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 OpenSSL license (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) ([0-9]+\.[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 $code=<<___;
  88 .text
  89
  90 .type   _aesni_ctr32_ghash_6x,\@abi-omnipotent
  91 .align  32
  92 _aesni_ctr32_ghash_6x:
  93 .cfi_startproc
  94         vmovdqu         0x20($const),$T2        # borrow $T2, .Lone_msb
  95         sub             \$6,$len
  96         vpxor           $Z0,$Z0,$Z0             # $Z0   = 0
  97         vmovdqu         0x00-0x80($key),$rndkey
  98         vpaddb          $T2,$T1,$inout1
  99         vpaddb          $T2,$inout1,$inout2
 100         vpaddb          $T2,$inout2,$inout3
 101         vpaddb          $T2,$inout3,$inout4
 102         vpaddb          $T2,$inout4,$inout5
 103         vpxor           $rndkey,$T1,$inout0
 104         vmovdqu         $Z0,16+8(%rsp)          # "$Z3" = 0
 105         jmp             .Loop6x
 106
 107 .align  32
 108 .Loop6x:
 109         add             \$`6<<24`,$counter
 110         jc              .Lhandle_ctr32          # discard $inout[1-5]?
 111         vmovdqu         0x00-0x20($Xip),$Hkey   # $Hkey^1
 112           vpaddb        $T2,$inout5,$T1         # next counter value
 113           vpxor         $rndkey,$inout1,$inout1
 114           vpxor         $rndkey,$inout2,$inout2
 115
 116 .Lresume_ctr32:
 117         vmovdqu         $T1,($ivp)              # save next counter value
 118         vpclmulqdq      \$0x10,$Hkey,$Z3,$Z1
 119           vpxor         $rndkey,$inout3,$inout3
 120           vmovups       0x10-0x80($key),$T2     # borrow $T2 for $rndkey
 121         vpclmulqdq      \$0x01,$Hkey,$Z3,$Z2
 122         xor             %r12,%r12
 123         cmp             $in0,$end0
 124
 125           vaesenc       $T2,$inout0,$inout0
 126         vmovdqu         0x30+8(%rsp),$Ii        # I[4]
 127           vpxor         $rndkey,$inout4,$inout4
 128         vpclmulqdq      \$0x00,$Hkey,$Z3,$T1
 129           vaesenc       $T2,$inout1,$inout1
 130           vpxor         $rndkey,$inout5,$inout5
 131         setnc           %r12b
 132         vpclmulqdq      \$0x11,$Hkey,$Z3,$Z3
 133           vaesenc       $T2,$inout2,$inout2
 134         vmovdqu         0x10-0x20($Xip),$Hkey   # $Hkey^2
 135         neg             %r12
 136           vaesenc       $T2,$inout3,$inout3
 137          vpxor          $Z1,$Z2,$Z2
 138         vpclmulqdq      \$0x00,$Hkey,$Ii,$Z1
 139          vpxor          $Z0,$Xi,$Xi             # modulo-scheduled
 140           vaesenc       $T2,$inout4,$inout4
 141          vpxor          $Z1,$T1,$Z0
 142         and             \$0x60,%r12
 143           vmovups       0x20-0x80($key),$rndkey
 144         vpclmulqdq      \$0x10,$Hkey,$Ii,$T1
 145           vaesenc       $T2,$inout5,$inout5
 146
 147         vpclmulqdq      \$0x01,$Hkey,$Ii,$T2
 148         lea             ($in0,%r12),$in0
 149           vaesenc       $rndkey,$inout0,$inout0
 150          vpxor          16+8(%rsp),$Xi,$Xi      # modulo-scheduled [vpxor $Z3,$Xi,$Xi]
 151         vpclmulqdq      \$0x11,$Hkey,$Ii,$Hkey
 152          vmovdqu        0x40+8(%rsp),$Ii        # I[3]
 153           vaesenc       $rndkey,$inout1,$inout1
 154         movbe           0x58($in0),%r13
 155           vaesenc       $rndkey,$inout2,$inout2
 156         movbe           0x50($in0),%r12
 157           vaesenc       $rndkey,$inout3,$inout3
 158         mov             %r13,0x20+8(%rsp)
 159           vaesenc       $rndkey,$inout4,$inout4
 160         mov             %r12,0x28+8(%rsp)
 161         vmovdqu         0x30-0x20($Xip),$Z1     # borrow $Z1 for $Hkey^3
 162           vaesenc       $rndkey,$inout5,$inout5
 163
 164           vmovups       0x30-0x80($key),$rndkey
 165          vpxor          $T1,$Z2,$Z2
 166         vpclmulqdq      \$0x00,$Z1,$Ii,$T1
 167           vaesenc       $rndkey,$inout0,$inout0
 168          vpxor          $T2,$Z2,$Z2
 169         vpclmulqdq      \$0x10,$Z1,$Ii,$T2
 170           vaesenc       $rndkey,$inout1,$inout1
 171          vpxor          $Hkey,$Z3,$Z3
 172         vpclmulqdq      \$0x01,$Z1,$Ii,$Hkey
 173           vaesenc       $rndkey,$inout2,$inout2
 174         vpclmulqdq      \$0x11,$Z1,$Ii,$Z1
 175          vmovdqu        0x50+8(%rsp),$Ii        # I[2]
 176           vaesenc       $rndkey,$inout3,$inout3
 177           vaesenc       $rndkey,$inout4,$inout4
 178          vpxor          $T1,$Z0,$Z0
 179         vmovdqu         0x40-0x20($Xip),$T1     # borrow $T1 for $Hkey^4
 180           vaesenc       $rndkey,$inout5,$inout5
 181
 182           vmovups       0x40-0x80($key),$rndkey
 183          vpxor          $T2,$Z2,$Z2
 184         vpclmulqdq      \$0x00,$T1,$Ii,$T2
