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 # [1] http://rt.openssl.org/Ticket/Display.html?id=2900&user=guest&pass=guest
  39 # [2] http://www.intel.com/content/dam/www/public/us/en/documents/software-support/enabling-high-performance-gcm.pdf
  40
  41 $flavour = shift;
  42 $output  = shift;
  43 if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
  44
  45 $win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);
  46
  47 $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
  48 ( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
  49 ( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or
  50 die "can't locate x86_64-xlate.pl";
  51
  52 if (`$ENV{CC} -Wa,-v -c -o /dev/null -x assembler /dev/null 2>&1`
  53                 =~ /GNU assembler version ([2-9]\.[0-9]+)/) {
  54         $avx = ($1>=2.20) + ($1>=2.22);
  55 }
  56
  57 if (!$avx && $win64 && ($flavour =~ /nasm/ || $ENV{ASM} =~ /nasm/) &&
  58             `nasm -v 2>&1` =~ /NASM version ([2-9]\.[0-9]+)/) {
  59         $avx = ($1>=2.09) + ($1>=2.10);
  60 }
  61
  62 if (!$avx && $win64 && ($flavour =~ /masm/ || $ENV{ASM} =~ /ml64/) &&
  63             `ml64 2>&1` =~ /Version ([0-9]+)\./) {
  64         $avx = ($1>=10) + ($1>=11);
  65 }
  66
  67 if (!$avx && `$ENV{CC} -v 2>&1` =~ /((?:^clang|LLVM) version|.*based on LLVM) ([3-9]\.[0-9]+)/) {
  68         $avx = ($2>=3.0) + ($2>3.0);
  69 }
  70
  71 open OUT,"| \"$^X\" \"$xlate\" $flavour \"$output\"";
  72 *STDOUT=*OUT;
  73
  74 if ($avx>1) {{{
  75
  76 ($inp,$out,$len,$key,$ivp,$Xip)=("%rdi","%rsi","%rdx","%rcx","%r8","%r9");
  77
  78 ($Ii,$T1,$T2,$Hkey,
  79  $Z0,$Z1,$Z2,$Z3,$Xi) = map("%xmm$_",(0..8));
  80
  81 ($inout0,$inout1,$inout2,$inout3,$inout4,$inout5,$rndkey) = map("%xmm$_",(9..15));
  82
  83 ($counter,$rounds,$ret,$const,$in0,$end0)=("%ebx","%ebp","%r10","%r11","%r14","%r15");
  84
  85 $code=<<___;
  86 .text
  87
  88 .type   _aesni_ctr32_ghash_6x,\@abi-omnipotent
  89 .align  32
  90 _aesni_ctr32_ghash_6x:
  91         vmovdqu         0x20($const),$T2        # borrow $T2, .Lone_msb
  92         sub             \$6,$len
  93         vpxor           $Z0,$Z0,$Z0             # $Z0   = 0
  94         vmovdqu         0x00-0x80($key),$rndkey
  95         vpaddb          $T2,$T1,$inout1
  96         vpaddb          $T2,$inout1,$inout2
  97         vpaddb          $T2,$inout2,$inout3
  98         vpaddb          $T2,$inout3,$inout4
  99         vpaddb          $T2,$inout4,$inout5
 100         vpxor           $rndkey,$T1,$inout0
 101         vmovdqu         $Z0,16+8(%rsp)          # "$Z3" = 0
 102         jmp             .Loop6x
 103
 104 .align  32
 105 .Loop6x:
 106         add             \$`6<<24`,$counter
 107         jc              .Lhandle_ctr32          # discard $inout[1-5]?
 108         vmovdqu         0x00-0x20($Xip),$Hkey   # $Hkey^1
 109           vpaddb        $T2,$inout5,$T1         # next counter value
 110           vpxor         $rndkey,$inout1,$inout1
 111           vpxor         $rndkey,$inout2,$inout2
 112
 113 .Lresume_ctr32:
 114         vmovdqu         $T1,($ivp)              # save next counter value
 115         vpclmulqdq      \$0x10,$Hkey,$Z3,$Z1
 116           vpxor         $rndkey,$inout3,$inout3
 117           vmovups       0x10-0x80($key),$T2     # borrow $T2 for $rndkey
 118         vpclmulqdq      \$0x01,$Hkey,$Z3,$Z2
 119
 120         # At this point, the current block of 96 (0x60) bytes has already been
 121         # loaded into registers. Concurrently with processing it, we want to
 122         # load the next 96 bytes of input for the next round. Obviously, we can
 123         # only do this if there are at least 96 more bytes of input beyond the
 124         # input we're currently processing, or else we'd read past the end of
 125         # the input buffer. Here, we set |%r12| to 96 if there are at least 96
 126         # bytes of input beyond the 96 bytes we're already processing, and we
 127         # set |%r12| to 0 otherwise. In the case where we set |%r12| to 96,
 128         # we'll read in the next block so that it is in registers for the next
 129         # loop iteration. In the case where we set |%r12| to 0, we'll re-read
 130         # the current block and then ignore what we re-read.
