       $rndkey,$inout4,$inout4
 305           vaesenc       $rndkey,$inout5,$inout5
 306
 307           vaesenc       $T1,$inout0,$inout0
 308           vaesenc       $T1,$inout1,$inout1
 309           vaesenc       $T1,$inout2,$inout2
 310           vaesenc       $T1,$inout3,$inout3
 311           vaesenc       $T1,$inout4,$inout4
 312           vmovups       0xd0-0x80($key),$rndkey
 313           vaesenc       $T1,$inout5,$inout5
 314           vmovups       0xe0-0x80($key),$T1
 315           jmp           .Lenc_tail              # 256-bit key
 316
 317 .align  32
 318 .Lhandle_ctr32:
 319         vmovdqu         ($const),$Ii            # borrow $Ii for .Lbswap_mask
 320           vpshufb       $Ii,$T1,$Z2             # byte-swap counter
 321           vmovdqu       0x30($const),$Z1        # borrow $Z1, .Ltwo_lsb
 322           vpaddd        0x40($const),$Z2,$inout1        # .Lone_lsb
 323           vpaddd        $Z1,$Z2,$inout2
 324         vmovdqu         0x00-0x20($Xip),$Hkey   # $Hkey^1
 325           vpaddd        $Z1,$inout1,$inout3
 326           vpshufb       $Ii,$inout1,$inout1
 327           vpaddd        $Z1,$inout2,$inout4
 328           vpshufb       $Ii,$inout2,$inout2
 329           vpxor         $rndkey,$inout1,$inout1
 330           vpaddd        $Z1,$inout3,$inout5
 331           vpshufb       $Ii,$inout3,$inout3
 332           vpxor         $rndkey,$inout2,$inout2
 333           vpaddd        $Z1,$inout4,$T1         # byte-swapped next counter value
 334           vpshufb       $Ii,$inout4,$inout4
 335           vpshufb       $Ii,$inout5,$inout5
 336           vpshufb       $Ii,$T1,$T1             # next counter value
 337         jmp             .Lresume_ctr32
 338
 339 .align  32
 340 .Lenc_tail:
 341           vaesenc       $rndkey,$inout0,$inout0
 342         vmovdqu         $Z3,16+8(%rsp)          # postpone vpxor $Z3,$Xi,$Xi
 343         vpalignr        \$8,$Z0,$Z0,$Xi         # 2nd phase
 344           vaesenc       $rndkey,$inout1,$inout1
 345         vpclmulqdq      \$0x10,$Hkey,$Z0,$Z0
 346           vpxor         0x00($inp),$T1,$T2
 347           vaesenc       $rndkey,$inout2,$inout2
 348           vpxor         0x10($inp),$T1,$Ii
 349           vaesenc       $rndkey,$inout3,$inout3
 350           vpxor         0x20($inp),$T1,$Z1
 351           vaesenc       $rndkey,$inout4,$inout4
 352           vpxor         0x30($inp),$T1,$Z2
 353           vaesenc       $rndkey,$inout5,$inout5
 354           vpxor         0x40($inp),$T1,$Z3
 355           vpxor         0x50($inp),$T1,$Hkey
 356           vmovdqu       ($ivp),$T1              # load next counter value
 357
 358           vaesenclast   $T2,$inout0,$inout0
 359           vmovdqu       0x20($const),$T2        # borrow $T2, .Lone_msb
 360           vaesenclast   $Ii,$inout1,$inout1
 361          vpaddb         $T2,$T1,$Ii
 362         mov             %r13,0x70+8(%rsp)
 363         lea             0x60($inp),$inp
 364           vaesenclast   $Z1,$inout2,$inout2
 365          vpaddb         $T2,$Ii,$Z1
 366         mov             %r12,0x78+8(%rsp)
 367         lea             0x60($out),$out
 368           vmovdqu       0x00-0x80($key),$rndkey
 369           vaesenclast   $Z2,$inout3,$inout3
 370          vpaddb         $T2,$Z1,$Z2
 371           vaesenclast   $Z3, $inout4,$inout4
 372          vpaddb         $T2,$Z2,$Z3
 373           vaesenclast   $Hkey,$inout5,$inout5
 374          vpaddb         $T2,$Z3,$Hkey
 375
 376         add             \$0x60,$ret
 377         sub             \$0x6,$len
 378         jc              .L6x_done
 379
 380           vmovups       $inout0,-0x60($out)     # save output
 381          vpxor          $rndkey,$T1,$inout0
 382           vmovups       $inout1,-0x50($out)
 383          vmovdqa        $Ii,$inout1             # 0 latency
 384           vmovups       $inout2,-0x40($out)
 385          vmovdqa        $Z1,$inout2             # 0 latency
 386           vmovups       $inout3,-0x30($out)
 387          vmovdqa        $Z2,$inout3             # 0 latency
 388           vmovups       $inout4,-0x20($out)
