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 # $output is the last argument if it looks like a file (it has an extension)
  44 # $flavour is the first argument if it doesn't look like a file
  45 $output = $#ARGV >= 0 && $ARGV[$#ARGV] =~ m|\.\w+$| ? pop : undef;
  46 $flavour = $#ARGV >= 0 && $ARGV[0] !~ m|\.| ? shift : undef;
  47
  48 $win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);
  49
  50 $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
  51 ( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
  52 ( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or
  53 die "can't locate x86_64-xlate.pl";
  54
  55 if (`$ENV{CC} -Wa,-v -c -o /dev/null -x assembler /dev/null 2>&1`
  56                 =~ /GNU assembler version ([2-9]\.[0-9]+)/) {
  57         $avx = ($1>=2.20) + ($1>=2.22);
  58 }
  59
  60 if (!$avx && $win64 && ($flavour =~ /nasm/ || $ENV{ASM} =~ /nasm/) &&
  61             `nasm -v 2>&1` =~ /NASM version ([2-9]\.[0-9]+)/) {
  62         $avx = ($1>=2.09) + ($1>=2.10);
  63 }
  64
  65 if (!$avx && $win64 && ($flavour =~ /masm/ || $ENV{ASM} =~ /ml64/) &&
  66             `ml64 2>&1` =~ /Version ([0-9]+)\./) {
  67         $avx = ($1>=10) + ($1>=11);
  68 }
  69
  70 if (!$avx && `$ENV{CC} -v 2>&1` =~ /((?:^clang|LLVM) version|.*based on LLVM) ([3-9]\.[0-9]+)/) {
  71         $avx = ($2>=3.0) + ($2>3.0);
  72 }
  73
  74 open OUT,"| \"$^X\" \"$xlate\" $flavour \"$output\""
  75     or die "can't call $xlate: $!";
  76 *STDOUT=*OUT;
  77
  78 if ($avx>1) {{{
  79
  80 ($inp,$out,$len,$key,$ivp,$Xip)=("%rdi","%rsi","%rdx","%rcx","%r8","%r9");
  81
  82 ($Ii,$T1,$T2,$Hkey,
  83  $Z0,$Z1,$Z2,$Z3,$Xi) = map("%xmm$_",(0..8));
  84
  85 ($inout0,$inout1,$inout2,$inout3,$inout4,$inout5,$rndkey) = map("%xmm$_",(9..15));
  86
  87 ($counter,$rounds,$ret,$const,$in0,$end0)=("%ebx","%ebp","%r10","%r11","%r14","%r15");
  88
  89 $code=<<___;
  90 .text
  91
  92 .type   _aesni_ctr32_ghash_6x,\@abi-omnipotent
  93 .align  32
  94 _aesni_ctr32_ghash_6x:
  95         vmovdqu         0x20($const),$T2        # borrow $T2, .Lone_msb
  96         sub             \$6,$len
  97         vpxor           $Z0,$Z0,$Z0             # $Z0   = 0
  98         vmovdqu         0x00-0x80($key),$rndkey
  99         vpaddb          $T2,$T1,$inout1
 100         vpaddb          $T2,$inout1,$inout2
 101         vpaddb          $T2,$inout2,$inout3
 102         vpaddb          $T2,$inout3,$inout4
 103         vpaddb          $T2,$inout4,$inout5
 104         vpxor           $rndkey,$T1,$inout0
 105         vmovdqu         $Z0,16+8(%rsp)          # "$Z3" = 0
 106         jmp             .Loop6x
 107
 108 .align  32
 109 .Loop6x:
 110         add             \$`6<<24`,$counter
 111         jc              .Lhandle_ctr32          # discard $inout[1-5]?
 112         vmovdqu         0x00-0x20($Xip),$Hkey   # $Hkey^1
 113           vpaddb        $T2,$inout5,$T1         # next counter value
 114           vpxor         $rndkey,$inout1,$inout1
 115           vpxor         $rndkey,$inout2,$inout2
 116
 117 .Lresume_ctr32:
 118         vmovdqu         $T1,($ivp)              # save next counter value
 119         vpclmulqdq      \$0x10,$Hkey,$Z3,$Z1
 120           vpxor         $rndkey,$inout3,$inout3
 121           vmovups       0x10-0x80($key),$T2     # borrow $T2 for $rndkey
 122         vpclmulqdq      \$0x01,$Hkey,$Z3,$Z2
 123         xor             %r12,%r12
 124         cmp             $in0,$end0
 125
 126           vaesenc       $T2,$inout0,$inout0
 127         vmovdqu         0x30+8(%rsp),$Ii        # I[4]
 128           vpxor         $rndkey,$inout4,$inout4
 129         vpclmulqdq      \$0x00,$Hkey,$Z3,$T1
 130           vaesenc       $T2,$inout1,$inout1
 131           vpxor         $rndkey,$inout5,$inout5
 132         setnc           %r12b
 133         vpclmulqdq      \$0x11,$Hkey,$Z3,$Z3
 134           vaesenc       $T2,$inout2,$inout2
 135         vmovdqu         0x10-0x20($Xip),$Hkey   # $Hkey^2
 136         neg             %r12
 137           vaesenc       $T2,$inout3,$inout3
 138          vpxor          $Z1,$Z2,$Z2
 139         vpclmulqdq      \$0x00,$Hkey,$Ii,$Z1
 140          vpxor          $Z0,$Xi,$Xi             # modulo-scheduled
 141           vaesenc       $T2,$inout4,$inout4
 142          vpxor          $Z1,$T1,$Z0
 143         and             \$0x60,%r12
 144           vmovups       0x20-0x80($key),$rndkey
 145         vpclmulqdq      \$0x10,$Hkey,$Ii,$T1
 146           vaesenc       $T2,$inout5,$inout5
 147
 148         vpclmulqdq      \$0x01,$Hkey,$Ii,$T2
 149         lea             ($in0,%r12),$in0
 150           vaesenc       $rndkey,$inout0,$inout0
 151          vpxor          16+8(%rsp),$Xi,$Xi      # modulo-scheduled [vpxor $Z3,$Xi,$Xi]
 152         vpclmulqdq      \$0x11,$Hkey,$Ii,$Hkey
 153          vmovdqu        0x40+8(%rsp),$Ii        # I[3]
 154           vaesenc       $rndkey,$inout1,$inout1
 155         movbe           0x58($in0),%r13
 156           vaesenc       $rndkey,$inout2,$inout2
 157         movbe           0x50($in0),%r12
 158           vaesenc       $rndkey,$inout3,$inout3
 159         mov             %r13,0x20+8(%rsp)
 160           vaesenc       $rndkey,$inout4,$inout4
 161         mov             %r12,0x28+8(%rsp)
 162         vmovdqu         0x30-0x20($Xip),$Z1     # borrow $Z1 for $Hkey^3
 163           vaesenc       $rndkey,$inout5,$inout5
 164
 165           vmovups       0x30-0x80($key),$rndkey
 166          vpxor          $T1,$Z2,$Z2
 167         vpclmulqdq      \$0x00,$Z1,$Ii,$T1
 168           vaesenc       $rndkey,$inout0,$inout0