 131         #
 132         # At this point, |$in0| points to the current (already read into
 133         # registers) block, and |$end0| points to 2*96 bytes before the end of
 134         # the input. Thus, |$in0| > |$end0| means that we do not have the next
 135         # 96-byte block to read in, and |$in0| <= |$end0| means we do.
 136         xor             %r12,%r12
 137         cmp             $in0,$end0
 138
 139           vaesenc       $T2,$inout0,$inout0
 140         vmovdqu         0x30+8(%rsp),$Ii        # I[4]
 141           vpxor         $rndkey,$inout4,$inout4
 142         vpclmulqdq      \$0x00,$Hkey,$Z3,$T1
 143           vaesenc       $T2,$inout1,$inout1
 144           vpxor         $rndkey,$inout5,$inout5
 145         setnc           %r12b
 146         vpclmulqdq      \$0x11,$Hkey,$Z3,$Z3
 147           vaesenc       $T2,$inout2,$inout2
 148         vmovdqu         0x10-0x20($Xip),$Hkey   # $Hkey^2
 149         neg             %r12
 150           vaesenc       $T2,$inout3,$inout3
 151          vpxor          $Z1,$Z2,$Z2
 152         vpclmulqdq      \$0x00,$Hkey,$Ii,$Z1
 153          vpxor          $Z0,$Xi,$Xi             # modulo-scheduled
 154           vaesenc       $T2,$inout4,$inout4
 155          vpxor          $Z1,$T1,$Z0
 156         and             \$0x60,%r12
 157           vmovups       0x20-0x80($key),$rndkey
 158         vpclmulqdq      \$0x10,$Hkey,$Ii,$T1
 159           vaesenc       $T2,$inout5,$inout5
 160
 161         vpclmulqdq      \$0x01,$Hkey,$Ii,$T2
 162         lea             ($in0,%r12),$in0
 163           vaesenc       $rndkey,$inout0,$inout0
 164          vpxor          16+8(%rsp),$Xi,$Xi      # modulo-scheduled [vpxor $Z3,$Xi,$Xi]
 165         vpclmulqdq      \$0x11,$Hkey,$Ii,$Hkey
 166          vmovdqu        0x40+8(%rsp),$Ii        # I[3]
 167           vaesenc       $rndkey,$inout1,$inout1
 168         movbe           0x58($in0),%r13
 169           vaesenc       $rndkey,$inout2,$inout2
 170         movbe           0x50($in0),%r12
 171           vaesenc       $rndkey,$inout3,$inout3
 172         mov             %r13,0x20+8(%rsp)
 173           vaesenc       $rndkey,$inout4,$inout4
 174         mov             %r12,0x28+8(%rsp)
 175         vmovdqu         0x30-0x20($Xip),$Z1     # borrow $Z1 for $Hkey^3
 176           vaesenc       $rndkey,$inout5,$inout5
 177
 178           vmovups       0x30-0x80($key),$rndkey
 179          vpxor          $T1,$Z2,$Z2
 180         vpclmulqdq      \$0x00,$Z1,$Ii,$T1
 181           vaesenc       $rndkey,$inout0,$inout0
 182          vpxor          $T2,$Z2,$Z2
 183         vpclmulqdq      \$0x10,$Z1,$Ii,$T2
 184           vaesenc       $rndkey,$inout1,$inout1
 185          vpxor          $Hkey,$Z3,$Z3
 186         vpclmulqdq      \$0x01,$Z1,$Ii,$Hkey
 187           vaesenc       $rndkey,$inout2,$inout2
 188         vpclmulqdq      \$0x11,$Z1,$Ii,$Z1
 189          vmovdqu        0x50+8(%rsp),$Ii        # I[2]
 190           vaesenc       $rndkey,$inout3,$inout3
 191           vaesenc       $rndkey,$inout4,$inout4
 192          vpxor          $T1,$Z0,$Z0
 193         vmovdqu         0x40-0x20($Xip),$T1     # borrow $T1 for $Hkey^4
 194           vaesenc       $rndkey,$inout5,$inout5
 195
 196           vmovups       0x40-0x80($key),$rndkey
 197          vpxor          $T2,$Z2,$Z2
 198         vpclmulqdq      \$0x00,$T1,$Ii,$T2
 199           vaesenc       $rndkey,$inout0,$inout0
 200          vpxor          $Hkey,$Z2,$Z2
 201         vpclmulqdq      \$0x10,$T1,$Ii,$Hkey
 202           vaesenc       $rndkey,$inout1,$inout1