 389          vmovdqa        $Z3,$inout4             # 0 latency
 390           vmovups       $inout5,-0x10($out)
 391          vmovdqa        $Hkey,$inout5           # 0 latency
 392         vmovdqu         0x20+8(%rsp),$Z3        # I[5]
 393         jmp             .Loop6x
 394
 395 .L6x_done:
 396         vpxor           16+8(%rsp),$Xi,$Xi      # modulo-scheduled
 397         vpxor           $Z0,$Xi,$Xi             # modulo-scheduled
 398
 399         ret
 400 .size   _aesni_ctr32_ghash_6x,.-_aesni_ctr32_ghash_6x
 401 ___
 402 ######################################################################
 403 #
 404 # size_t aesni_gcm_[en|de]crypt(const void *inp, void *out, size_t len,
 405 #               const AES_KEY *key, unsigned char iv[16],
 406 #               struct { u128 Xi,H,Htbl[9]; } *Xip);
 407 $code.=<<___;
 408 .globl  aesni_gcm_decrypt
 409 .type   aesni_gcm_decrypt,\@function,6
 410 .align  32
 411 aesni_gcm_decrypt:
 412 .cfi_startproc
 413         xor     $ret,$ret
 414         cmp     \$0x60,$len                     # minimal accepted length
 415         jb      .Lgcm_dec_abort
 416
 417         lea     (%rsp),%rax                     # save stack pointer
 418 .cfi_def_cfa_register   %rax
 419         push    %rbx
 420 .cfi_push       %rbx
 421         push    %rbp
 422 .cfi_push       %rbp
 423         push    %r12
 424 .cfi_push       %r12
 425         push    %r13
 426 .cfi_push       %r13
 427         push    %r14
 428 .cfi_push       %r14
 429         push    %r15
 430 .cfi_push       %r15
 431 ___
 432 $code.=<<___ if ($win64);
 433         lea     -0xa8(%rsp),%rsp
 434         movaps  %xmm6,-0xd8(%rax)
 435         movaps  %xmm7,-0xc8(%rax)
 436         movaps  %xmm8,-0xb8(%rax)
 437         movaps  %xmm9,-0xa8(%rax)
 438         movaps  %xmm10,-0x98(%rax)
 439         movaps  %xmm11,-0x88(%rax)
 440         movaps  %xmm12,-0x78(%rax)
 441         movaps  %xmm13,-0x68(%rax)
 442         movaps  %xmm14,-0x58(%rax)
 443         movaps  %xmm15,-0x48(%rax)
 444 .Lgcm_dec_body:
 445 ___
 446 $code.=<<___;
 447         vzeroupper
 448
 449         vmovdqu         ($ivp),$T1              # input counter value
 450         add             \$-128,%rsp
 451         mov             12($ivp),$counter
 452         lea             .Lbswap_mask(%rip),$const
 453         lea             -0x80($key),$in0        # borrow $in0
 454         mov             \$0xf80,$end0           # borrow $end0
 455         vmovdqu         ($Xip),$Xi              # load Xi
 456         and             \$-128,%rsp             # ensure stack alignment
 457         vmovdqu         ($const),$Ii            # borrow $Ii for .Lbswap_mask
 458         lea             0x80($key),$key         # size optimization
 459         lea             0x20+0x20($Xip),$Xip    # size optimization
 460         mov             0xf0-0x80($key),$rounds
 461         vpshufb         $Ii,$Xi,$Xi
 462
 463         and             $end0,$in0
 464         and             %rsp,$end0
 465         sub             $in0,$end0
 466         jc              .Ldec_no_key_aliasing
 467         cmp             \$768,$end0
 468         jnc             .Ldec_no_key_aliasing
 469         sub             $end0,%rsp              # avoid aliasing with key
 470 .Ldec_no_key_aliasing:
 471
 472         vmovdqu         0x50($inp),$Z3          # I[5]
 473         lea             ($inp),$in0
 474         vmovdqu         0x40($inp),$Z0
 475         lea             -0xc0($inp,$len),$end0
 476         vmovdqu         0x30($inp),$Z1
 477         shr             \$4,$len
 478         xor             $ret,$ret
 479         vmovdqu         0x20($inp),$Z2
 480          vpshufb        $Ii,$Z3,$Z3             # passed to _aesni_ctr32_ghash_6x
 481         vmovdqu         0x10($inp),$T2
 482          vpshufb        $Ii,$Z0,$Z0
 483         vmovdqu         ($inp),$Hkey
 484          vpshufb        $Ii,$Z1,$Z1
 485         vmovdqu         $Z0,0x30(%rsp)
 486          vpshufb        $Ii,$Z2,$Z2
 487         vmovdqu         $Z1,0x40(%rsp)
 488          vpshufb        $Ii,$T2,$T2