 169          vpxor          $T2,$Z2,$Z2
 170         vpclmulqdq      \$0x10,$Z1,$Ii,$T2
 171           vaesenc       $rndkey,$inout1,$inout1
 172          vpxor          $Hkey,$Z3,$Z3
 173         vpclmulqdq      \$0x01,$Z1,$Ii,$Hkey
 174           vaesenc       $rndkey,$inout2,$inout2
 175         vpclmulqdq      \$0x11,$Z1,$Ii,$Z1
 176          vmovdqu        0x50+8(%rsp),$Ii        # I[2]
 177           vaesenc       $rndkey,$inout3,$inout3
 178           vaesenc       $rndkey,$inout4,$inout4
 179          vpxor          $T1,$Z0,$Z0
 180         vmovdqu         0x40-0x20($Xip),$T1     # borrow $T1 for $Hkey^4
 181           vaesenc       $rndkey,$inout5,$inout5
 182
 183           vmovups       0x40-0x80($key),$rndkey
 184          vpxor          $T2,$Z2,$Z2
 185         vpclmulqdq      \$0x00,$T1,$Ii,$T2
 186           vaesenc       $rndkey,$inout0,$inout0
 187          vpxor          $Hkey,$Z2,$Z2
 188         vpclmulqdq      \$0x10,$T1,$Ii,$Hkey
 189           vaesenc       $rndkey,$inout1,$inout1
 190         movbe           0x48($in0),%r13
 191          vpxor          $Z1,$Z3,$Z3
 192         vpclmulqdq      \$0x01,$T1,$Ii,$Z1
 193           vaesenc       $rndkey,$inout2,$inout2
 194         movbe           0x40($in0),%r12
 195         vpclmulqdq      \$0x11,$T1,$Ii,$T1
 196          vmovdqu        0x60+8(%rsp),$Ii        # I[1]
 197           vaesenc       $rndkey,$inout3,$inout3
 198         mov             %r13,0x30+8(%rsp)
 199           vaesenc       $rndkey,$inout4,$inout4
 200         mov             %r12,0x38+8(%rsp)
 201          vpxor          $T2,$Z0,$Z0
 202         vmovdqu         0x60-0x20($Xip),$T2     # borrow $T2 for $Hkey^5
 203           vaesenc       $rndkey,$inout5,$inout5
 204
 205           vmovups       0x50-0x80($key),$rndkey
 206          vpxor          $Hkey,$Z2,$Z2
 207         vpclmulqdq      \$0x00,$T2,$Ii,$Hkey
 208           vaesenc       $rndkey,$inout0,$inout0
 209          vpxor          $Z1,$Z2,$Z2
 210         vpclmulqdq      \$0x10,$T2,$Ii,$Z1
 211           vaesenc       $rndkey,$inout1,$inout1
 212         movbe           0x38($in0),%r13
 213          vpxor          $T1,$Z3,$Z3
 214         vpclmulqdq      \$0x01,$T2,$Ii,$T1
 215          vpxor          0x70+8(%rsp),$Xi,$Xi    # accumulate I[0]
 216           vaesenc       $rndkey,$inout2,$inout2
 217         movbe           0x30($in0),%r12
 218         vpclmulqdq      \$0x11,$T2,$Ii,$T2
 219           vaesenc       $rndkey,$inout3,$inout3
 220         mov             %r13,0x40+8(%rsp)
 221           vaesenc       $rndkey,$inout4,$inout4
 222         mov             %r12,0x48+8(%rsp)
 223          vpxor          $Hkey,$Z0,$Z0
 224          vmovdqu        0x70-0x20($Xip),$Hkey   # $Hkey^6
 225           vaesenc       $rndkey,$inout5,$inout5
 226
 227           vmovups       0x60-0x80($key),$rndkey
 228          vpxor          $Z1,$Z2,$Z2
 229         vpclmulqdq      \$0x10,$Hkey,$Xi,$Z1
 230           vaesenc       $rndkey,$inout0,$inout0
 231          vpxor          $T1,$Z2,$Z2
 232         vpclmulqdq      \$0x01,$Hkey,$Xi,$T1
 233           vaesenc       $rndkey,$inout1,$inout1
 234         movbe           0x28($in0),%r13
 235          vpxor          $T2,$Z3,$Z3
 236         vpclmulqdq      \$0x00,$Hkey,$Xi,$T2
 237           vaesenc       $rndkey,$inout2,$inout2
 238         movbe           0x20($in0),%r12
 239         vpclmulqdq      \$0x11,$Hkey,$Xi,$Xi
 240           vaesenc       $rndkey,$inout3,$inout3
 241         mov             %r13,0x50+8(%rsp)
 242           vaesenc       $rndkey,$inout4,$inout4
 243         mov             %r12,0x58+8(%rsp)
 244         vpxor           $Z1,$Z2,$Z2
 245           vaesenc       $rndkey,$inout5,$inout5
 246         vpxor           $T1,$Z2,$Z2
 247
 248           vmovups       0x70-0x80($key),$rndkey
 249         vpslldq         \$8,$Z2,$Z1
 250         vpxor           $T2,$Z0,$Z0
 251         vmovdqu         0x10($const),$Hkey      # .Lpoly
 252
 253           vaesenc       $rndkey,$inout0,$inout0
 254         vpxor           $Xi,$Z3,$Z3
 255           vaesenc       $rndkey,$inout1,$inout1
 256         vpxor           $Z1,$Z0,$Z0
 257         movbe           0x18($in0),%r13
 258           vaesenc       $rndkey,$inout2,$inout2
 259         movbe           0x10($in0),%r12
 260         vpalignr        \$8,$Z0,$Z0,$Ii         # 1st phase
 261         vpclmulqdq      \$0x10,$Hkey,$Z0,$Z0
 262         mov             %r13,0x60+8(%rsp)
 263           vaesenc       $rndkey,$inout3,$inout3
 264         mov             %r12,0x68+8(%rsp)
 265           vaesenc       $rndkey,$inout4,$inout4
 266           vmovups       0x80-0x80($key),$T1     # borrow $T1 for $rndkey
 267           vaesenc       $rndkey,$inout5,$inout5
 268
 269           vaesenc       $T1,$inout0,$inout0
 270           vmovups       0x90-0x80($key),$rndkey
 271           vaesenc       $T1,$inout1,$inout1
 272         vpsrldq         \$8,$Z2,$Z2
 273           vaesenc       $T1,$inout2,$inout2
 274         vpxor           $Z2,$Z3,$Z3
 275           vaesenc       $T1,$inout3,$inout3
 276         vpxor           $Ii,$Z0,$Z0
 277         movbe           0x08($in0),%r13
 278           vaesenc       $T1,$inout4,$inout4
 279         movbe           0x00($in0),%r12
 280           vaesenc       $T1,$inout5,$inout5
 281           vmovups       0xa0-0x80($key),$T1
 282           cmp           \$11,$rounds
 283           jb            .Lenc_tail              # 128-bit key
 284
 285           vaesenc       $rndkey,$inout0,$inout0
 286           vaesenc       $rndkey,$inout1,$inout1
 287           vaesenc       $rndkey,$inout2,$inout2