 203         movbe           0x48($in0),%r13
 204          vpxor          $Z1,$Z3,$Z3
 205         vpclmulqdq      \$0x01,$T1,$Ii,$Z1
 206           vaesenc       $rndkey,$inout2,$inout2
 207         movbe           0x40($in0),%r12
 208         vpclmulqdq      \$0x11,$T1,$Ii,$T1
 209          vmovdqu        0x60+8(%rsp),$Ii        # I[1]
 210           vaesenc       $rndkey,$inout3,$inout3
 211         mov             %r13,0x30+8(%rsp)
 212           vaesenc       $rndkey,$inout4,$inout4
 213         mov             %r12,0x38+8(%rsp)
 214          vpxor          $T2,$Z0,$Z0
 215         vmovdqu         0x60-0x20($Xip),$T2     # borrow $T2 for $Hkey^5
 216           vaesenc       $rndkey,$inout5,$inout5
 217
 218           vmovups       0x50-0x80($key),$rndkey
 219          vpxor          $Hkey,$Z2,$Z2
 220         vpclmulqdq      \$0x00,$T2,$Ii,$Hkey
 221           vaesenc       $rndkey,$inout0,$inout0
 222          vpxor          $Z1,$Z2,$Z2
 223         vpclmulqdq      \$0x10,$T2,$Ii,$Z1
 224           vaesenc       $rndkey,$inout1,$inout1
 225         movbe           0x38($in0),%r13
 226          vpxor          $T1,$Z3,$Z3
 227         vpclmulqdq      \$0x01,$T2,$Ii,$T1
 228          vpxor          0x70+8(%rsp),$Xi,$Xi    # accumulate I[0]
 229           vaesenc       $rndkey,$inout2,$inout2
 230         movbe           0x30($in0),%r12
 231         vpclmulqdq      \$0x11,$T2,$Ii,$T2
 232           vaesenc       $rndkey,$inout3,$inout3
 233         mov             %r13,0x40+8(%rsp)
 234           vaesenc       $rndkey,$inout4,$inout4
 235         mov             %r12,0x48+8(%rsp)
 236          vpxor          $Hkey,$Z0,$Z0
 237          vmovdqu        0x70-0x20($Xip),$Hkey   # $Hkey^6
 238           vaesenc       $rndkey,$inout5,$inout5
 239
 240           vmovups       0x60-0x80($key),$rndkey
 241          vpxor          $Z1,$Z2,$Z2
 242         vpclmulqdq      \$0x10,$Hkey,$Xi,$Z1
 243           vaesenc       $rndkey,$inout0,$inout0
 244          vpxor          $T1,$Z2,$Z2
 245         vpclmulqdq      \$0x01,$Hkey,$Xi,$T1
 246           vaesenc       $rndkey,$inout1,$inout1
 247         movbe           0x28($in0),%r13
 248          vpxor          $T2,$Z3,$Z3
 249         vpclmulqdq      \$0x00,$Hkey,$Xi,$T2
 250           vaesenc       $rndkey,$inout2,$inout2
 251         movbe           0x20($in0),%r12
 252         vpclmulqdq      \$0x11,$Hkey,$Xi,$Xi
 253           vaesenc       $rndkey,$inout3,$inout3
 254         mov             %r13,0x50+8(%rsp)
 255           vaesenc       $rndkey,$inout4,$inout4
 256         mov             %r12,0x58+8(%rsp)
 257         vpxor           $Z1,$Z2,$Z2
 258           vaesenc       $rndkey,$inout5,$inout5
 259         vpxor           $T1,$Z2,$Z2
 260
 261           vmovups       0x70-0x80($key),$rndkey
 262         vpslldq         \$8,$Z2,$Z1
 263         vpxor           $T2,$Z0,$Z0
 264         vmovdqu         0x10($const),$Hkey      # .Lpoly
 265
 266           vaesenc       $rndkey,$inout0,$inout0
 267         vpxor           $Xi,$Z3,$Z3
 268           vaesenc       $rndkey,$inout1,$inout1
 269         vpxor           $Z1,$Z0,$Z0
 270         movbe           0x18($in0),%r13
 271           vaesenc       $rndkey,$inout2,$inout2
 272         movbe           0x10($in0),%r12
 273         vpalignr        \$8,$Z0,$Z0,$Ii         # 1st phase
 274         vpclmulqdq      \$0x10,$Hkey,$Z0,$Z0
 275         mov             %r13,0x60+8(%rsp)
 276           vaesenc       $rndkey,$inout3,$inout3
 277         mov             %r12,0x68+8(%rsp)