 489         vmovdqu         $Z2,0x50(%rsp)
 490          vpshufb        $Ii,$Hkey,$Hkey
 491         vmovdqu         $T2,0x60(%rsp)
 492         vmovdqu         $Hkey,0x70(%rsp)
 493
 494         call            _aesni_ctr32_ghash_6x
 495
 496         vmovups         $inout0,-0x60($out)     # save output
 497         vmovups         $inout1,-0x50($out)
 498         vmovups         $inout2,-0x40($out)
 499         vmovups         $inout3,-0x30($out)
 500         vmovups         $inout4,-0x20($out)
 501         vmovups         $inout5,-0x10($out)
 502
 503         vpshufb         ($const),$Xi,$Xi        # .Lbswap_mask
 504         vmovdqu         $Xi,-0x40($Xip)         # output Xi
 505
 506         vzeroupper
 507 ___
 508 $code.=<<___ if ($win64);
 509         movaps  -0xd8(%rax),%xmm6
 510         movaps  -0xc8(%rax),%xmm7
 511         movaps  -0xb8(%rax),%xmm8
 512         movaps  -0xa8(%rax),%xmm9
 513         movaps  -0x98(%rax),%xmm10
 514         movaps  -0x88(%rax),%xmm11
 515         movaps  -0x78(%rax),%xmm12
 516         movaps  -0x68(%rax),%xmm13
 517         movaps  -0x58(%rax),%xmm14
 518         movaps  -0x48(%rax),%xmm15
 519 ___
 520 $code.=<<___;
 521         mov     -48(%rax),%r15
 522 .cfi_restore    %r15
 523         mov     -40(%rax),%r14
 524 .cfi_restore    %r14
 525         mov     -32(%rax),%r13
 526 .cfi_restore    %r13
 527         mov     -24(%rax),%r12
 528 .cfi_restore    %r12
 529         mov     -16(%rax),%rbp
 530 .cfi_restore    %rbp
 531         mov     -8(%rax),%rbx
 532 .cfi_restore    %rbx
 533         lea     (%rax),%rsp             # restore %rsp
 534 .cfi_def_cfa_register   %rsp
 535 .Lgcm_dec_abort:
 536         mov     $ret,%rax               # return value
 537         ret
 538 .cfi_endproc
 539 .size   aesni_gcm_decrypt,.-aesni_gcm_decrypt
 540 ___
 541
 542 $code.=<<___;
 543 .type   _aesni_ctr32_6x,\@abi-omnipotent
 544 .align  32
 545 _aesni_ctr32_6x:
 546         vmovdqu         0x00-0x80($key),$Z0     # borrow $Z0 for $rndkey
 547         vmovdqu         0x20($const),$T2        # borrow $T2, .Lone_msb
 548         lea             -1($rounds),%r13
 549         vmovups         0x10-0x80($key),$rndkey
 550         lea             0x20-0x80($key),%r12
 551         vpxor           $Z0,$T1,$inout0
 552         add             \$`6<<24`,$counter
 553         jc              .Lhandle_ctr32_2
 554         vpaddb          $T2,$T1,$inout1
 555         vpaddb          $T2,$inout1,$inout2
 556         vpxor           $Z0,$inout1,$inout1
 557         vpaddb          $T2,$inout2,$inout3
 558         vpxor           $Z0,$inout2,$inout2
 559         vpaddb          $T2,$inout3,$inout4
 560         vpxor           $Z0,$inout3,$inout3
 561         vpaddb          $T2,$inout4,$inout5
 562         vpxor           $Z0,$inout4,$inout4
 563         vpaddb          $T2,$inout5,$T1
 564         vpxor           $Z0,$inout5,$inout5
 565         jmp             .Loop_ctr32
 566
 567 .align  16
 568 .Loop_ctr32:
 569         vaesenc         $rndkey,$inout0,$inout0
 570         vaesenc         $rndkey,$inout1,$inout1
 571         vaesenc         $rndkey,$inout2,$inout2
 572         vaesenc         $rndkey,$inout3,$inout3
 573         vaesenc         $rndkey,$inout4,$inout4
 574         vaesenc         $rndkey,$inout5,$inout5
 575         vmovups         (%r12),$rndkey
 576         lea             0x10(%r12),%r12
 577         dec             %r13d
 578         jnz             .Loop_ctr32
 579
 580         vmovdqu         (%r12),$Hkey            # last round key
 581         vaesenc         $rndkey,$inout0,$inout0
 582         vpxor           0x00($inp),$Hkey,$Z0
 583         vaesenc         $rndkey,$inout1,$inout1
 584         vpxor           0x10($inp),$Hkey,$Z1
 585         vaesenc         $rndkey,$inout2,$inout2
 586         vpxor           0x20($inp),$Hkey,$Z2
 587         vaesenc         $rndkey,$inout3,$inout3
 588         vpxor           0x30($inp),$Hkey,$Xi
 589         vaesenc         $rndkey,$inout4,$inout4
 590         vpxor           0x40($inp),$Hkey,$T2
 591         vaesenc         $rndkey,$inout5,$inout5