 185           vaesenc       $rndkey,$inout0,$inout0
 186          vpxor          $Hkey,$Z2,$Z2
 187         vpclmulqdq      \$0x10,$T1,$Ii,$Hkey
 188           vaesenc       $rndkey,$inout1,$inout1
 189         movbe           0x48($in0),%r13
 190          vpxor          $Z1,$Z3,$Z3
 191         vpclmulqdq      \$0x01,$T1,$Ii,$Z1
 192           vaesenc       $rndkey,$inout2,$inout2
 193         movbe           0x40($in0),%r12
 194         vpclmulqdq      \$0x11,$T1,$Ii,$T1
 195          vmovdqu        0x60+8(%rsp),$Ii        # I[1]
 196           vaesenc       $rndkey,$inout3,$inout3
 197         mov             %r13,0x30+8(%rsp)
 198           vaesenc       $rndkey,$inout4,$inout4
 199         mov             %r12,0x38+8(%rsp)
 200          vpxor          $T2,$Z0,$Z0
 201         vmovdqu         0x60-0x20($Xip),$T2     # borrow $T2 for $Hkey^5
 202           vaesenc       $rndkey,$inout5,$inout5
 203
 204           vmovups       0x50-0x80($key),$rndkey
 205          vpxor          $Hkey,$Z2,$Z2
 206         vpclmulqdq      \$0x00,$T2,$Ii,$Hkey
 207           vaesenc       $rndkey,$inout0,$inout0
 208          vpxor          $Z1,$Z2,$Z2
 209         vpclmulqdq      \$0x10,$T2,$Ii,$Z1
 210           vaesenc       $rndkey,$inout1,$inout1
 211         movbe           0x38($in0),%r13
 212          vpxor          $T1,$Z3,$Z3
 213         vpclmulqdq      \$0x01,$T2,$Ii,$T1
 214          vpxor          0x70+8(%rsp),$Xi,$Xi    # accumulate I[0]
 215           vaesenc       $rndkey,$inout2,$inout2
 216         movbe           0x30($in0),%r12
 217         vpclmulqdq      \$0x11,$T2,$Ii,$T2
 218           vaesenc       $rndkey,$inout3,$inout3
 219         mov             %r13,0x40+8(%rsp)
 220           vaesenc       $rndkey,$inout4,$inout4
 221         mov             %r12,0x48+8(%rsp)
 222          vpxor          $Hkey,$Z0,$Z0
 223          vmovdqu        0x70-0x20($Xip),$Hkey   # $Hkey^6
 224           vaesenc       $rndkey,$inout5,$inout5
 225
 226           vmovups       0x60-0x80($key),$rndkey
 227          vpxor          $Z1,$Z2,$Z2
 228         vpclmulqdq      \$0x10,$Hkey,$Xi,$Z1
 229           vaesenc       $rndkey,$inout0,$inout0
 230          vpxor          $T1,$Z2,$Z2
 231         vpclmulqdq      \$0x01,$Hkey,$Xi,$T1
 232           vaesenc       $rndkey,$inout1,$inout1
 233         movbe           0x28($in0),%r13
 234          vpxor          $T2,$Z3,$Z3
 235         vpclmulqdq      \$0x00,$Hkey,$Xi,$T2
 236           vaesenc       $rndkey,$inout2,$inout2
 237         movbe           0x20($in0),%r12
 238         vpclmulqdq      \$0x11,$Hkey,$Xi,$Xi
 239           vaesenc       $rndkey,$inout3,$inout3
 240         mov             %r13,0x50+8(%rsp)
 241           vaesenc       $rndkey,$inout4,$inout4
 242         mov             %r12,0x58+8(%rsp)
 243         vpxor           $Z1,$Z2,$Z2
 244           vaesenc       $rndkey,$inout5,$inout5
 245         vpxor           $T1,$Z2,$Z2
 246
 247           vmovups       0x70-0x80($key),$rndkey
 248         vpslldq         \$8,$Z2,$Z1
 249         vpxor           $T2,$Z0,$Z0
 250         vmovdqu         0x10($const),$Hkey      # .Lpoly
 251
 252           vaesenc       $rndkey,$inout0,$inout0
 253         vpxor           $Xi,$Z3,$Z3
 254           vaesenc       $rndkey,$inout1,$inout1
 255         vpxor           $Z1,$Z0,$Z0
 256         movbe           0x18($in0),%r13
 257           vaesenc       $rndkey,$inout2,$inout2
 258         movbe           0x10($in0),%r12
 259         vpalignr        \$8,$Z0,$Z0,$Ii         # 1st phase
 260         vpclmulqdq      \$0x10,$Hkey,$Z0,$Z0
 261         mov             %r13,0x60+8(%rsp)
 262           vaesenc       $rndkey,$inout3,$inout3
 263         mov             %r12,0x68+8(%rsp)
 264           vaesenc       $rndkey,$inout4,$inout4
 265           vmovups       0x80-0x80($key),$T1     # borrow $T1 for $rndkey
 266           vaesenc       $rndkey,$inout5,$inout5
 267
 268           vaesenc       $T1,$inout0,$inout0
 269           vmovups       0x90-0x80($key),$rndkey
 270           vaesenc       $T1,$inout1,$inout1
 271         vpsrldq         \$8,$Z2,$Z2
 272           vaesenc       $T1,$inout2,$inout2
 273         vpxor           $Z2,$Z3,$Z3
 274           vaesenc       $T1,$inout3,$inout3
 275         vpxor           $Ii,$Z0,$Z0
 276         movbe           0x08($in0),%r13
 277           vaesenc       $T1,$inout4,$inout4
 278         movbe           0x00($in0),%r12
 279           vaesenc       $T1,$inout5,$inout5
 280           vmovups       0xa0-0x80($key),$T1
 281           cmp           \$11,$rounds
 282           jb            .Lenc_tail              # 128-bit key
 283
 284           vaesenc       $rndkey,$inout0,$inout0
 285           vaesenc       $rndkey,$inout1,$inout1