 278           vaesenc       $rndkey,$inout4,$inout4
 279           vmovups       0x80-0x80($key),$T1     # borrow $T1 for $rndkey
 280           vaesenc       $rndkey,$inout5,$inout5
 281
 282           vaesenc       $T1,$inout0,$inout0
 283           vmovups       0x90-0x80($key),$rndkey
 284           vaesenc       $T1,$inout1,$inout1
 285         vpsrldq         \$8,$Z2,$Z2
 286           vaesenc       $T1,$inout2,$inout2
 287         vpxor           $Z2,$Z3,$Z3
 288           vaesenc       $T1,$inout3,$inout3
 289         vpxor           $Ii,$Z0,$Z0
 290         movbe           0x08($in0),%r13
 291           vaesenc       $T1,$inout4,$inout4
 292         movbe           0x00($in0),%r12
 293           vaesenc       $T1,$inout5,$inout5
 294           vmovups       0xa0-0x80($key),$T1
 295           cmp           \$11,$rounds
 296           jb            .Lenc_tail              # 128-bit key
 297
 298           vaesenc       $rndkey,$inout0,$inout0
 299           vaesenc       $rndkey,$inout1,$inout1
 300           vaesenc       $rndkey,$inout2,$inout2
 301           vaesenc       $rndkey,$inout3,$inout3
 302           vaesenc       $rndkey,$inout4,$inout4
 303           vaesenc       $rndkey,$inout5,$inout5
 304
 305           vaesenc       $T1,$inout0,$inout0
 306           vaesenc       $T1,$inout1,$inout1
 307           vaesenc       $T1,$inout2,$inout2
 308           vaesenc       $T1,$inout3,$inout3
 309           vaesenc       $T1,$inout4,$inout4
 310           vmovups       0xb0-0x80($key),$rndkey
 311           vaesenc       $T1,$inout5,$inout5
 312           vmovups       0xc0-0x80($key),$T1
 313           je            .Lenc_tail              # 192-bit key
 314
 315           vaesenc       $rndkey,$inout0,$inout0
 316           vaesenc       $rndkey,$inout1,$inout1
 317           vaesenc       $rndkey,$inout2,$inout2
 318           vaesenc       $rndkey,$inout3,$inout3
 319           vaesenc       $rndkey,$inout4,$inout4
 320           vaesenc       $rndkey,$inout5,$inout5
 321
 322           vaesenc       $T1,$inout0,$inout0
 323           vaesenc       $T1,$inout1,$inout1
 324           vaesenc       $T1,$inout2,$inout2
 325           vaesenc       $T1,$inout3,$inout3
 326           vaesenc       $T1,$inout4,$inout4
 327           vmovups       0xd0-0x80($key),$rndkey
 328           vaesenc       $T1,$inout5,$inout5
 329           vmovups       0xe0-0x80($key),$T1
 330           jmp           .Lenc_tail              # 256-bit key
 331
 332 .align  32
 333 .Lhandle_ctr32:
 334         vmovdqu         ($const),$Ii            # borrow $Ii for .Lbswap_mask
 335           vpshufb       $Ii,$T1,$Z2             # byte-swap counter
 336           vmovdqu       0x30($const),$Z1        # borrow $Z1, .Ltwo_lsb
 337           vpaddd        0x40($const),$Z2,$inout1        # .Lone_lsb
 338           vpaddd        $Z1,$Z2,$inout2
 339         vmovdqu         0x00-0x20($Xip),$Hkey   # $Hkey^1
 340           vpaddd        $Z1,$inout1,$inout3
 341           vpshufb       $Ii,$inout1,$inout1
 342           vpaddd        $Z1,$inout2,$inout4
 343           vpshufb       $Ii,$inout2,$inout2
 344           vpxor         $rndkey,$inout1,$inout1
 345           vpaddd        $Z1,$inout3,$inout5
 346           vpshufb       $Ii,$inout3,$inout3
 347           vpxor         $rndkey,$inout2,$inout2
 348           vpaddd        $Z1,$inout4,$T1         # byte-swapped next counter value
 349           vpshufb       $Ii,$inout4,$inout4
 350           vpshufb       $Ii,$inout5,$inout5
 351           vpshufb       $Ii,$T1,$T1             # next counter value