 592         vpxor           0x50($inp),$Hkey,$Hkey
 593         lea             0x60($inp),$inp
 594
 595         vaesenclast     $Z0,$inout0,$inout0
 596         vaesenclast     $Z1,$inout1,$inout1
 597         vaesenclast     $Z2,$inout2,$inout2
 598         vaesenclast     $Xi,$inout3,$inout3
 599         vaesenclast     $T2,$inout4,$inout4
 600         vaesenclast     $Hkey,$inout5,$inout5
 601         vmovups         $inout0,0x00($out)
 602         vmovups         $inout1,0x10($out)
 603         vmovups         $inout2,0x20($out)
 604         vmovups         $inout3,0x30($out)
 605         vmovups         $inout4,0x40($out)
 606         vmovups         $inout5,0x50($out)
 607         lea             0x60($out),$out
 608
 609         ret
 610 .align  32
 611 .Lhandle_ctr32_2:
 612         vpshufb         $Ii,$T1,$Z2             # byte-swap counter
 613         vmovdqu         0x30($const),$Z1        # borrow $Z1, .Ltwo_lsb
 614         vpaddd          0x40($const),$Z2,$inout1        # .Lone_lsb
 615         vpaddd          $Z1,$Z2,$inout2
 616         vpaddd          $Z1,$inout1,$inout3
 617         vpshufb         $Ii,$inout1,$inout1
 618         vpaddd          $Z1,$inout2,$inout4
 619         vpshufb         $Ii,$inout2,$inout2
 620         vpxor           $Z0,$inout1,$inout1
 621         vpaddd          $Z1,$inout3,$inout5
 622         vpshufb         $Ii,$inout3,$inout3
 623         vpxor           $Z0,$inout2,$inout2
 624         vpaddd          $Z1,$inout4,$T1         # byte-swapped next counter value
 625         vpshufb         $Ii,$inout4,$inout4
 626         vpxor           $Z0,$inout3,$inout3
 627         vpshufb         $Ii,$inout5,$inout5
 628         vpxor           $Z0,$inout4,$inout4
 629         vpshufb         $Ii,$T1,$T1             # next counter value
 630         vpxor           $Z0,$inout5,$inout5
 631         jmp     .Loop_ctr32
 632 .size   _aesni_ctr32_6x,.-_aesni_ctr32_6x
 633
 634 .globl  aesni_gcm_encrypt
 635 .type   aesni_gcm_encrypt,\@function,6
 636 .align  32
 637 aesni_gcm_encrypt:
 638 .cfi_startproc
 639         xor     $ret,$ret
 640         cmp     \$0x60*3,$len                   # minimal accepted length
 641         jb      .Lgcm_enc_abort
 642
 643         lea     (%rsp),%rax                     # save stack pointer
 644 .cfi_def_cfa_register   %rax
 645         push    %rbx
 646 .cfi_push       %rbx
 647         push    %rbp
 648 .cfi_push       %rbp
 649         push    %r12
 650 .cfi_push       %r12
 651         push    %r13
 652 .cfi_push       %r13
 653         push    %r14
 654 .cfi_push       %r14
 655         push    %r15
 656 .cfi_push       %r15
 657 ___
 658 $code.=<<___ if ($win64);
 659         lea     -0xa8(%rsp),%rsp
 660         movaps  %xmm6,-0xd8(%rax)
 661         movaps  %xmm7,-0xc8(%rax)
 662         movaps  %xmm8,-0xb8(%rax)
 663         movaps  %xmm9,-0xa8(%rax)
 664         movaps  %xmm10,-0x98(%rax)
 665         movaps  %xmm11,-0x88(%rax)
 666         movaps  %xmm12,-0x78(%rax)
 667         movaps  %xmm13,-0x68(%rax)
 668         movaps  %xmm14,-0x58(%rax)
 669         movaps  %xmm15,-0x48(%rax)
 670 .Lgcm_enc_body:
 671 ___
 672 $code.=<<___;
 673         vzeroupper
 674
 675         vmovdqu         ($ivp),$T1              # input counter value
 676         add             \$-128,%rsp
 677         mov             12($ivp),$counter
 678         lea             .Lbswap_mask(%rip),$const
 679         lea             -0x80($key),$in0        # borrow $in0
 680         mov             \$0xf80,$end0           # borrow $end0
 681         lea             0x80($key),$key         # size optimization
 682         vmovdqu         ($const),$Ii            # borrow $Ii for .Lbswap_mask
 683         and             \$-128,%rsp             # ensure stack alignment
 684         mov             0xf0-0x80($key),$rounds
 685
 686         and             $end0,$in0
 687         and             %rsp,$end0
 688         sub             $in0,$end0
 689         jc              .Lenc_no_key_aliasing
 690         cmp             \$768,$end0
 691         jnc             .Lenc_no_key_aliasing