 286           vaesenc       $rndkey,$inout2,$inout2
 287           vaesenc       $rndkey,$inout3,$inout3
 288           vaesenc       $rndkey,$inout4,$inout4
 289           vaesenc       $rndkey,$inout5,$inout5
 290
 291           vaesenc       $T1,$inout0,$inout0
 292           vaesenc       $T1,$inout1,$inout1
 293           vaesenc       $T1,$inout2,$inout2
 294           vaesenc       $T1,$inout3,$inout3
 295           vaesenc       $T1,$inout4,$inout4
 296           vmovups       0xb0-0x80($key),$rndkey
 297           vaesenc       $T1,$inout5,$inout5
 298           vmovups       0xc0-0x80($key),$T1
 299           je            .Lenc_tail              # 192-bit key
 300
 301           vaesenc       $rndkey,$inout0,$inout0
 302           vaesenc       $rndkey,$inout1,$inout1
 303           vaesenc       $rndkey,$inout2,$inout2
 304           vaesenc       $rndkey,$inout3,$inout3
 305           vaesenc       $rndkey,$inout4,$inout4
 306           vaesenc       $rndkey,$inout5,$inout5
 307
 308           vaesenc       $T1,$inout0,$inout0
 309           vaesenc       $T1,$inout1,$inout1
 310           vaesenc       $T1,$inout2,$inout2
 311           vaesenc       $T1,$inout3,$inout3
 312           vaesenc       $T1,$inout4,$inout4
 313           vmovups       0xd0-0x80($key),$rndkey
 314           vaesenc       $T1,$inout5,$inout5
 315           vmovups       0xe0-0x80($key),$T1
 316           jmp           .Lenc_tail              # 256-bit key
 317
 318 .align  32
 319 .Lhandle_ctr32:
 320         vmovdqu         ($const),$Ii            # borrow $Ii for .Lbswap_mask
 321           vpshufb       $Ii,$T1,$Z2             # byte-swap counter
 322           vmovdqu       0x30($const),$Z1        # borrow $Z1, .Ltwo_lsb
 323           vpaddd        0x40($const),$Z2,$inout1        # .Lone_lsb
 324           vpaddd        $Z1,$Z2,$inout2
 325         vmovdqu         0x00-0x20($Xip),$Hkey   # $Hkey^1
 326           vpaddd        $Z1,$inout1,$inout3
 327           vpshufb       $Ii,$inout1,$inout1
 328           vpaddd        $Z1,$inout2,$inout4
 329           vpshufb       $Ii,$inout2,$inout2
 330           vpxor         $rndkey,$inout1,$inout1
 331           vpaddd        $Z1,$inout3,$inout5
 332           vpshufb       $Ii,$inout3,$inout3
 333           vpxor         $rndkey,$inout2,$inout2
 334           vpaddd        $Z1,$inout4,$T1         # byte-swapped next counter value
 335           vpshufb       $Ii,$inout4,$inout4
 336           vpshufb       $Ii,$inout5,$inout5
 337           vpshufb       $Ii,$T1,$T1             # next counter value
 338         jmp             .Lresume_ctr32
 339
 340 .align  32
 341 .Lenc_tail:
 342           vaesenc       $rndkey,$inout0,$inout0
 343         vmovdqu         $Z3,16+8(%rsp)          # postpone vpxor $Z3,$Xi,$Xi
 344         vpalignr        \$8,$Z0,$Z0,$Xi         # 2nd phase
 345           vaesenc       $rndkey,$inout1,$inout1
 346         vpclmulqdq      \$0x10,$Hkey,$Z0,$Z0
 347           vpxor         0x00($inp),$T1,$T2
 348           vaesenc       $rndkey,$inout2,$inout2
 349           vpxor         0x10($inp),$T1,$Ii
 350           vaesenc       $rndkey,$inout3,$inout3
 351           vpxor         0x20($inp),$T1,$Z1
 352           vaesenc       $rndkey,$inout4,$inout4
 353           vpxor         0x30($inp),$T1,$Z2
 354           vaesenc       $rndkey,$inout5,$inout5
 355           vpxor         0x40($inp),$T1,$Z3
 356           vpxor         0x50($inp),$T1,$Hkey
 357           vmovdqu       ($ivp),$T1              # load next counter value
 358
 359           vaesenclast   $T2,$inout0,$inout0
 360           vmovdqu       0x20($const),$T2        # borrow $T2, .Lone_msb
 361           vaesenclast   $Ii,$inout1,$inout1
 362          vpaddb         $T2,$T1,$Ii
 363         mov             %r13,0x70+8(%rsp)
 364         lea             0x60($inp),$inp
 365           vaesenclast   $Z1,$inout2,$inout2
 366          vpaddb         $T2,$Ii,$Z1
 367         mov             %r12,0x78+8(%rsp)
 368         lea             0x60($out),$out
 369           vmovdqu       0x00-0x80($key),$rndkey
 370           vaesenclast   $Z2,$inout3,$inout3
 371          vpaddb         $T2,$Z1,$Z2
 372           vaesenclast   $Z3, $inout4,$inout4
 373          vpaddb         $T2,$Z2,$Z3
 374           vaesenclast   $Hkey,$inout5,$inout5
 375          vpaddb         $T2,$Z3,$Hkey
 376
 377         add             \$0x60,$ret
 378         sub             \$0x6,$len
 379         jc              .L6x_done
 380
 381           vmovups       $inout0,-0x60($out)     # save output
 382          vpxor          $rndkey,$T1,$inout0
 383           vmovups       $inout1,-0x50($out)
 384          vmovdqa        $Ii,$inout1             # 0 latency