 352         jmp             .Lresume_ctr32
 353
 354 .align  32
 355 .Lenc_tail:
 356           vaesenc       $rndkey,$inout0,$inout0
 357         vmovdqu         $Z3,16+8(%rsp)          # postpone vpxor $Z3,$Xi,$Xi
 358         vpalignr        \$8,$Z0,$Z0,$Xi         # 2nd phase
 359           vaesenc       $rndkey,$inout1,$inout1
 360         vpclmulqdq      \$0x10,$Hkey,$Z0,$Z0
 361           vpxor         0x00($inp),$T1,$T2
 362           vaesenc       $rndkey,$inout2,$inout2
 363           vpxor         0x10($inp),$T1,$Ii
 364           vaesenc       $rndkey,$inout3,$inout3
 365           vpxor         0x20($inp),$T1,$Z1
 366           vaesenc       $rndkey,$inout4,$inout4
 367           vpxor         0x30($inp),$T1,$Z2
 368           vaesenc       $rndkey,$inout5,$inout5
 369           vpxor         0x40($inp),$T1,$Z3
 370           vpxor         0x50($inp),$T1,$Hkey
 371           vmovdqu       ($ivp),$T1              # load next counter value
 372
 373           vaesenclast   $T2,$inout0,$inout0
 374           vmovdqu       0x20($const),$T2        # borrow $T2, .Lone_msb
 375           vaesenclast   $Ii,$inout1,$inout1
 376          vpaddb         $T2,$T1,$Ii
 377         mov             %r13,0x70+8(%rsp)
 378         lea             0x60($inp),$inp
 379           vaesenclast   $Z1,$inout2,$inout2
 380          vpaddb         $T2,$Ii,$Z1
 381         mov             %r12,0x78+8(%rsp)
 382         lea             0x60($out),$out
 383           vmovdqu       0x00-0x80($key),$rndkey
 384           vaesenclast   $Z2,$inout3,$inout3
 385          vpaddb         $T2,$Z1,$Z2
 386           vaesenclast   $Z3, $inout4,$inout4
 387          vpaddb         $T2,$Z2,$Z3
 388           vaesenclast   $Hkey,$inout5,$inout5
 389          vpaddb         $T2,$Z3,$Hkey
 390
 391         add             \$0x60,$ret
 392         sub             \$0x6,$len
 393         jc              .L6x_done
 394
 395           vmovups       $inout0,-0x60($out)     # save output
 396          vpxor          $rndkey,$T1,$inout0
 397           vmovups       $inout1,-0x50($out)
 398          vmovdqa        $Ii,$inout1             # 0 latency
 399           vmovups       $inout2,-0x40($out)
 400          vmovdqa        $Z1,$inout2             # 0 latency
 401           vmovups       $inout3,-0x30($out)
 402          vmovdqa        $Z2,$inout3             # 0 latency
 403           vmovups       $inout4,-0x20($out)
 404          vmovdqa        $Z3,$inout4             # 0 latency
 405           vmovups       $inout5,-0x10($out)
 406          vmovdqa        $Hkey,$inout5           # 0 latency
 407         vmovdqu         0x20+8(%rsp),$Z3        # I[5]
 408         jmp             .Loop6x
 409
 410 .L6x_done:
 411         vpxor           16+8(%rsp),$Xi,$Xi      # modulo-scheduled
 412         vpxor           $Z0,$Xi,$Xi             # modulo-scheduled
 413
 414         ret
 415 .size   _aesni_ctr32_ghash_6x,.-_aesni_ctr32_ghash_6x
 416 ___
 417 ######################################################################
 418 #
 419 # size_t aesni_gcm_[en|de]crypt(const void *inp, void *out, size_t len,
 420 #               const AES_KEY *key, unsigned char iv[16],
 421 #               struct { u128 Xi,H,Htbl[9]; } *Xip);
 422 $code.=<<___;
 423 .globl  aesni_gcm_decrypt
 424 .type   aesni_gcm_decrypt,\@function,6
 425 .align  32
 426 aesni_gcm_decrypt:
 427 .cfi_startproc
 428         xor     $ret,$ret
 429
 430         # We call |_aesni_ctr32_ghash_6x|, which requires at least 96 (0x60)
 431         # bytes of input.