 288           vaesenc       $rndkey,$inout3,$inout3
 289           vaesenc       $rndkey,$inout4,$inout4
 290           vaesenc       $rndkey,$inout5,$inout5
 291
 292           vaesenc       $T1,$inout0,$inout0
 293           vaesenc       $T1,$inout1,$inout1
 294           vaesenc       $T1,$inout2,$inout2
 295           vaesenc       $T1,$inout3,$inout3
 296           vaesenc       $T1,$inout4,$inout4
 297           vmovups       0xb0-0x80($key),$rndkey
 298           vaesenc       $T1,$inout5,$inout5
 299           vmovups       0xc0-0x80($key),$T1
 300           je            .Lenc_tail              # 192-bit key
 301
 302           vaesenc       $rndkey,$inout0,$inout0
 303           vaesenc       $rndkey,$inout1,$inout1
 304           vaesenc       $rndkey,$inout2,$inout2
 305           vaesenc       $rndkey,$inout3,$inout3
 306           vaesenc       $rndkey,$inout4,$inout4
 307           vaesenc       $rndkey,$inout5,$inout5
 308
 309           vaesenc       $T1,$inout0,$inout0
 310           vaesenc       $T1,$inout1,$inout1
 311           vaesenc       $T1,$inout2,$inout2
 312           vaesenc       $T1,$inout3,$inout3
 313           vaesenc       $T1,$inout4,$inout4
 314           vmovups       0xd0-0x80($key),$rndkey
 315           vaesenc       $T1,$inout5,$inout5
 316           vmovups       0xe0-0x80($key),$T1
 317           jmp           .Lenc_tail              # 256-bit key
 318
 319 .align  32
 320 .Lhandle_ctr32:
 321         vmovdqu         ($const),$Ii            # borrow $Ii for .Lbswap_mask
 322           vpshufb       $Ii,$T1,$Z2             # byte-swap counter
 323           vmovdqu       0x30($const),$Z1        # borrow $Z1, .Ltwo_lsb
 324           vpaddd        0x40($const),$Z2,$inout1        # .Lone_lsb
 325           vpaddd        $Z1,$Z2,$inout2
 326         vmovdqu         0x00-0x20($Xip),$Hkey   # $Hkey^1
 327           vpaddd        $Z1,$inout1,$inout3
 328           vpshufb       $Ii,$inout1,$inout1
 329           vpaddd        $Z1,$inout2,$inout4
 330           vpshufb       $Ii,$inout2,$inout2
 331           vpxor         $rndkey,$inout1,$inout1
 332           vpaddd        $Z1,$inout3,$inout5
 333           vpshufb       $Ii,$inout3,$inout3
 334           vpxor         $rndkey,$inout2,$inout2
 335           vpaddd        $Z1,$inout4,$T1         # byte-swapped next counter value
 336           vpshufb       $Ii,$inout4,$inout4
 337           vpshufb       $Ii,$inout5,$inout5
 338           vpshufb       $Ii,$T1,$T1             # next counter value
 339         jmp             .Lresume_ctr32
 340
 341 .align  32
 342 .Lenc_tail:
 343           vaesenc       $rndkey,$inout0,$inout0
 344         vmovdqu         $Z3,16+8(%rsp)          # postpone vpxor $Z3,$Xi,$Xi
 345         vpalignr        \$8,$Z0,$Z0,$Xi         # 2nd phase
 346           vaesenc       $rndkey,$inout1,$inout1
 347         vpclmulqdq      \$0x10,$Hkey,$Z0,$Z0
 348           vpxor         0x00($inp),$T1,$T2
 349           vaesenc       $rndkey,$inout2,$inout2
 350           vpxor         0x10($inp),$T1,$Ii
 351           vaesenc       $rndkey,$inout3,$inout3
 352           vpxor         0x20($inp),$T1,$Z1
 353           vaesenc       $rndkey,$inout4,$inout4
 354           vpxor         0x30($inp),$T1,$Z2
 355           vaesenc       $rndkey,$inout5,$inout5
 356           vpxor         0x40($inp),$T1,$Z3
 357           vpxor         0x50($inp),$T1,$Hkey
 358           vmovdqu       ($ivp),$T1              # load next counter value
 359
 360           vaesenclast   $T2,$inout0,$inout0
 361           vmovdqu       0x20($const),$T2        # borrow $T2, .Lone_msb
 362           vaesenclast   $Ii,$inout1,$inout1
 363          vpaddb         $T2,$T1,$Ii
 364         mov             %r13,0x70+8(%rsp)
 365         lea             0x60($inp),$inp
 366           vaesenclast   $Z1,$inout2,$inout2
 367          vpaddb         $T2,$Ii,$Z1
 368         mov             %r12,0x78+8(%rsp)
 369         lea             0x60($out),$out
 370           vmovdqu       0x00-0x80($key),$rndkey
 371           vaesenclast   $Z2,$inout3,$inout3
 372          vpaddb         $T2,$Z1,$Z2
 373           vaesenclast   $Z3, $inout4,$inout4
 374          vpaddb         $T2,$Z2,$Z3
 375           vaesenclast   $Hkey,$inout5,$inout5
 376          vpaddb         $T2,$Z3,$Hkey
 377
 378         add             \$0x60,$ret
 379         sub             \$0x6,$len
 380         jc              .L6x_done
 381
 382           vmovups       $inout0,-0x60($out)     # save output
 383          vpxor          $rndkey,$T1,$inout0
 384           vmovups       $inout1,-0x50($out)
 385          vmovdqa        $Ii,$inout1             # 0 latency
 386           vmovups       $inout2,-0x40($out)
 387          vmovdqa        $Z1,$inout2             # 0 latency
 388           vmovups       $inout3,-0x30($out)
 389          vmovdqa        $Z2,$inout3             # 0 latency
 390           vmovups       $inout4,-0x20($out)
 391          vmovdqa        $Z3,$inout4             # 0 latency
 392           vmovups       $inout5,-0x10($out)
 393          vmovdqa        $Hkey,$inout5           # 0 latency
 394         vmovdqu         0x20+8(%rsp),$Z3        # I[5]
 395         jmp             .Loop6x
 396
 397 .L6x_done:
 398         vpxor           16+8(%rsp),$Xi,$Xi      # modulo-scheduled
 399         vpxor           $Z0,$Xi,$Xi             # modulo-scheduled
 400
 401         ret
 402 .size   _aesni_ctr32_ghash_6x,.-_aesni_ctr32_ghash_6x
 403 ___
 404 ######################################################################
 405 #