 692         sub             $end0,%rsp              # avoid aliasing with key
 693 .Lenc_no_key_aliasing:
 694
 695         lea             ($out),$in0
 696         lea             -0xc0($out,$len),$end0
 697         shr             \$4,$len
 698
 699         call            _aesni_ctr32_6x
 700         vpshufb         $Ii,$inout0,$Xi         # save bswapped output on stack
 701         vpshufb         $Ii,$inout1,$T2
 702         vmovdqu         $Xi,0x70(%rsp)
 703         vpshufb         $Ii,$inout2,$Z0
 704         vmovdqu         $T2,0x60(%rsp)
 705         vpshufb         $Ii,$inout3,$Z1
 706         vmovdqu         $Z0,0x50(%rsp)
 707         vpshufb         $Ii,$inout4,$Z2
 708         vmovdqu         $Z1,0x40(%rsp)
 709         vpshufb         $Ii,$inout5,$Z3         # passed to _aesni_ctr32_ghash_6x
 710         vmovdqu         $Z2,0x30(%rsp)
 711
 712         call            _aesni_ctr32_6x
 713
 714         vmovdqu         ($Xip),$Xi              # load Xi
 715         lea             0x20+0x20($Xip),$Xip    # size optimization
 716         sub             \$12,$len
 717         mov             \$0x60*2,$ret
 718         vpshufb         $Ii,$Xi,$Xi
 719
 720         call            _aesni_ctr32_ghash_6x
 721         vmovdqu         0x20(%rsp),$Z3          # I[5]
 722          vmovdqu        ($const),$Ii            # borrow $Ii for .Lbswap_mask
 723         vmovdqu         0x00-0x20($Xip),$Hkey   # $Hkey^1
 724         vpunpckhqdq     $Z3,$Z3,$T1
 725         vmovdqu         0x20-0x20($Xip),$rndkey # borrow $rndkey for $HK
 726          vmovups        $inout0,-0x60($out)     # save output
 727          vpshufb        $Ii,$inout0,$inout0     # but keep bswapped copy
 728         vpxor           $Z3,$T1,$T1
 729          vmovups        $inout1,-0x50($out)
 730          vpshufb        $Ii,$inout1,$inout1
 731          vmovups        $inout2,-0x40($out)
 732          vpshufb        $Ii,$inout2,$inout2
 733          vmovups        $inout3,-0x30($out)
 734          vpshufb        $Ii,$inout3,$inout3
 735          vmovups        $inout4,-0x20($out)
 736          vpshufb        $Ii,$inout4,$inout4
 737          vmovups        $inout5,-0x10($out)
 738          vpshufb        $Ii,$inout5,$inout5
 739          vmovdqu        $inout0,0x10(%rsp)      # free $inout0
 740 ___
 741 { my ($HK,$T3)=($rndkey,$inout0);
 742
 743 $code.=<<___;
 744          vmovdqu        0x30(%rsp),$Z2          # I[4]
 745          vmovdqu        0x10-0x20($Xip),$Ii     # borrow $Ii for $Hkey^2
 746          vpunpckhqdq    $Z2,$Z2,$T2
 747         vpclmulqdq      \$0x00,$Hkey,$Z3,$Z1
 748          vpxor          $Z2,$T2,$T2
 749         vpclmulqdq      \$0x11,$Hkey,$Z3,$Z3
 750         vpclmulqdq      \$0x00,$HK,$T1,$T1
 751
 752          vmovdqu        0x40(%rsp),$T3          # I[3]
 753         vpclmulqdq      \$0x00,$Ii,$Z2,$Z0
 754          vmovdqu        0x30-0x20($Xip),$Hkey   # $Hkey^3
 755         vpxor           $Z1,$Z0,$Z0
 756          vpunpckhqdq    $T3,$T3,$Z1
 757         vpclmulqdq      \$0x11,$Ii,$Z2,$Z2
 758          vpxor          $T3,$Z1,$Z1
 759         vpxor           $Z3,$Z2,$Z2
 760         vpclmulqdq      \$0x10,$HK,$T2,$T2
 761          vmovdqu        0x50-0x20($Xip),$HK
 762         vpxor           $T1,$T2,$T2
 763
 764          vmovdqu        0x50(%rsp),$T1          # I[2]
 765         vpclmulqdq      \$0x00,$Hkey,$T3,$Z3
 766          vmovdqu        0x40-0x20($Xip),$Ii     # borrow $Ii for $Hkey^4
 767         vpxor           $Z0,$Z3,$Z3
 768          vpunpckhqdq    $T1,$T1,$Z0
 769         vpclmulqdq      \$0x11,$Hkey,$T3,$T3
 770          vpxor          $T1,$Z0,$Z0
 771         vpxor           $Z2,$T3,$T3
 772         vpclmulqdq      \$0x00,$HK,$Z1,$Z1
 773         vpxor           $T2,$Z1,$Z1
 774
 775          vmovdqu        0x60(%rsp),$T2          # I[1]
 776         vpclmulqdq      \$0x00,$Ii,$T1,$Z2
 777          vmovdqu        0x60-0x20($Xip),$Hkey   # $Hkey^5
 778         vpxor           $Z3,$Z2,$Z2
 779          vpunpckhqdq    $T2,$T2,$Z3
 780         vpclmulqdq      \$0x11,$Ii,$T1,$T1