 432         cmp     \$0x60,$len                     # minimal accepted length
 433         jb      .Lgcm_dec_abort
 434
 435         lea     (%rsp),%rax                     # save stack pointer
 436 .cfi_def_cfa_register   %rax
 437         push    %rbx
 438 .cfi_push       %rbx
 439         push    %rbp
 440 .cfi_push       %rbp
 441         push    %r12
 442 .cfi_push       %r12
 443         push    %r13
 444 .cfi_push       %r13
 445         push    %r14
 446 .cfi_push       %r14
 447         push    %r15
 448 .cfi_push       %r15
 449 ___
 450 $code.=<<___ if ($win64);
 451         lea     -0xa8(%rsp),%rsp
 452         movaps  %xmm6,-0xd8(%rax)
 453         movaps  %xmm7,-0xc8(%rax)
 454         movaps  %xmm8,-0xb8(%rax)
 455         movaps  %xmm9,-0xa8(%rax)
 456         movaps  %xmm10,-0x98(%rax)
 457         movaps  %xmm11,-0x88(%rax)
 458         movaps  %xmm12,-0x78(%rax)
 459         movaps  %xmm13,-0x68(%rax)
 460         movaps  %xmm14,-0x58(%rax)
 461         movaps  %xmm15,-0x48(%rax)
 462 .Lgcm_dec_body:
 463 ___
 464 $code.=<<___;
 465         vzeroupper
 466
 467         vmovdqu         ($ivp),$T1              # input counter value
 468         add             \$-128,%rsp
 469         mov             12($ivp),$counter
 470         lea             .Lbswap_mask(%rip),$const
 471         lea             -0x80($key),$in0        # borrow $in0
 472         mov             \$0xf80,$end0           # borrow $end0
 473         vmovdqu         ($Xip),$Xi              # load Xi
 474         and             \$-128,%rsp             # ensure stack alignment
 475         vmovdqu         ($const),$Ii            # borrow $Ii for .Lbswap_mask
 476         lea             0x80($key),$key         # size optimization
 477         lea             0x20+0x20($Xip),$Xip    # size optimization
 478         mov             0xf0-0x80($key),$rounds
 479         vpshufb         $Ii,$Xi,$Xi
 480
 481         and             $end0,$in0
 482         and             %rsp,$end0
 483         sub             $in0,$end0
 484         jc              .Ldec_no_key_aliasing
 485         cmp             \$768,$end0
 486         jnc             .Ldec_no_key_aliasing
 487         sub             $end0,%rsp              # avoid aliasing with key
 488 .Ldec_no_key_aliasing:
 489
 490         vmovdqu         0x50($inp),$Z3          # I[5]
 491         lea             ($inp),$in0
 492         vmovdqu         0x40($inp),$Z0
 493
 494         # |_aesni_ctr32_ghash_6x| requires |$end0| to point to 2*96 (0xc0)
 495         # bytes before the end of the input. Note, in particular, that this is
 496         # correct even if |$len| is not an even multiple of 96 or 16. XXX: This
 497         # seems to require that |$inp| + |$len| >= 2*96 (0xc0); i.e. |$inp| must
 498         # not be near the very beginning of the address space when |$len| < 2*96
 499         # (0xc0).