 385           vmovups       $inout2,-0x40($out)
 386          vmovdqa        $Z1,$inout2             # 0 latency
 387           vmovups       $inout3,-0x30($out)
 388          vmovdqa        $Z2,$inout3             # 0 latency
 389           vmovups       $inout4,-0x20($out)
 390          vmovdqa        $Z3,$inout4             # 0 latency
 391           vmovups       $inout5,-0x10($out)
 392          vmovdqa        $Hkey,$inout5           # 0 latency
 393         vmovdqu         0x20+8(%rsp),$Z3        # I[5]
 394         jmp             .Loop6x
 395
 396 .L6x_done:
 397         vpxor           16+8(%rsp),$Xi,$Xi      # modulo-scheduled
 398         vpxor           $Z0,$Xi,$Xi             # modulo-scheduled
 399
 400         ret
 401 .cfi_endproc
 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 $code.=<<___;
 545 .type   _aesni_ctr32_6x,\@abi-omnipotent
 546 .align  32
 547 _aesni_ctr32_6x:
 548 .cfi_startproc
 549         vmovdqu         0x00-0x80($key),$Z0     # borrow $Z0 for $rndkey
 550         vmovdqu         0x20($const),$T2        # borrow $T2, .Lone_msb
 551         lea             -1($rounds),%r13
 552         vmovups         0x10-0x80($key),$rndkey
 553         lea             0x20-0x80($key),%r12
 554         vpxor           $Z0,$T1,$inout0
 555         add             \$`6<<24`,$counter
 556         jc              .Lhandle_ctr32_2
 557         vpaddb          $T2,$T1,$inout1
 558         vpaddb          $T2,$inout1,$inout2
 559         vpxor           $Z0,$inout1,$inout1
 560         vpaddb          $T2,$inout2,$inout3
 561         vpxor           $Z0,$inout2,$inout2
 562         vpaddb          $T2,$inout3,$inout4
 563         vpxor           $Z0,$inout3,$inout3
 564         vpaddb          $T2,$inout4,$inout5
 565         vpxor           $Z0,$inout4,$inout4
 566         vpaddb          $T2,$inout5,$T1
 567         vpxor           $Z0,$inout5,$inout5
 568         jmp             .Loop_ctr32
 569
 570 .align  16
 571 .Loop_ctr32:
 572         vaesenc         $rndkey,$inout0,$inout0
 573         vaesenc         $rndkey,$inout1,$inout1
 574         vaesenc         $rndkey,$inout2,$inout2
 575         vaesenc         $rndkey,$inout3,$inout3
 576         vaesenc         $rndkey,$inout4,$inout4
 577         vaesenc         $rndkey,$inout5,$inout5
 578         vmovups         (%r12),$rndkey
 579         lea             0x10(%r12),%r12
 580         dec             %r13d
 581         jnz             .Loop_ctr32
 582
 583         vmovdqu         (%r12),$Hkey            # last round key
 584         vaesenc         $rndkey,$inout0,$inout0
 585         vpxor           0x00($inp),$Hkey,$Z0
 586         vaesenc         $rndkey,$inout1,$inout1
 587         vpxor           0x10($inp),$Hkey,$Z1
 588         vaesenc         $rndkey,$inout2,$inout2
 589         vpxor           0x20($inp),$Hkey,$Z2
 590         vaesenc         $rndkey,$inout3,$inout3
 591         vpxor           0x30($inp),$Hkey,$Xi
 592         vaesenc         $rndkey,$inout4,$inout4
 593         vpxor           0x40($inp),$Hkey,$T2
 594         vaesenc         $rndkey,$inout5,$inout5
 595         vpxor           0x50($inp),$Hkey,$Hkey
 596         lea             0x60($inp),$inp
 597
 598         vaesenclast     $Z0,$inout0,$inout0
 599         vaesenclast     $Z1,$inout1,$inout1
 600         vaesenclast     $Z2,$inout2,$inout2
 601         vaesenclast     $Xi,$inout3,$inout3
 602         vaesenclast     $T2,$inout4,$inout4
 603         vaesenclast     $Hkey,$inout5,$inout5
 604         vmovups         $inout0,0x00($out)
 605         vmovups         $inout1,0x10($out)
 606         vmovups         $inout2,0x20($out)
 607         vmovups         $inout3,0x30($out)
 608         vmovups         $inout4,0x40($out)
 609         vmovups         $inout5,0x50($out)
 610         lea             0x60($out),$out
 611
 612         ret
 613 .align  32
 614 .Lhandle_ctr32_2:
 615         vpshufb         $Ii,$T1,$Z2             # byte-swap counter
 616         vmovdqu         0x30($const),$Z1        # borrow $Z1, .Ltwo_lsb
 617         vpaddd          0x40($const),$Z2,$inout1        # .Lone_lsb
 618         vpaddd          $Z1,$Z2,$inout2
 619         vpaddd          $Z1,$inout1,$inout3
 620         vpshufb         $Ii,$inout1,$inout1
 621         vpaddd          $Z1,$inout2,$inout4
 622         vpshufb         $Ii,$inout2,$inout2
 623         vpxor           $Z0,$inout1,$inout1
 624         vpaddd          $Z1,$inout3,$inout5
 625         vpshufb         $Ii,$inout3,$inout3
 626         vpxor           $Z0,$inout2,$inout2
 627         vpaddd          $Z1,$inout4,$T1         # byte-swapped next counter value
 628         vpshufb         $Ii,$inout4,$inout4
 629         vpxor           $Z0,$inout3,$inout3