 406 # size_t aesni_gcm_[en|de]crypt(const void *inp, void *out, size_t len,
 407 #               const AES_KEY *key, unsigned char iv[16],
 408 #               struct { u128 Xi,H,Htbl[9]; } *Xip);
 409 $code.=<<___;
 410 .globl  aesni_gcm_decrypt
 411 .type   aesni_gcm_decrypt,\@function,6
 412 .align  32
 413 aesni_gcm_decrypt:
 414 .cfi_startproc
 415         xor     $ret,$ret
 416         cmp     \$0x60,$len                     # minimal accepted length
 417         jb      .Lgcm_dec_abort
 418
 419         lea     (%rsp),%rax                     # save stack pointer
 420 .cfi_def_cfa_register   %rax
 421         push    %rbx
 422 .cfi_push       %rbx
 423         push    %rbp
 424 .cfi_push       %rbp
 425         push    %r12
 426 .cfi_push       %r12
 427         push    %r13
 428 .cfi_push       %r13
 429         push    %r14
 430 .cfi_push       %r14
 431         push    %r15
 432 .cfi_push       %r15
 433 ___
 434 $code.=<<___ if ($win64);
 435         lea     -0xa8(%rsp),%rsp
 436         movaps  %xmm6,-0xd8(%rax)
 437         movaps  %xmm7,-0xc8(%rax)
 438         movaps  %xmm8,-0xb8(%rax)
 439         movaps  %xmm9,-0xa8(%rax)
 440         movaps  %xmm10,-0x98(%rax)
 441         movaps  %xmm11,-0x88(%rax)
 442         movaps  %xmm12,-0x78(%rax)
 443         movaps  %xmm13,-0x68(%rax)
 444         movaps  %xmm14,-0x58(%rax)
 445         movaps  %xmm15,-0x48(%rax)
 446 .Lgcm_dec_body:
 447 ___
 448 $code.=<<___;
 449         vzeroupper
 450
 451         vmovdqu         ($ivp),$T1              # input counter value
 452         add             \$-128,%rsp
 453         mov             12($ivp),$counter
 454         lea             .Lbswap_mask(%rip),$const
 455         lea             -0x80($key),$in0        # borrow $in0
 456         mov             \$0xf80,$end0           # borrow $end0
 457         vmovdqu         ($Xip),$Xi              # load Xi
 458         and             \$-128,%rsp             # ensure stack alignment
 459         vmovdqu         ($const),$Ii            # borrow $Ii for .Lbswap_mask
 460         lea             0x80($key),$key         # size optimization
 461         lea             0x20+0x20($Xip),$Xip    # size optimization
 462         mov             0xf0-0x80($key),$rounds
 463         vpshufb         $Ii,$Xi,$Xi
 464
 465         and             $end0,$in0
 466         and             %rsp,$end0
 467         sub             $in0,$end0
 468         jc              .Ldec_no_key_aliasing
 469         cmp             \$768,$end0
 470         jnc             .Ldec_no_key_aliasing
 471         sub             $end0,%rsp              # avoid aliasing with key
 472 .Ldec_no_key_aliasing:
 473
 474         vmovdqu         0x50($inp),$Z3          # I[5]
 475         lea             ($inp),$in0
 476         vmovdqu         0x40($inp),$Z0
 477         lea             -0xc0($inp,$len),$end0
 478         vmovdqu         0x30($inp),$Z1
 479         shr             \$4,$len
 480         xor             $ret,$ret
 481         vmovdqu         0x20($inp),$Z2
 482          vpshufb        $Ii,$Z3,$Z3             # passed to _aesni_ctr32_ghash_6x
 483         vmovdqu         0x10($inp),$T2
 484          vpshufb        $Ii,$Z0,$Z0
 485         vmovdqu         ($inp),$Hkey
 486          vpshufb        $Ii,$Z1,$Z1
 487         vmovdqu         $Z0,0x30(%rsp)
 488          vpshufb        $Ii,$Z2,$Z2
 489         vmovdqu         $Z1,0x40(%rsp)
 490          vpshufb        $Ii,$T2,$T2
 491         vmovdqu         $Z2,0x50(%rsp)
 492          vpshufb        $Ii,$Hkey,$Hkey
 493         vmovdqu         $T2,0x60(%rsp)
 494         vmovdqu         $Hkey,0x70(%rsp)
 495
 496         call            _aesni_ctr32_ghash_6x
 497
 498         vmovups         $inout0,-0x60($out)     # save output
 499         vmovups         $inout1,-0x50($out)
 500         vmovups         $inout2,-0x40($out)
 501         vmovups         $inout3,-0x30($out)
 502         vmovups         $inout4,-0x20($out)
 503         vmovups         $inout5,-0x10($out)
 504
 505         vpshufb         ($const),$Xi,$Xi        # .Lbswap_mask
 506         vmovdqu         $Xi,-0x40($Xip)         # output Xi
 507
 508         vzeroupper
 509 ___
 510 $code.=<<___ if ($win64);
 511         movaps  -0xd8(%rax),%xmm6
 512         movaps  -0xc8(%rax),%xmm7
 513         movaps  -0xb8(%rax),%xmm8
 514         movaps  -0xa8(%rax),%xmm9
 515         movaps  -0x98(%rax),%xmm10
 516         movaps  -0x88(%rax),%xmm11
 517         movaps  -0x78(%rax),%xmm12
 518         movaps  -0x68(%rax),%xmm13
 519         movaps  -0x58(%rax),%xmm14
 520         movaps  -0x48(%rax),%xmm15
 521 ___
 522 $code.=<<___;
 523         mov     -48(%rax),%r15
 524 .cfi_restore    %r15
 525         mov     -40(%rax),%r14
 526 .cfi_restore    %r14
 527         mov     -32(%rax),%r13
 528 .cfi_restore    %r13
 529         mov     -24(%rax),%r12
 530 .cfi_restore    %r12
 531         mov     -16(%rax),%rbp
 532 .cfi_restore    %rbp
 533         mov     -8(%rax),%rbx
 534 .cfi_restore    %rbx
 535         lea     (%rax),%rsp             # restore %rsp
 536 .cfi_def_cfa_register   %rsp
 537 .Lgcm_dec_abort:
 538         mov     $ret,%rax               # return value
 539         ret
 540 .cfi_endproc
 541 .size   aesni_gcm_decrypt,.-aesni_gcm_decrypt
 542 ___
 543
 544 $code.=<<___;
 545 .type   _aesni_ctr32_6x,\@abi-omnipotent
 546 .align  32
 547 _aesni_ctr32_6x:
 548         vmovdqu         0x00-0x80($key),$Z0     # borrow $Z0 for $rndkey
 549         vmovdqu         0x20($const),$T2        # borrow $T2, .Lone_msb
 550         lea             -1($rounds),%r13
 551         vmovups         0x10-0x80($key),$rndkey