 500         lea             -0xc0($inp,$len),$end0
 501
 502         vmovdqu         0x30($inp),$Z1
 503         shr             \$4,$len
 504         xor             $ret,$ret
 505         vmovdqu         0x20($inp),$Z2
 506          vpshufb        $Ii,$Z3,$Z3             # passed to _aesni_ctr32_ghash_6x
 507         vmovdqu         0x10($inp),$T2
 508          vpshufb        $Ii,$Z0,$Z0
 509         vmovdqu         ($inp),$Hkey
 510          vpshufb        $Ii,$Z1,$Z1
 511         vmovdqu         $Z0,0x30(%rsp)
 512          vpshufb        $Ii,$Z2,$Z2
 513         vmovdqu         $Z1,0x40(%rsp)
 514          vpshufb        $Ii,$T2,$T2
 515         vmovdqu         $Z2,0x50(%rsp)
 516          vpshufb        $Ii,$Hkey,$Hkey
 517         vmovdqu         $T2,0x60(%rsp)
 518         vmovdqu         $Hkey,0x70(%rsp)
 519
 520         call            _aesni_ctr32_ghash_6x
 521
 522         vmovups         $inout0,-0x60($out)     # save output
 523         vmovups         $inout1,-0x50($out)
 524         vmovups         $inout2,-0x40($out)
 525         vmovups         $inout3,-0x30($out)
 526         vmovups         $inout4,-0x20($out)
 527         vmovups         $inout5,-0x10($out)
 528
 529         vpshufb         ($const),$Xi,$Xi        # .Lbswap_mask
 530         vmovdqu         $Xi,-0x40($Xip)         # output Xi
 531
 532         vzeroupper
 533 ___
 534 $code.=<<___ if ($win64);
 535         movaps  -0xd8(%rax),%xmm6
 536         movaps  -0xc8(%rax),%xmm7
 537         movaps  -0xb8(%rax),%xmm8
 538         movaps  -0xa8(%rax),%xmm9
 539         movaps  -0x98(%rax),%xmm10
 540         movaps  -0x88(%rax),%xmm11
 541         movaps  -0x78(%rax),%xmm12
 542         movaps  -0x68(%rax),%xmm13
 543         movaps  -0x58(%rax),%xmm14
 544         movaps  -0x48(%rax),%xmm15
 545 ___
 546 $code.=<<___;
 547         mov     -48(%rax),%r15
 548 .cfi_restore    %r15
 549         mov     -40(%rax),%r14
 550 .cfi_restore    %r14
 551         mov     -32(%rax),%r13
 552 .cfi_restore    %r13
 553         mov     -24(%rax),%r12
 554 .cfi_restore    %r12
 555         mov     -16(%rax),%rbp
 556 .cfi_restore    %rbp
 557         mov     -8(%rax),%rbx
 558 .cfi_restore    %rbx
 559         lea     (%rax),%rsp             # restore %rsp
 560 .cfi_def_cfa_register   %rsp
 561 .Lgcm_dec_abort:
 562         mov     $ret,%rax               # return value
 563         ret
 564 .cfi_endproc
 565 .size   aesni_gcm_decrypt,.-aesni_gcm_decrypt
 566 ___
 567
 568 $code.=<<___;
 569 .type   _aesni_ctr32_6x,\@abi-omnipotent
 570 .align  32
 571 _aesni_ctr32_6x:
 572         vmovdqu         0x00-0x80($key),$Z0     # borrow $Z0 for $rndkey
 573         vmovdqu         0x20($const),$T2        # borrow $T2, .Lone_msb
 574         lea             -1($rounds),%r13
 575         vmovups         0x10-0x80($key),$rndkey
 576         lea             0x20-0x80($key),%r12
 577         vpxor           $Z0,$T1,$inout0
 578         add             \$`6<<24`,$counter
 579         jc              .Lhandle_ctr32_2
 580         vpaddb          $T2,$T1,$inout1
 581         vpaddb          $T2,$inout1,$inout2
 582         vpxor           $Z0,$inout1,$inout1
 583         vpaddb          $T2,$inout2,$inout3
 584         vpxor           $Z0,$inout2,$inout2
 585         vpaddb          $T2,$inout3,$inout4
 586         vpxor           $Z0,$inout3,$inout3
 587         vpaddb          $T2,$inout4,$inout5
 588         vpxor           $Z0,$inout4,$inout4
 589         vpaddb          $T2,$inout5,$T1
 590         vpxor           $Z0,$inout5,$inout5
 591         jmp             .Loop_ctr32
 592
 593 .align  16