 781          vpxor          $T2,$Z3,$Z3
 782         vpxor           $T3,$T1,$T1
 783         vpclmulqdq      \$0x10,$HK,$Z0,$Z0
 784          vmovdqu        0x80-0x20($Xip),$HK
 785         vpxor           $Z1,$Z0,$Z0
 786
 787          vpxor          0x70(%rsp),$Xi,$Xi      # accumulate I[0]
 788         vpclmulqdq      \$0x00,$Hkey,$T2,$Z1
 789          vmovdqu        0x70-0x20($Xip),$Ii     # borrow $Ii for $Hkey^6
 790          vpunpckhqdq    $Xi,$Xi,$T3
 791         vpxor           $Z2,$Z1,$Z1
 792         vpclmulqdq      \$0x11,$Hkey,$T2,$T2
 793          vpxor          $Xi,$T3,$T3
 794         vpxor           $T1,$T2,$T2
 795         vpclmulqdq      \$0x00,$HK,$Z3,$Z3
 796         vpxor           $Z0,$Z3,$Z0
 797
 798         vpclmulqdq      \$0x00,$Ii,$Xi,$Z2
 799          vmovdqu        0x00-0x20($Xip),$Hkey   # $Hkey^1
 800          vpunpckhqdq    $inout5,$inout5,$T1
 801         vpclmulqdq      \$0x11,$Ii,$Xi,$Xi
 802          vpxor          $inout5,$T1,$T1
 803         vpxor           $Z1,$Z2,$Z1
 804         vpclmulqdq      \$0x10,$HK,$T3,$T3
 805          vmovdqu        0x20-0x20($Xip),$HK
 806         vpxor           $T2,$Xi,$Z3
 807         vpxor           $Z0,$T3,$Z2
 808
 809          vmovdqu        0x10-0x20($Xip),$Ii     # borrow $Ii for $Hkey^2
 810           vpxor         $Z1,$Z3,$T3             # aggregated Karatsuba post-processing
 811         vpclmulqdq      \$0x00,$Hkey,$inout5,$Z0
 812           vpxor         $T3,$Z2,$Z2
 813          vpunpckhqdq    $inout4,$inout4,$T2
 814         vpclmulqdq      \$0x11,$Hkey,$inout5,$inout5
 815          vpxor          $inout4,$T2,$T2
 816           vpslldq       \$8,$Z2,$T3
 817         vpclmulqdq      \$0x00,$HK,$T1,$T1
 818           vpxor         $T3,$Z1,$Xi
 819           vpsrldq       \$8,$Z2,$Z2
 820           vpxor         $Z2,$Z3,$Z3
 821
 822         vpclmulqdq      \$0x00,$Ii,$inout4,$Z1
 823          vmovdqu        0x30-0x20($Xip),$Hkey   # $Hkey^3
 824         vpxor           $Z0,$Z1,$Z1
 825          vpunpckhqdq    $inout3,$inout3,$T3
 826         vpclmulqdq      \$0x11,$Ii,$inout4,$inout4
 827          vpxor          $inout3,$T3,$T3
 828         vpxor           $inout5,$inout4,$inout4
 829           vpalignr      \$8,$Xi,$Xi,$inout5     # 1st phase
 830         vpclmulqdq      \$0x10,$HK,$T2,$T2
 831          vmovdqu        0x50-0x20($Xip),$HK
 832         vpxor           $T1,$T2,$T2
 833
 834         vpclmulqdq      \$0x00,$Hkey,$inout3,$Z0
 835          vmovdqu        0x40-0x20($Xip),$Ii     # borrow $Ii for $Hkey^4
 836         vpxor           $Z1,$Z0,$Z0
 837          vpunpckhqdq    $inout2,$inout2,$T1
 838         vpclmulqdq      \$0x11,$Hkey,$inout3,$inout3
 839          vpxor          $inout2,$T1,$T1
 840         vpxor           $inout4,$inout3,$inout3
 841           vxorps        0x10(%rsp),$Z3,$Z3      # accumulate $inout0
 842         vpclmulqdq      \$0x00,$HK,$T3,$T3
 843         vpxor           $T2,$T3,$T3
 844
 845           vpclmulqdq    \$0x10,0x10($const),$Xi,$Xi
 846           vxorps        $inout5,$Xi,$Xi
 847
 848         vpclmulqdq      \$0x00,$Ii,$inout2,$Z1
 849          vmovdqu        0x60-0x20($Xip),$Hkey   # $Hkey^5
 850         vpxor           $Z0,$Z1,$Z1
 851          vpunpckhqdq    $inout1,$inout1,$T2
 852         vpclmulqdq      \$0x11,$Ii,$inout2,$inout2
 853          vpxor          $inout1,$T2,$T2
 854           vpalignr      \$8,$Xi,$Xi,$inout5     # 2nd phase
 855         vpxor           $inout3,$inout2,$inout2
 856         vpclmulqdq      \$0x10,$HK,$T1,$T1
 857          vmovdqu        0x80-0x20($Xip),$HK
 858         vpxor           $T3,$T1,$T1
 859
 860           vxorps        $Z3,$inout5,$inout5
 861           vpclmulqdq    \$0x10,0x10($const),$Xi,$Xi
 862           vxorps        $inout5,$Xi,$Xi
 863
 864         vpclmulqdq      \$0x00,$Hkey,$inout1,$Z0
 865          vmovdqu        0x70-0x20($Xip),$Ii     # borrow $Ii for $Hkey^6
 866         vpxor           $Z1,$Z0,$Z0
 867          vpunpckhqdq    $Xi,$Xi,$T3
 868         vpclmulqdq      \$0x11,$Hkey,$inout1,$inout1
 869          vpxor          $Xi,$T3,$T3