 552         lea             0x20-0x80($key),%r12
 553         vpxor           $Z0,$T1,$inout0
 554         add             \$`6<<24`,$counter
 555         jc              .Lhandle_ctr32_2
 556         vpaddb          $T2,$T1,$inout1
 557         vpaddb          $T2,$inout1,$inout2
 558         vpxor           $Z0,$inout1,$inout1
 559         vpaddb          $T2,$inout2,$inout3
 560         vpxor           $Z0,$inout2,$inout2
 561         vpaddb          $T2,$inout3,$inout4
 562         vpxor           $Z0,$inout3,$inout3
 563         vpaddb          $T2,$inout4,$inout5
 564         vpxor           $Z0,$inout4,$inout4
 565         vpaddb          $T2,$inout5,$T1
 566         vpxor           $Z0,$inout5,$inout5
 567         jmp             .Loop_ctr32
 568
 569 .align  16
 570 .Loop_ctr32:
 571         vaesenc         $rndkey,$inout0,$inout0
 572         vaesenc         $rndkey,$inout1,$inout1
 573         vaesenc         $rndkey,$inout2,$inout2
 574         vaesenc         $rndkey,$inout3,$inout3
 575         vaesenc         $rndkey,$inout4,$inout4
 576         vaesenc         $rndkey,$inout5,$inout5
 577         vmovups         (%r12),$rndkey
 578         lea             0x10(%r12),%r12
 579         dec             %r13d
 580         jnz             .Loop_ctr32
 581
 582         vmovdqu         (%r12),$Hkey            # last round key
 583         vaesenc         $rndkey,$inout0,$inout0
 584         vpxor           0x00($inp),$Hkey,$Z0
 585         vaesenc         $rndkey,$inout1,$inout1
 586         vpxor           0x10($inp),$Hkey,$Z1
 587         vaesenc         $rndkey,$inout2,$inout2
 588         vpxor           0x20($inp),$Hkey,$Z2
 589         vaesenc         $rndkey,$inout3,$inout3
 590         vpxor           0x30($inp),$Hkey,$Xi
 591         vaesenc         $rndkey,$inout4,$inout4
 592         vpxor           0x40($inp),$Hkey,$T2
 593         vaesenc         $rndkey,$inout5,$inout5
 594         vpxor           0x50($inp),$Hkey,$Hkey
 595         lea             0x60($inp),$inp
 596
 597         vaesenclast     $Z0,$inout0,$inout0
 598         vaesenclast     $Z1,$inout1,$inout1
 599         vaesenclast     $Z2,$inout2,$inout2
 600         vaesenclast     $Xi,$inout3,$inout3
 601         vaesenclast     $T2,$inout4,$inout4
 602         vaesenclast     $Hkey,$inout5,$inout5
 603         vmovups         $inout0,0x00($out)
 604         vmovups         $inout1,0x10($out)
 605         vmovups         $inout2,0x20($out)
 606         vmovups         $inout3,0x30($out)
 607         vmovups         $inout4,0x40($out)
 608         vmovups         $inout5,0x50($out)
 609         lea             0x60($out),$out
 610
 611         ret
 612 .align  32
 613 .Lhandle_ctr32_2:
 614         vpshufb         $Ii,$T1,$Z2             # byte-swap counter
 615         vmovdqu         0x30($const),$Z1        # borrow $Z1, .Ltwo_lsb
 616         vpaddd          0x40($const),$Z2,$inout1        # .Lone_lsb
 617         vpaddd          $Z1,$Z2,$inout2
 618         vpaddd          $Z1,$inout1,$inout3
 619         vpshufb         $Ii,$inout1,$inout1
 620         vpaddd          $Z1,$inout2,$inout4
 621         vpshufb         $Ii,$inout2,$inout2
 622         vpxor           $Z0,$inout1,$inout1
 623         vpaddd          $Z1,$inout3,$inout5
 624         vpshufb         $Ii,$inout3,$inout3
 625         vpxor           $Z0,$inout2,$inout2
 626         vpaddd          $Z1,$inout4,$T1         # byte-swapped next counter value
 627         vpshufb         $Ii,$inout4,$inout4
 628         vpxor           $Z0,$inout3,$inout3
 629         vpshufb         $Ii,$inout5,$inout5
 630         vpxor           $Z0,$inout4,$inout4
 631         vpshufb         $Ii,$T1,$T1             # next counter value
 632         vpxor           $Z0,$inout5,$inout5
 633         jmp     .Loop_ctr32
 634 .size   _aesni_ctr32_6x,.-_aesni_ctr32_6x
 635
 636 .globl  aesni_gcm_encrypt
 637 .type   aesni_gcm_encrypt,\@function,6
 638 .align  32
 639 aesni_gcm_encrypt:
 640 .cfi_startproc
 641         xor     $ret,$ret
 642         cmp     \$0x60*3,$len                   # minimal accepted length
 643         jb      .Lgcm_enc_abort
 644
 645         lea     (%rsp),%rax                     # save stack pointer
 646 .cfi_def_cfa_register   %rax
 647         push    %rbx
 648 .cfi_push       %rbx
 649         push    %rbp
 650 .cfi_push       %rbp
 651         push    %r12
 652 .cfi_push       %r12
 653         push    %r13
 654 .cfi_push       %r13
 655         push    %r14
 656 .cfi_push       %r14
 657         push    %r15
 658 .cfi_push       %r15
 659 ___
 660 $code.=<<___ if ($win64);
 661         lea     -0xa8(%rsp),%rsp
 662         movaps  %xmm6,-0xd8(%rax)
 663         movaps  %xmm7,-0xc8(%rax)
 664         movaps  %xmm8,-0xb8(%rax)
 665         movaps  %xmm9,-0xa8(%rax)
 666         movaps  %xmm10,-0x98(%rax)
 667         movaps  %xmm11,-0x88(%rax)
 668         movaps  %xmm12,-0x78(%rax)
 669         movaps  %xmm13,-0x68(%rax)
 670         movaps  %xmm14,-0x58(%rax)
 671         movaps  %xmm15,-0x48(%rax)
 672 .Lgcm_enc_body:
 673 ___
 674 $code.=<<___;
 675         vzeroupper
 676
 677         vmovdqu         ($ivp),$T1              # input counter value
 678         add             \$-128,%rsp
 679         mov             12($ivp),$counter
 680         lea             .Lbswap_mask(%rip),$const
 681         lea             -0x80($key),$in0        # borrow $in0
 682         mov             \$0xf80,$end0           # borrow $end0
 683         lea             0x80($key),$key         # size optimization