 630         vpshufb         $Ii,$inout5,$inout5
 631         vpxor           $Z0,$inout4,$inout4
 632         vpshufb         $Ii,$T1,$T1             # next counter value
 633         vpxor           $Z0,$inout5,$inout5
 634         jmp     .Loop_ctr32
 635 .cfi_endproc
 636 .size   _aesni_ctr32_6x,.-_aesni_ctr32_6x
 637
 638 .globl  aesni_gcm_encrypt
 639 .type   aesni_gcm_encrypt,\@function,6
 640 .align  32
 641 aesni_gcm_encrypt:
 642 .cfi_startproc
 643         xor     $ret,$ret
 644         cmp     \$0x60*3,$len                   # minimal accepted length
 645         jb      .Lgcm_enc_abort
 646
 647         lea     (%rsp),%rax                     # save stack pointer
 648 .cfi_def_cfa_register   %rax
 649         push    %rbx
 650 .cfi_push       %rbx
 651         push    %rbp
 652 .cfi_push       %rbp
 653         push    %r12
 654 .cfi_push       %r12
 655         push    %r13
 656 .cfi_push       %r13
 657         push    %r14
 658 .cfi_push       %r14
 659         push    %r15
 660 .cfi_push       %r15
 661 ___
 662 $code.=<<___ if ($win64);
 663         lea     -0xa8(%rsp),%rsp
 664         movaps  %xmm6,-0xd8(%rax)
 665         movaps  %xmm7,-0xc8(%rax)
 666         movaps  %xmm8,-0xb8(%rax)
 667         movaps  %xmm9,-0xa8(%rax)
 668         movaps  %xmm10,-0x98(%rax)
 669         movaps  %xmm11,-0x88(%rax)
 670         movaps  %xmm12,-0x78(%rax)
 671         movaps  %xmm13,-0x68(%rax)
 672         movaps  %xmm14,-0x58(%rax)
 673         movaps  %xmm15,-0x48(%rax)
 674 .Lgcm_enc_body:
 675 ___
 676 $code.=<<___;
 677         vzeroupper
 678
 679         vmovdqu         ($ivp),$T1              # input counter value
 680         add             \$-128,%rsp
 681         mov             12($ivp),$counter
 682         lea             .Lbswap_mask(%rip),$const
 683         lea             -0x80($key),$in0        # borrow $in0
 684         mov             \$0xf80,$end0           # borrow $end0
 685         lea             0x80($key),$key         # size optimization
 686         vmovdqu         ($const),$Ii            # borrow $Ii for .Lbswap_mask
 687         and             \$-128,%rsp             # ensure stack alignment
 688         mov             0xf0-0x80($key),$rounds
 689
 690         and             $end0,$in0
 691         and             %rsp,$end0
 692         sub             $in0,$end0
 693         jc              .Lenc_no_key_aliasing
 694         cmp             \$768,$end0
 695         jnc             .Lenc_no_key_aliasing
 696         sub             $end0,%rsp              # avoid aliasing with key
 697 .Lenc_no_key_aliasing:
 698
 699         lea             ($out),$in0
 700         lea             -0xc0($out,$len),$end0
 701         shr             \$4,$len
 702
 703         call            _aesni_ctr32_6x
 704         vpshufb         $Ii,$inout0,$Xi         # save bswapped output on stack
 705         vpshufb         $Ii,$inout1,$T2
 706         vmovdqu         $Xi,0x70(%rsp)
 707         vpshufb         $Ii,$inout2,$Z0
 708         vmovdqu         $T2,0x60(%rsp)
 709         vpshufb         $Ii,$inout3,$Z1
 710         vmovdqu         $Z0,0x50(%rsp)
 711         vpshufb         $Ii,$inout4,$Z2
 712         vmovdqu         $Z1,0x40(%rsp)
 713         vpshufb         $Ii,$inout5,$Z3         # passed to _aesni_ctr32_ghash_6x
 714         vmovdqu         $Z2,0x30(%rsp)
 715
 716         call            _aesni_ctr32_6x
 717
 718         vmovdqu         ($Xip),$Xi              # load Xi
 719         lea             0x20+0x20($Xip),$Xip    # size optimization
 720         sub             \$12,$len
 721         mov             \$0x60*2,$ret
 722         vpshufb         $Ii,$Xi,$Xi
 723
 724         call            _aesni_ctr32_ghash_6x
 725         vmovdqu         0x20(%rsp),$Z3          # I[5]
 726          vmovdqu        ($const),$Ii            # borrow $Ii for .Lbswap_mask
 727         vmovdqu         0x00-0x20($Xip),$Hkey   # $Hkey^1
 728         vpunpckhqdq     $Z3,$Z3,$T1
 729         vmovdqu         0x20-0x20($Xip),$rndkey # borrow $rndkey for $HK
 730          vmovups        $inout0,-0x60($out)     # save output
 731          vpshufb        $Ii,$inout0,$inout0     # but keep bswapped copy
 732         vpxor           $Z3,$T1,$T1
 733          vmovups        $inout1,-0x50($out)
 734          vpshufb        $Ii,$inout1,$inout1
 735          vmovups        $inout2,-0x40($out)
 736          vpshufb        $Ii,$inout2,$inout2
 737          vmovups        $inout3,-0x30($out)
 738          vpshufb        $Ii,$inout3,$inout3
 739          vmovups        $inout4,-0x20($out)
 740          vpshufb        $Ii,$inout4,$inout4
 741          vmovups        $inout5,-0x10($out)
 742          vpshufb        $Ii,$inout5,$inout5