 870         vpxor           $inout2,$inout1,$inout1
 871         vpclmulqdq      \$0x00,$HK,$T2,$T2
 872         vpxor           $T1,$T2,$T2
 873
 874         vpclmulqdq      \$0x00,$Ii,$Xi,$Z1
 875         vpclmulqdq      \$0x11,$Ii,$Xi,$Z3
 876         vpxor           $Z0,$Z1,$Z1
 877         vpclmulqdq      \$0x10,$HK,$T3,$Z2
 878         vpxor           $inout1,$Z3,$Z3
 879         vpxor           $T2,$Z2,$Z2
 880
 881         vpxor           $Z1,$Z3,$Z0             # aggregated Karatsuba post-processing
 882         vpxor           $Z0,$Z2,$Z2
 883         vpslldq         \$8,$Z2,$T1
 884         vmovdqu         0x10($const),$Hkey      # .Lpoly
 885         vpsrldq         \$8,$Z2,$Z2
 886         vpxor           $T1,$Z1,$Xi
 887         vpxor           $Z2,$Z3,$Z3
 888
 889         vpalignr        \$8,$Xi,$Xi,$T2         # 1st phase
 890         vpclmulqdq      \$0x10,$Hkey,$Xi,$Xi
 891         vpxor           $T2,$Xi,$Xi
 892
 893         vpalignr        \$8,$Xi,$Xi,$T2         # 2nd phase
 894         vpclmulqdq      \$0x10,$Hkey,$Xi,$Xi
 895         vpxor           $Z3,$T2,$T2
 896         vpxor           $T2,$Xi,$Xi
 897 ___
 898 }
 899 $code.=<<___;
 900         vpshufb         ($const),$Xi,$Xi        # .Lbswap_mask
 901         vmovdqu         $Xi,-0x40($Xip)         # output Xi
 902
 903         vzeroupper
 904 ___
 905 $code.=<<___ if ($win64);
 906         movaps  -0xd8(%rax),%xmm6
 907         movaps  -0xc8(%rax),%xmm7
 908         movaps  -0xb8(%rax),%xmm8
 909         movaps  -0xa8(%rax),%xmm9
 910         movaps  -0x98(%rax),%xmm10
 911         movaps  -0x88(%rax),%xmm11
 912         movaps  -0x78(%rax),%xmm12
 913         movaps  -0x68(%rax),%xmm13
 914         movaps  -0x58(%rax),%xmm14
 915         movaps  -0x48(%rax),%xmm15
 916 ___
 917 $code.=<<___;
 918         mov     -48(%rax),%r15
 919 .cfi_restore    %r15
 920         mov     -40(%rax),%r14
 921 .cfi_restore    %r14
 922         mov     -32(%rax),%r13
 923 .cfi_restore    %r13
 924         mov     -24(%rax),%r12
 925 .cfi_restore    %r12
 926         mov     -16(%rax),%rbp
 927 .cfi_restore    %rbp
 928         mov     -8(%rax),%rbx
 929 .cfi_restore    %rbx
 930         lea     (%rax),%rsp             # restore %rsp
 931 .cfi_def_cfa_register   %rsp
 932 .Lgcm_enc_abort:
 933         mov     $ret,%rax               # return value
 934         ret
 935 .cfi_endproc
 936 .size   aesni_gcm_encrypt,.-aesni_gcm_encrypt
 937 ___
 938
 939 $code.=<<___;
 940 .align  64
 941 .Lbswap_mask:
 942         .byte   15,14,13,12,11,10,9,8,7,6,5,4,3,2,1,0
 943 .Lpoly:
 944         .byte   0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0xc2
 945 .Lone_msb:
 946         .byte   0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1
 947 .Ltwo_lsb:
 948         .byte   2,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0
 949 .Lone_lsb:
 950         .byte   1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0
 951 .asciz  "AES-NI GCM module for x86_64, CRYPTOGAMS by <appro\@openssl.org>"
 952 .align  64
 953 ___
 954 if ($win64) {
 955 $rec="%rcx";
 956 $frame="%rdx";
 957 $context="%r8";
 958 $disp="%r9";
 959
 960 $code.=<<___
 961 .extern __imp_RtlVirtualUnwind
 962 .type   gcm_se_handler,\@abi-omnipotent
 963 .align  16
 964 gcm_se_handler:
 965         push    %rsi
 966         push    %rdi
 967         push    %rbx
 968         push    %rbp
 969         push    %r12
 970         push    %r13
 971         push    %r14
 972         push    %r15
 973         pushfq
 974         sub     \$64,%rsp
 975
 976         mov     120($context),%rax      # pull context->Rax
 977         mov     248($context),%rbx      # pull context->Rip
 978
 979         mov     8($disp),%rsi           # disp->ImageBase
 980         mov     56($disp),%r11          # disp->HandlerData
 981
 982         mov     0(%r11),%r10d           # HandlerData[0]
 983         lea     (%rsi,%r10),%r10        # prologue label
 984         cmp     %r10,%rbx               # context->Rip<prologue label
 985         jb      .Lcommon_seh_tail
 986