 684         vmovdqu         ($const),$Ii            # borrow $Ii for .Lbswap_mask
 685         and             \$-128,%rsp             # ensure stack alignment
 686         mov             0xf0-0x80($key),$rounds
 687
 688         and             $end0,$in0
 689         and             %rsp,$end0
 690         sub             $in0,$end0
 691         jc              .Lenc_no_key_aliasing
 692         cmp             \$768,$end0
 693         jnc             .Lenc_no_key_aliasing
 694         sub             $end0,%rsp              # avoid aliasing with key
 695 .Lenc_no_key_aliasing:
 696
 697         lea             ($out),$in0
 698         lea             -0xc0($out,$len),$end0
 699         shr             \$4,$len
 700
 701         call            _aesni_ctr32_6x
 702         vpshufb         $Ii,$inout0,$Xi         # save bswapped output on stack
 703         vpshufb         $Ii,$inout1,$T2
 704         vmovdqu         $Xi,0x70(%rsp)
 705         vpshufb         $Ii,$inout2,$Z0
 706         vmovdqu         $T2,0x60(%rsp)
 707         vpshufb         $Ii,$inout3,$Z1
 708         vmovdqu         $Z0,0x50(%rsp)
 709         vpshufb         $Ii,$inout4,$Z2
 710         vmovdqu         $Z1,0x40(%rsp)
 711         vpshufb         $Ii,$inout5,$Z3         # passed to _aesni_ctr32_ghash_6x
 712         vmovdqu         $Z2,0x30(%rsp)
 713
 714         call            _aesni_ctr32_6x
 715
 716         vmovdqu         ($Xip),$Xi              # load Xi
 717         lea             0x20+0x20($Xip),$Xip    # size optimization
 718         sub             \$12,$len
 719         mov             \$0x60*2,$ret
 720         vpshufb         $Ii,$Xi,$Xi
 721
 722         call            _aesni_ctr32_ghash_6x
 723         vmovdqu         0x20(%rsp),$Z3          # I[5]
 724          vmovdqu        ($const),$Ii            # borrow $Ii for .Lbswap_mask
 725         vmovdqu         0x00-0x20($Xip),$Hkey   # $Hkey^1
 726         vpunpckhqdq     $Z3,$Z3,$T1
 727         vmovdqu         0x20-0x20($Xip),$rndkey # borrow $rndkey for $HK
 728          vmovups        $inout0,-0x60($out)     # save output
 729          vpshufb        $Ii,$inout0,$inout0     # but keep bswapped copy
 730         vpxor           $Z3,$T1,$T1
 731          vmovups        $inout1,-0x50($out)
 732          vpshufb        $Ii,$inout1,$inout1
 733          vmovups        $inout2,-0x40($out)
 734          vpshufb        $Ii,$inout2,$inout2
 735          vmovups        $inout3,-0x30($out)
 736          vpshufb        $Ii,$inout3,$inout3
 737          vmovups        $inout4,-0x20($out)
 738          vpshufb        $Ii,$inout4,$inout4
 739          vmovups        $inout5,-0x10($out)
 740          vpshufb        $Ii,$inout5,$inout5
 741          vmovdqu        $inout0,0x10(%rsp)      # free $inout0
 742 ___
 743 { my ($HK,$T3)=($rndkey,$inout0);
 744
 745 $code.=<<___;
 746          vmovdqu        0x30(%rsp),$Z2          # I[4]
 747          vmovdqu        0x10-0x20($Xip),$Ii     # borrow $Ii for $Hkey^2
 748          vpunpckhqdq    $Z2,$Z2,$T2
 749         vpclmulqdq      \$0x00,$Hkey,$Z3,$Z1
 750          vpxor          $Z2,$T2,$T2
 751         vpclmulqdq      \$0x11,$Hkey,$Z3,$Z3
 752         vpclmulqdq      \$0x00,$HK,$T1,$T1
 753
 754          vmovdqu        0x40(%rsp),$T3          # I[3]
 755         vpclmulqdq      \$0x00,$Ii,$Z2,$Z0
 756          vmovdqu        0x30-0x20($Xip),$Hkey   # $Hkey^3
 757         vpxor           $Z1,$Z0,$Z0
 758          vpunpckhqdq    $T3,$T3,$Z1
 759         vpclmulqdq      \$0x11,$Ii,$Z2,$Z2
 760          vpxor          $T3,$Z1,$Z1
 761         vpxor           $Z3,$Z2,$Z2
 762         vpclmulqdq      \$0x10,$HK,$T2,$T2
 763          vmovdqu        0x50-0x20($Xip),$HK
 764         vpxor           $T1,$T2,$T2
 765
 766          vmovdqu        0x50(%rsp),$T1          # I[2]
 767         vpclmulqdq      \$0x00,$Hkey,$T3,$Z3
 768          vmovdqu        0x40-0x20($Xip),$Ii     # borrow $Ii for $Hkey^4
 769         vpxor           $Z0,$Z3,$Z3
 770          vpunpckhqdq    $T1,$T1,$Z0
 771         vpclmulqdq      \$0x11,$Hkey,$T3,$T3
 772          vpxor          $T1,$Z0,$Z0
 773         vpxor           $Z2,$T3,$T3
 774         vpclmulqdq      \$0x00,$HK,$Z1,$Z1
 775         vpxor           $T2,$Z1,$Z1
 776
 777          vmovdqu        0x60(%rsp),$T2          # I[1]
 778         vpclmulqdq      \$0x00,$Ii,$T1,$Z2
 779          vmovdqu        0x60-0x20($Xip),$Hkey   # $Hkey^5
 780         vpxor           $Z3,$Z2,$Z2
 781          vpunpckhqdq    $T2,$T2,$Z3
 782         vpclmulqdq      \$0x11,$Ii,$T1,$T1
 783          vpxor          $T2,$Z3,$Z3
 784         vpxor           $T3,$T1,$T1
 785         vpclmulqdq      \$0x10,$HK,$Z0,$Z0
 786          vmovdqu        0x80-0x20($Xip),$HK
 787         vpxor           $Z1,$Z0,$Z0
 788
 789          vpxor          0x70(%rsp),$Xi,$Xi      # accumulate I[0]
 790         vpclmulqdq      \$0x00,$Hkey,$T2,$Z1
 791          vmovdqu        0x70-0x20($Xip),$Ii     # borrow $Ii for $Hkey^6
 792          vpunpckhqdq    $Xi,$Xi,$T3
 793         vpxor           $Z2,$Z1,$Z1
 794         vpclmulqdq      \$0x11,$Hkey,$T2,$T2
 795          vpxor          $Xi,$T3,$T3
 796         vpxor           $T1,$T2,$T2
 797         vpclmulqdq      \$0x00,$HK,$Z3,$Z3
 798         vpxor           $Z0,$Z3,$Z0
 799
 800         vpclmulqdq      \$0x00,$Ii,$Xi,$Z2
 801          vmovdqu        0x00-0x20($Xip),$Hkey   # $Hkey^1
 802          vpunpckhqdq    $inout5,$inout5,$T1
 803         vpclmulqdq      \$0x11,$Ii,$Xi,$Xi
 804          vpxor          $inout5,$T1,$T1
 805         vpxor           $Z1,$Z2,$Z1
 806         vpclmulqdq      \$0x10,$HK,$T3,$T3