 743          vmovdqu        $inout0,0x10(%rsp)      # free $inout0
 744 ___
 745 { my ($HK,$T3)=($rndkey,$inout0);
 746
 747 $code.=<<___;
 748          vmovdqu        0x30(%rsp),$Z2          # I[4]
 749          vmovdqu        0x10-0x20($Xip),$Ii     # borrow $Ii for $Hkey^2
 750          vpunpckhqdq    $Z2,$Z2,$T2
 751         vpclmulqdq      \$0x00,$Hkey,$Z3,$Z1
 752          vpxor          $Z2,$T2,$T2
 753         vpclmulqdq      \$0x11,$Hkey,$Z3,$Z3
 754         vpclmulqdq      \$0x00,$HK,$T1,$T1
 755
 756          vmovdqu        0x40(%rsp),$T3          # I[3]
 757         vpclmulqdq      \$0x00,$Ii,$Z2,$Z0
 758          vmovdqu        0x30-0x20($Xip),$Hkey   # $Hkey^3
 759         vpxor           $Z1,$Z0,$Z0
 760          vpunpckhqdq    $T3,$T3,$Z1
 761         vpclmulqdq      \$0x11,$Ii,$Z2,$Z2
 762          vpxor          $T3,$Z1,$Z1
 763         vpxor           $Z3,$Z2,$Z2
 764         vpclmulqdq      \$0x10,$HK,$T2,$T2
 765          vmovdqu        0x50-0x20($Xip),$HK
 766         vpxor           $T1,$T2,$T2
 767
 768          vmovdqu        0x50(%rsp),$T1          # I[2]
 769         vpclmulqdq      \$0x00,$Hkey,$T3,$Z3
 770          vmovdqu        0x40-0x20($Xip),$Ii     # borrow $Ii for $Hkey^4
 771         vpxor           $Z0,$Z3,$Z3
 772          vpunpckhqdq    $T1,$T1,$Z0
 773         vpclmulqdq      \$0x11,$Hkey,$T3,$T3
 774          vpxor          $T1,$Z0,$Z0
 775         vpxor           $Z2,$T3,$T3
 776         vpclmulqdq      \$0x00,$HK,$Z1,$Z1
 777         vpxor           $T2,$Z1,$Z1
 778
 779          vmovdqu        0x60(%rsp),$T2          # I[1]
 780         vpclmulqdq      \$0x00,$Ii,$T1,$Z2
 781          vmovdqu        0x60-0x20($Xip),$Hkey   # $Hkey^5
 782         vpxor           $Z3,$Z2,$Z2
 783          vpunpckhqdq    $T2,$T2,$Z3
 784         vpclmulqdq      \$0x11,$Ii,$T1,$T1
 785          vpxor          $T2,$Z3,$Z3
 786         vpxor           $T3,$T1,$T1
 787         vpclmulqdq      \$0x10,$HK,$Z0,$Z0
 788          vmovdqu        0x80-0x20($Xip),$HK
 789         vpxor           $Z1,$Z0,$Z0
 790
 791          vpxor          0x70(%rsp),$Xi,$Xi      # accumulate I[0]
 792         vpclmulqdq      \$0x00,$Hkey,$T2,$Z1
 793          vmovdqu        0x70-0x20($Xip),$Ii     # borrow $Ii for $Hkey^6
 794          vpunpckhqdq    $Xi,$Xi,$T3
 795         vpxor           $Z2,$Z1,$Z1
 796         vpclmulqdq      \$0x11,$Hkey,$T2,$T2
 797          vpxor          $Xi,$T3,$T3
 798         vpxor           $T1,$T2,$T2
 799         vpclmulqdq      \$0x00,$HK,$Z3,$Z3
 800         vpxor           $Z0,$Z3,$Z0
 801
 802         vpclmulqdq      \$0x00,$Ii,$Xi,$Z2
 803          vmovdqu        0x00-0x20($Xip),$Hkey   # $Hkey^1
 804          vpunpckhqdq    $inout5,$inout5,$T1
 805         vpclmulqdq      \$0x11,$Ii,$Xi,$Xi
 806          vpxor          $inout5,$T1,$T1
 807         vpxor           $Z1,$Z2,$Z1
 808         vpclmulqdq      \$0x10,$HK,$T3,$T3
 809          vmovdqu        0x20-0x20($Xip),$HK
 810         vpxor           $T2,$Xi,$Z3
 811         vpxor           $Z0,$T3,$Z2
 812
 813          vmovdqu        0x10-0x20($Xip),$Ii     # borrow $Ii for $Hkey^2
 814           vpxor         $Z1,$Z3,$T3             # aggregated Karatsuba post-processing
 815         vpclmulqdq      \$0x00,$Hkey,$inout5,$Z0
 816           vpxor         $T3,$Z2,$Z2
 817          vpunpckhqdq    $inout4,$inout4,$T2
 818         vpclmulqdq      \$0x11,$Hkey,$inout5,$inout5
 819          vpxor          $inout4,$T2,$T2
 820           vpslldq       \$8,$Z2,$T3
 821         vpclmulqdq      \$0x00,$HK,$T1,$T1
 822           vpxor         $T3,$Z1,$Xi
 823           vpsrldq       \$8,$Z2,$Z2
 824           vpxor         $Z2,$Z3,$Z3
 825
 826         vpclmulqdq      \$0x00,$Ii,$inout4,$Z1
 827          vmovdqu        0x30-0x20($Xip),$Hkey   # $Hkey^3
 828         vpxor           $Z0,$Z1,$Z1
 829          vpunpckhqdq    $inout3,$inout3,$T3
 830         vpclmulqdq      \$0x11,$Ii,$inout4,$inout4
 831          vpxor          $inout3,$T3,$T3
 832         vpxor           $inout5,$inout4,$inout4
 833           vpalignr      \$8,$Xi,$Xi,$inout5     # 1st phase
 834         vpclmulqdq      \$0x10,$HK,$T2,$T2
 835          vmovdqu        0x50-0x20($Xip),$HK
 836         vpxor           $T1,$T2,$T2
 837
 838         vpclmulqdq      \$0x00,$Hkey,$inout3,$Z0
 839          vmovdqu        0x40-0x20($Xip),$Ii     # borrow $Ii for $Hkey^4
 840         vpxor           $Z1,$Z0,$Z0
 841          vpunpckhqdq    $inout2,$inout2,$T1
 842         vpclmulqdq      \$0x11,$Hkey,$inout3,$inout3
 843          vpxor          $inout2,$T1,$T1
 844         vpxor           $inout4,$inout3,$inout3
 845           vxorps        0x10(%rsp),$Z3,$Z3      # accumulate $inout0