 987         mov     152($context),%rax      # pull context->Rsp
 988
 989         mov     4(%r11),%r10d           # HandlerData[1]
 990         lea     (%rsi,%r10),%r10        # epilogue label
 991         cmp     %r10,%rbx               # context->Rip>=epilogue label
 992         jae     .Lcommon_seh_tail
 993
 994         mov     120($context),%rax      # pull context->Rax
 995
 996         mov     -48(%rax),%r15
 997         mov     -40(%rax),%r14
 998         mov     -32(%rax),%r13
 999         mov     -24(%rax),%r12
1000         mov     -16(%rax),%rbp
1001         mov     -8(%rax),%rbx
1002         mov     %r15,240($context)
1003         mov     %r14,232($context)
1004         mov     %r13,224($context)
1005         mov     %r12,216($context)
1006         mov     %rbp,160($context)
1007         mov     %rbx,144($context)
1008
1009         lea     -0xd8(%rax),%rsi        # %xmm save area
1010         lea     512($context),%rdi      # & context.Xmm6
1011         mov     \$20,%ecx               # 10*sizeof(%xmm0)/sizeof(%rax)
1012         .long   0xa548f3fc              # cld; rep movsq
1013
1014 .Lcommon_seh_tail:
1015         mov     8(%rax),%rdi
1016         mov     16(%rax),%rsi
1017         mov     %rax,152($context)      # restore context->Rsp
1018         mov     %rsi,168($context)      # restore context->Rsi
1019         mov     %rdi,176($context)      # restore context->Rdi
1020
1021         mov     40($disp),%rdi          # disp->ContextRecord
1022         mov     $context,%rsi           # context
1023         mov     \$154,%ecx              # sizeof(CONTEXT)
1024         .long   0xa548f3fc              # cld; rep movsq
1025
1026         mov     $disp,%rsi
1027         xor     %rcx,%rcx               # arg1, UNW_FLAG_NHANDLER
1028         mov     8(%rsi),%rdx            # arg2, disp->ImageBase
1029         mov     0(%rsi),%r8             # arg3, disp->ControlPc
1030         mov     16(%rsi),%r9            # arg4, disp->FunctionEntry
1031         mov     40(%rsi),%r10           # disp->ContextRecord
1032         lea     56(%rsi),%r11           # &disp->HandlerData
1033         lea     24(%rsi),%r12           # &disp->EstablisherFrame
1034         mov     %r10,32(%rsp)           # arg5
1035         mov     %r11,40(%rsp)           # arg6
1036         mov     %r12,48(%rsp)           # arg7
1037         mov     %rcx,56(%rsp)           # arg8, (NULL)
1038         call    *__imp_RtlVirtualUnwind(%rip)
1039
1040         mov     \$1,%eax                # ExceptionContinueSearch
1041         add     \$64,%rsp
1042         popfq
1043         pop     %r15
1044         pop     %r14
1045         pop     %r13
1046         pop     %r12
1047         pop     %rbp
1048         pop     %rbx
1049         pop     %rdi
1050         pop     %rsi
1051         ret
1052 .size   gcm_se_handler,.-gcm_se_handler
1053
1054 .section        .pdata
1055 .align  4
1056         .rva    .LSEH_begin_aesni_gcm_decrypt
1057         .rva    .LSEH_end_aesni_gcm_decrypt
1058         .rva    .LSEH_gcm_dec_info
1059
1060         .rva    .LSEH_begin_aesni_gcm_encrypt
1061         .rva    .LSEH_end_aesni_gcm_encrypt
1062         .rva    .LSEH_gcm_enc_info
1063 .section        .xdata
1064 .align  8
1065 .LSEH_gcm_dec_info:
1066         .byte   9,0,0,0
1067         .rva    gcm_se_handler
1068         .rva    .Lgcm_dec_body,.Lgcm_dec_abort
1069 .LSEH_gcm_enc_info:
1070         .byte   9,0,0,0
1071         .rva    gcm_se_handler
1072         .rva    .Lgcm_enc_body,.Lgcm_enc_abort
1073 ___
1074 }
1075 }}} else {{{
1076 $code=<<___;    # assembler is too old
1077 .text
1078
1079 .globl  aesni_gcm_encrypt
1080 .type   aesni_gcm_encrypt,\@abi-omnipotent
1081 aesni_gcm_encrypt:
1082         xor     %eax,%eax
1083         ret
1084 .size   aesni_gcm_encrypt,.-aesni_gcm_encrypt
1085
1086 .globl  aesni_gcm_decrypt
1087 .type   aesni_gcm_decrypt,\@abi-omnipotent
1088 aesni_gcm_decrypt:
1089         xor     %eax,%eax
1090         ret
1091 .size   aesni_gcm_decrypt,.-aesni_gcm_decrypt
1092 ___
1093 }}}
1094
1095 $code =~ s/\`([^\`]*)\`/eval($1)/gem;
1096
1097 print $code;
1098
1099 close STDOUT;