 807          vmovdqu        0x20-0x20($Xip),$HK
 808         vpxor           $T2,$Xi,$Z3
 809         vpxor           $Z0,$T3,$Z2
 810
 811          vmovdqu        0x10-0x20($Xip),$Ii     # borrow $Ii for $Hkey^2
 812           vpxor         $Z1,$Z3,$T3             # aggregated Karatsuba post-processing
 813         vpclmulqdq      \$0x00,$Hkey,$inout5,$Z0
 814           vpxor         $T3,$Z2,$Z2
 815          vpunpckhqdq    $inout4,$inout4,$T2
 816         vpclmulqdq      \$0x11,$Hkey,$inout5,$inout5
 817          vpxor          $inout4,$T2,$T2
 818           vpslldq       \$8,$Z2,$T3
 819         vpclmulqdq      \$0x00,$HK,$T1,$T1
 820           vpxor         $T3,$Z1,$Xi
 821           vpsrldq       \$8,$Z2,$Z2
 822           vpxor         $Z2,$Z3,$Z3
 823
 824         vpclmulqdq      \$0x00,$Ii,$inout4,$Z1
 825          vmovdqu        0x30-0x20($Xip),$Hkey   # $Hkey^3
 826         vpxor           $Z0,$Z1,$Z1
 827          vpunpckhqdq    $inout3,$inout3,$T3
 828         vpclmulqdq      \$0x11,$Ii,$inout4,$inout4
 829          vpxor          $inout3,$T3,$T3
 830         vpxor           $inout5,$inout4,$inout4
 831           vpalignr      \$8,$Xi,$Xi,$inout5     # 1st phase
 832         vpclmulqdq      \$0x10,$HK,$T2,$T2
 833          vmovdqu        0x50-0x20($Xip),$HK
 834         vpxor           $T1,$T2,$T2
 835
 836         vpclmulqdq      \$0x00,$Hkey,$inout3,$Z0
 837          vmovdqu        0x40-0x20($Xip),$Ii     # borrow $Ii for $Hkey^4
 838         vpxor           $Z1,$Z0,$Z0
 839          vpunpckhqdq    $inout2,$inout2,$T1
 840         vpclmulqdq      \$0x11,$Hkey,$inout3,$inout3
 841          vpxor          $inout2,$T1,$T1
 842         vpxor           $inout4,$inout3,$inout3
 843           vxorps        0x10(%rsp),$Z3,$Z3      # accumulate $inout0
 844         vpclmulqdq      \$0x00,$HK,$T3,$T3
 845         vpxor           $T2,$T3,$T3
 846
 847           vpclmulqdq    \$0x10,0x10($const),$Xi,$Xi
 848           vxorps        $inout5,$Xi,$Xi
 849
 850         vpclmulqdq      \$0x00,$Ii,$inout2,$Z1
 851          vmovdqu        0x60-0x20($Xip),$Hkey   # $Hkey^5
 852         vpxor           $Z0,$Z1,$Z1
 853          vpunpckhqdq    $inout1,$inout1,$T2
 854         vpclmulqdq      \$0x11,$Ii,$inout2,$inout2
 855          vpxor          $inout1,$T2,$T2
 856           vpalignr      \$8,$Xi,$Xi,$inout5     # 2nd phase
 857         vpxor           $inout3,$inout2,$inout2
 858         vpclmulqdq      \$0x10,$HK,$T1,$T1
 859          vmovdqu        0x80-0x20($Xip),$HK
 860         vpxor           $T3,$T1,$T1
 861
 862           vxorps        $Z3,$inout5,$inout5
 863           vpclmulqdq    \$0x10,0x10($const),$Xi,$Xi
 864           vxorps        $inout5,$Xi,$Xi
 865
 866         vpclmulqdq      \$0x00,$Hkey,$inout1,$Z0
 867          vmovdqu        0x70-0x20($Xip),$Ii     # borrow $Ii for $Hkey^6
 868         vpxor           $Z1,$Z0,$Z0
 869          vpunpckhqdq    $Xi,$Xi,$T3
 870         vpclmulqdq      \$0x11,$Hkey,$inout1,$inout1
 871          vpxor          $Xi,$T3,$T3
 872         vpxor           $inout2,$inout1,$inout1
 873         vpclmulqdq      \$0x00,$HK,$T2,$T2
 874         vpxor           $T1,$T2,$T2
 875
 876         vpclmulqdq      \$0x00,$Ii,$Xi,$Z1
 877         vpclmulqdq      \$0x11,$Ii,$Xi,$Z3
 878         vpxor           $Z0,$Z1,$Z1
 879         vpclmulqdq      \$0x10,$HK,$T3,$Z2
 880         vpxor           $inout1,$Z3,$Z3
 881         vpxor           $T2,$Z2,$Z2
 882
 883         vpxor           $Z1,$Z3,$Z0             # aggregated Karatsuba post-processing
 884         vpxor           $Z0,$Z2,$Z2
 885         vpslldq         \$8,$Z2,$T1
 886         vmovdqu         0x10($const),$Hkey      # .Lpoly
 887         vpsrldq         \$8,$Z2,$Z2
 888         vpxor           $T1,$Z1,$Xi
 889         vpxor           $Z2,$Z3,$Z3
 890
 891         vpalignr        \$8,$Xi,$Xi,$T2         # 1st phase
 892         vpclmulqdq      \$0x10,$Hkey,$Xi,$Xi
 893         vpxor           $T2,$Xi,$Xi
 894
 895         vpalignr        \$8,$Xi,$Xi,$T2         # 2nd phase
 896         vpclmulqdq      \$0x10,$Hkey,$Xi,$Xi
 897         vpxor           $Z3,$T2,$T2
 898         vpxor           $T2,$Xi,$Xi
 899 ___
 900 }
 901 $code.=<<___;
 902         vpshufb         ($const),$Xi,$Xi        # .Lbswap_mask
 903         vmovdqu         $Xi,-0x40($Xip)         # output Xi
 904
 905         vzeroupper
 906 ___
 907 $code.=<<___ if ($win64);
 908         movaps  -0xd8(%rax),%xmm6
 909         movaps  -0xc8(%rax),%xmm7
 910         movaps  -0xb8(%rax),%xmm8
 911         movaps  -0xa8(%rax),%xmm9
 912         movaps  -0x98(%rax),%xmm10
 913         movaps  -0x88(%rax),%xmm11
 914         movaps  -0x78(%rax),%xmm12
 915         movaps  -0x68(%rax),%xmm13
 916         movaps  -0x58(%rax),%xmm14
 917         movaps  -0x48(%rax),%xmm15
 918 ___
 919 $code.=<<___;
 920         mov     -48(%rax),%r15
 921 .cfi_restore    %r15
 922         mov     -40(%rax),%r14
 923 .cfi_restore    %r14
 924         mov     -32(%rax),%r13
 925 .cfi_restore    %r13
 926         mov     -24(%rax),%r12
 927 .cfi_restore    %r12
 928         mov     -16(%rax),%rbp
 929 .cfi_restore    %rbp
 930         mov     -8(%rax),%rbx
 931 .cfi_restore    %rbx
 932         lea     (%rax),%rsp             # restore %rsp
 933 .cfi_def_cfa_register   %rsp
 934 .Lgcm_enc_abort:
 935         mov     $ret,%rax               # return value
 936         ret
 937 .cfi_endproc
 938 .size   aesni_gcm_encrypt,.-aesni_gcm_encrypt
 939 ___
 940
 941 $code.=<<___;
 942 .align  64
 943 .Lbswap_mask:
 944         .byte   15,14,13,12,11,10,9,8,7,6,5,4,3,2,1,0
 945 .Lpoly:
 946         .byte   0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0xc2
 947 .Lone_msb:
 948         .byte   0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1
 949 .Ltwo_lsb:
 950         .byte   2,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0
 951 .Lone_lsb:
 952         .byte   1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0
 953 .asciz  "AES-NI GCM module for x86_64, CRYPTOGAMS by <appro\@openssl.org>"
 954 .align  64
 955 ___
 956 if ($win64) {
 957 $rec="%rcx";
 958 $frame="%rdx";
 959 $context="%r8";
 960 $disp="%r9";
 961
 962 $code.=<<___
 963 .extern __imp_RtlVirtualUnwind
 964 .type   gcm_se_handler,\@abi-omnipotent
 965 .align  16
 966 gcm_se_handler:
 967         push    %rsi
 968         push    %rdi
 969         push    %rbx
 970         push    %rbp
 971         push    %r12
 972         push    %r13
 973         push    %r14
 974         push    %r15
 975         pushfq
 976         sub     \$64,%rsp
 977
 978         mov     120($context),%rax      # pull context->Rax
 979         mov     248($context),%rbx      # pull context->Rip
 980
 981         mov     8($disp),%rsi           # disp->ImageBase
 982         mov     56($disp),%r11          # disp->HandlerData
 983
 984         mov     0(%r11),%r10d           # HandlerData[0]
 985         lea     (%rsi,%r10),%r10        # prologue label
 986         cmp     %r10,%rbx               # context->Rip<prologue label
 987         jb      .Lcommon_seh_tail
 988
 989         mov     152($context),%rax      # pull context->Rsp
 990
 991         mov     4(%r11),%r10d           # HandlerData[1]
 992         lea     (%rsi,%r10),%r10        # epilogue label
 993         cmp     %r10,%rbx               # context->Rip>=epilogue label
 994         jae     .Lcommon_seh_tail
 995
 996         mov     120($context),%rax      # pull context->Rax
 997
 998         mov     -48(%rax),%r15
 999         mov     -40(%rax),%r14
1000         mov     -32(%rax),%r13
1001         mov     -24(%rax),%r12
1002         mov     -16(%rax),%rbp
1003         mov     -8(%rax),%rbx
1004         mov     %r15,240($context)
1005         mov     %r14,232($context)
1006         mov     %r13,224($context)
1007         mov     %r12,216($context)
1008         mov     %rbp,160($context)
1009         mov     %rbx,144($context)
1010
1011         lea     -0xd8(%rax),%rsi        # %xmm save area
1012         lea     512($context),%rdi      # & context.Xmm6
1013         mov     \$20,%ecx               # 10*sizeof(%xmm0)/sizeof(%rax)
1014         .long   0xa548f3fc              # cld; rep movsq
1015
1016 .Lcommon_seh_tail:
1017         mov     8(%rax),%rdi
1018         mov     16(%rax),%rsi
1019         mov     %rax,152($context)      # restore context->Rsp
1020         mov     %rsi,168($context)      # restore context->Rsi
1021         mov     %rdi,176($context)      # restore context->Rdi
1022
1023         mov     40($disp),%rdi          # disp->ContextRecord
1024         mov     $context,%rsi           # context
1025         mov     \$154,%ecx              # sizeof(CONTEXT)
1026         .long   0xa548f3fc              # cld; rep movsq
1027
1028         mov     $disp,%rsi
1029         xor     %rcx,%rcx               # arg1, UNW_FLAG_NHANDLER
1030         mov     8(%rsi),%rdx            # arg2, disp->ImageBase
1031         mov     0(%rsi),%r8             # arg3, disp->ControlPc
1032         mov     16(%rsi),%r9            # arg4, disp->FunctionEntry
1033         mov     40(%rsi),%r10           # disp->ContextRecord
1034         lea     56(%rsi),%r11           # &disp->HandlerData
1035         lea     24(%rsi),%r12           # &disp->EstablisherFrame
1036         mov     %r10,32(%rsp)           # arg5
1037         mov     %r11,40(%rsp)           # arg6
1038         mov     %r12,48(%rsp)           # arg7
1039         mov     %rcx,56(%rsp)           # arg8, (NULL)
1040         call    *__imp_RtlVirtualUnwind(%rip)
1041
1042         mov     \$1,%eax                # ExceptionContinueSearch
1043         add     \$64,%rsp
1044         popfq
1045         pop     %r15
1046         pop     %r14
1047         pop     %r13
1048         pop     %r12
1049         pop     %rbp
1050         pop     %rbx
1051         pop     %rdi
1052         pop     %rsi
1053         ret
1054 .size   gcm_se_handler,.-gcm_se_handler
1055
1056 .section        .pdata
1057 .align  4
1058         .rva    .LSEH_begin_aesni_gcm_decrypt
1059         .rva    .LSEH_end_aesni_gcm_decrypt
1060         .rva    .LSEH_gcm_dec_info
1061
1062         .rva    .LSEH_begin_aesni_gcm_encrypt
1063         .rva    .LSEH_end_aesni_gcm_encrypt
1064         .rva    .LSEH_gcm_enc_info
1065 .section        .xdata
1066 .align  8
1067 .LSEH_gcm_dec_info:
1068         .byte   9,0,0,0
1069         .rva    gcm_se_handler
1070         .rva    .Lgcm_dec_body,.Lgcm_dec_abort
1071 .LSEH_gcm_enc_info:
1072         .byte   9,0,0,0
1073         .rva    gcm_se_handler
1074         .rva    .Lgcm_enc_body,.Lgcm_enc_abort
1075 ___
1076 }
1077 }}} else {{{
1078 $code=<<___;    # assembler is too old
1079 .text
1080
1081 .globl  aesni_gcm_encrypt
1082 .type   aesni_gcm_encrypt,\@abi-omnipotent
1083 aesni_gcm_encrypt:
1084         xor     %eax,%eax
1085         ret
1086 .size   aesni_gcm_encrypt,.-aesni_gcm_encrypt
1087
1088 .globl  aesni_gcm_decrypt
1089 .type   aesni_gcm_decrypt,\@abi-omnipotent
1090 aesni_gcm_decrypt:
1091         xor     %eax,%eax
1092         ret
1093 .size   aesni_gcm_decrypt,.-aesni_gcm_decrypt
1094 ___
1095 }}}
1096
1097 $code =~ s/\`([^\`]*)\`/eval($1)/gem;
1098
1099 print $code;
1100
1101 close STDOUT;