 846         vpclmulqdq      \$0x00,$HK,$T3,$T3
 847         vpxor           $T2,$T3,$T3
 848
 849           vpclmulqdq    \$0x10,0x10($const),$Xi,$Xi
 850           vxorps        $inout5,$Xi,$Xi
 851
 852         vpclmulqdq      \$0x00,$Ii,$inout2,$Z1
 853          vmovdqu        0x60-0x20($Xip),$Hkey   # $Hkey^5
 854         vpxor           $Z0,$Z1,$Z1
 855          vpunpckhqdq    $inout1,$inout1,$T2
 856         vpclmulqdq      \$0x11,$Ii,$inout2,$inout2
 857          vpxor          $inout1,$T2,$T2
 858           vpalignr      \$8,$Xi,$Xi,$inout5     # 2nd phase
 859         vpxor           $inout3,$inout2,$inout2
 860         vpclmulqdq      \$0x10,$HK,$T1,$T1
 861          vmovdqu        0x80-0x20($Xip),$HK
 862         vpxor           $T3,$T1,$T1
 863
 864           vxorps        $Z3,$inout5,$inout5
 865           vpclmulqdq    \$0x10,0x10($const),$Xi,$Xi
 866           vxorps        $inout5,$Xi,$Xi
 867
 868         vpclmulqdq      \$0x00,$Hkey,$inout1,$Z0
 869          vmovdqu        0x70-0x20($Xip),$Ii     # borrow $Ii for $Hkey^6
 870         vpxor           $Z1,$Z0,$Z0
 871          vpunpckhqdq    $Xi,$Xi,$T3
 872         vpclmulqdq      \$0x11,$Hkey,$inout1,$inout1
 873          vpxor          $Xi,$T3,$T3
 874         vpxor           $inout2,$inout1,$inout1
 875         vpclmulqdq      \$0x00,$HK,$T2,$T2
 876         vpxor           $T1,$T2,$T2
 877
 878         vpclmulqdq      \$0x00,$Ii,$Xi,$Z1
 879         vpclmulqdq      \$0x11,$Ii,$Xi,$Z3
 880         vpxor           $Z0,$Z1,$Z1
 881         vpclmulqdq      \$0x10,$HK,$T3,$Z2
 882         vpxor           $inout1,$Z3,$Z3
 883         vpxor           $T2,$Z2,$Z2
 884
 885         vpxor           $Z1,$Z3,$Z0             # aggregated Karatsuba post-processing
 886         vpxor           $Z0,$Z2,$Z2
 887         vpslldq         \$8,$Z2,$T1
 888         vmovdqu         0x10($const),$Hkey      # .Lpoly
 889         vpsrldq         \$8,$Z2,$Z2
 890         vpxor           $T1,$Z1,$Xi
 891         vpxor           $Z2,$Z3,$Z3
 892
 893         vpalignr        \$8,$Xi,$Xi,$T2         # 1st phase
 894         vpclmulqdq      \$0x10,$Hkey,$Xi,$Xi
 895         vpxor           $T2,$Xi,$Xi
 896
 897         vpalignr        \$8,$Xi,$Xi,$T2         # 2nd phase
 898         vpclmulqdq      \$0x10,$Hkey,$Xi,$Xi
 899         vpxor           $Z3,$T2,$T2
 900         vpxor           $T2,$Xi,$Xi
 901 ___
 902 }
 903 $code.=<<___;
 904         vpshufb         ($const),$Xi,$Xi        # .Lbswap_mask
 905         vmovdqu         $Xi,-0x40($Xip)         # output Xi
 906
 907         vzeroupper
 908 ___
 909 $code.=<<___ if ($win64);
 910         movaps  -0xd8(%rax),%xmm6
 911         movaps  -0xc8(%rax),%xmm7
 912         movaps  -0xb8(%rax),%xmm8
 913         movaps  -0xa8(%rax),%xmm9
 914         movaps  -0x98(%rax),%xmm10
 915         movaps  -0x88(%rax),%xmm11
 916         movaps  -0x78(%rax),%xmm12
 917         movaps  -0x68(%rax),%xmm13
 918         movaps  -0x58(%rax),%xmm14
 919         movaps  -0x48(%rax),%xmm15
 920 ___
 921 $code.=<<___;
 922         mov     -48(%rax),%r15
 923 .cfi_restore    %r15
 924         mov     -40(%rax),%r14
 925 .cfi_restore    %r14
 926         mov     -32(%rax),%r13
 927 .cfi_restore    %r13
 928         mov     -24(%rax),%r12
 929 .cfi_restore    %r12
 930         mov     -16(%rax),%rbp
 931 .cfi_restore    %rbp
 932         mov     -8(%rax),%rbx
 933 .cfi_restore    %rbx
 934         lea     (%rax),%rsp             # restore %rsp
 935 .cfi_def_cfa_register   %rsp
 936 .Lgcm_enc_abort:
 937         mov     $ret,%rax               # return value
 938         ret
 939 .cfi_endproc
 940 .size   aesni_gcm_encrypt,.-aesni_gcm_encrypt
 941 ___
 942
 943 $code.=<<___;
 944 .align  64
 945 .Lbswap_mask:
 946         .byte   15,14,13,12,11,10,9,8,7,6,5,4,3,2,1,0
 947 .Lpoly:
 948         .byte   0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0xc2
 949 .Lone_msb:
 950         .byte   0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1
 951 .Ltwo_lsb:
 952         .byte   2,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0
 953 .Lone_lsb:
 954         .byte   1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0
 955 .asciz  "AES-NI GCM module for x86_64, CRYPTOGAMS by <appro\@openssl.org>"
 956 .align  64
 957 ___
 958 if ($win64) {
 959 $rec="%rcx";
 960 $frame="%rdx";
 961 $context="%r8";
 962 $disp="%r9";
 963
 964 $code.=<<___
 965 .extern __imp_RtlVirtualUnwind
 966 .type   gcm_se_handler,\@abi-omnipotent
 967 .align  16
 968 gcm_se_handler:
 969         push    %rsi
 970         push    %rdi
 971         push    %rbx
 972         push    %rbp
 973         push    %r12
 974         push    %r13
 975         push    %r14
 976         push    %r15
 977         pushfq
 978         sub     \$64,%rsp
 979
 980         mov     120($context),%rax      # pull context->Rax
 981         mov     248($context),%rbx      # pull context->Rip
 982
 983         mov     8($disp),%rsi           # disp->ImageBase
 984         mov     56($disp),%r11          # disp->HandlerData
 985
 986         mov     0(%r11),%r10d           # HandlerData[0]
 987         lea     (%rsi,%r10),%r10        # prologue label
 988         cmp     %r10,%rbx               # context->Rip<prologue label
 989         jb      .Lcommon_seh_tail
 990
 991         mov     152($context),%rax      # pull context->Rsp
 992
 993         mov     4(%r11),%r10d           # HandlerData[1]
 994         lea     (%rsi,%r10),%r10        # epilogue label
 995         cmp     %r10,%rbx               # context->Rip>=epilogue label
 996         jae     .Lcommon_seh_tail
 997
 998         mov     120($context),%rax      # pull context->Rax
 999
1000         mov     -48(%rax),%r15
1001         mov     -40(%rax),%r14
1002         mov     -32(%rax),%r13
1003         mov     -24(%rax),%r12
1004         mov     -16(%rax),%rbp
1005         mov     -8(%rax),%rbx
1006         mov     %r15,240($context)
1007         mov     %r14,232($context)
1008         mov     %r13,224($context)
1009         mov     %r12,216($context)
1010         mov     %rbp,160($context)
1011         mov     %rbx,144($context)
1012
1013         lea     -0xd8(%rax),%rsi        # %xmm save area
1014         lea     512($context),%rdi      # & context.Xmm6
1015         mov     \$20,%ecx               # 10*sizeof(%xmm0)/sizeof(%rax)
1016         .long   0xa548f3fc              # cld; rep movsq
1017
1018 .Lcommon_seh_tail:
1019         mov     8(%rax),%rdi
1020         mov     16(%rax),%rsi
1021         mov     %rax,152($context)      # restore context->Rsp
1022         mov     %rsi,168($context)      # restore context->Rsi
1023         mov     %rdi,176($context)      # restore context->Rdi
1024
1025         mov     40($disp),%rdi          # disp->ContextRecord
1026         mov     $context,%rsi           # context
1027         mov     \$154,%ecx              # sizeof(CONTEXT)
1028         .long   0xa548f3fc              # cld; rep movsq
1029
1030         mov     $disp,%rsi
1031         xor     %rcx,%rcx               # arg1, UNW_FLAG_NHANDLER
1032         mov     8(%rsi),%rdx            # arg2, disp->ImageBase
1033         mov     0(%rsi),%r8             # arg3, disp->ControlPc
1034         mov     16(%rsi),%r9            # arg4, disp->FunctionEntry
1035         mov     40(%rsi),%r10           # disp->ContextRecord
1036         lea     56(%rsi),%r11           # &disp->HandlerData
1037         lea     24(%rsi),%r12           # &disp->EstablisherFrame
1038         mov     %r10,32(%rsp)           # arg5
1039         mov     %r11,40(%rsp)           # arg6
1040         mov     %r12,48(%rsp)           # arg7
1041         mov     %rcx,56(%rsp)           # arg8, (NULL)
1042         call    *__imp_RtlVirtualUnwind(%rip)
1043
1044         mov     \$1,%eax                # ExceptionContinueSearch
1045         add     \$64,%rsp
1046         popfq
1047         pop     %r15
1048         pop     %r14
1049         pop     %r13
1050         pop     %r12
1051         pop     %rbp
1052         pop     %rbx
1053         pop     %rdi
1054         pop     %rsi
1055         ret
1056 .size   gcm_se_handler,.-gcm_se_handler
1057
1058 .section        .pdata
1059 .align  4
1060         .rva    .LSEH_begin_aesni_gcm_decrypt
1061         .rva    .LSEH_end_aesni_gcm_decrypt
1062         .rva    .LSEH_gcm_dec_info
1063
1064         .rva    .LSEH_begin_aesni_gcm_encrypt
1065         .rva    .LSEH_end_aesni_gcm_encrypt
1066         .rva    .LSEH_gcm_enc_info
1067 .section        .xdata
1068 .align  8
1069 .LSEH_gcm_dec_info:
1070         .byte   9,0,0,0
1071         .rva    gcm_se_handler
1072         .rva    .Lgcm_dec_body,.Lgcm_dec_abort
1073 .LSEH_gcm_enc_info:
1074         .byte   9,0,0,0
1075         .rva    gcm_se_handler
1076         .rva    .Lgcm_enc_body,.Lgcm_enc_abort
1077 ___
1078 }
1079 }}} else {{{
1080 $code=<<___;    # assembler is too old
1081 .text
1082
1083 .globl  aesni_gcm_encrypt
1084 .type   aesni_gcm_encrypt,\@abi-omnipotent
1085 aesni_gcm_encrypt:
1086 .cfi_startproc
1087         xor     %eax,%eax
1088         ret
1089 .cfi_endproc
1090 .size   aesni_gcm_encrypt,.-aesni_gcm_encrypt
1091
1092 .globl  aesni_gcm_decrypt
1093 .type   aesni_gcm_decrypt,\@abi-omnipotent
1094 aesni_gcm_decrypt:
1095 .cfi_startproc
1096         xor     %eax,%eax
1097         ret
1098 .cfi_endproc
1099 .size   aesni_gcm_decrypt,.-aesni_gcm_decrypt
1100 ___
1101 }}}
1102
1103 $code =~ s/\`([^\`]*)\`/eval($1)/gem;
1104
1105 print $code;
1106
1107 close STDOUT;