 594 .Loop_ctr32:
 595         vaesenc         $rndkey,$inout0,$inout0
 596         vaesenc         $rndkey,$inout1,$inout1
 597         vaesenc         $rndkey,$inout2,$inout2
 598         vaesenc         $rndkey,$inout3,$inout3
 599         vaesenc         $rndkey,$inout4,$inout4
 600         vaesenc         $rndkey,$inout5,$inout5
 601         vmovups         (%r12),$rndkey
 602         lea             0x10(%r12),%r12
 603         dec             %r13d
 604         jnz             .Loop_ctr32
 605
 606         vmovdqu         (%r12),$Hkey            # last round key
 607         vaesenc         $rndkey,$inout0,$inout0
 608         vpxor           0x00($inp),$Hkey,$Z0
 609         vaesenc         $rndkey,$inout1,$inout1
 610         vpxor           0x10($inp),$Hkey,$Z1
 611         vaesenc         $rndkey,$inout2,$inout2
 612         vpxor           0x20($inp),$Hkey,$Z2
 613         vaesenc         $rndkey,$inout3,$inout3
 614         vpxor           0x30($inp),$Hkey,$Xi
 615         vaesenc         $rndkey,$inout4,$inout4
 616         vpxor           0x40($inp),$Hkey,$T2
 617         vaesenc         $rndkey,$inout5,$inout5
 618         vpxor           0x50($inp),$Hkey,$Hkey
 619         lea             0x60($inp),$inp
 620
 621         vaesenclast     $Z0,$inout0,$inout0
 622         vaesenclast     $Z1,$inout1,$inout1
 623         vaesenclast     $Z2,$inout2,$inout2
 624         vaesenclast     $Xi,$inout3,$inout3
 625         vaesenclast     $T2,$inout4,$inout4
 626         vaesenclast     $Hkey,$inout5,$inout5
 627         vmovups         $inout0,0x00($out)
 628         vmovups         $inout1,0x10($out)
 629         vmovups         $inout2,0x20($out)
 630         vmovups         $inout3,0x30($out)
 631         vmovups         $inout4,0x40($out)
 632         vmovups         $inout5,0x50($out)
 633         lea             0x60($out),$out
 634
 635         ret
 636 .align  32
 637 .Lhandle_ctr32_2:
 638         vpshufb         $Ii,$T1,$Z2             # byte-swap counter
 639         vmovdqu         0x30($const),$Z1        # borrow $Z1, .Ltwo_lsb
 640         vpaddd          0x40($const),$Z2,$inout1        # .Lone_lsb
 641         vpaddd          $Z1,$Z2,$inout2
 642         vpaddd          $Z1,$inout1,$inout3
 643         vpshufb         $Ii,$inout1,$inout1
 644         vpaddd          $Z1,$inout2,$inout4
 645         vpshufb         $Ii,$inout2,$inout2
 646         vpxor           $Z0,$inout1,$inout1
 647         vpaddd          $Z1,$inout3,$inout5
 648         vpshufb         $Ii,$inout3,$inout3
 649         vpxor           $Z0,$inout2,$inout2
 650         vpaddd          $Z1,$inout4,$T1         # byte-swapped next counter value
 651         vpshufb         $Ii,$inout4,$inout4
 652         vpxor           $Z0,$inout3,$inout3
 653         vpshufb         $Ii,$inout5,$inout5
 654         vpxor           $Z0,$inout4,$inout4
 655         vpshufb         $Ii,$T1,$T1             # next counter value
 656         vpxor           $Z0,$inout5,$inout5
 657         jmp     .Loop_ctr32
 658 .size   _aesni_ctr32_6x,.-_aesni_ctr32_6x
 659
 660 .globl  aesni_gcm_encrypt
 661 .type   aesni_gcm_encrypt,\@function,6
 662 .align  32
 663 aesni_gcm_encrypt:
 664 .cfi_startproc
 665         xor     $ret,$ret
 666
 667         # We call |_aesni_ctr32_6x| twice, each call consuming 96 bytes of
 668         # input. Then we call |_aesni_ctr32_ghash_6x|, which requires at
 669         # least 96 more bytes of input.
 670         cmp     \$0x60*3,$len                   # minimal accepted length
 671         jb      .Lgcm_enc_abort
 672
 673         lea     (%rsp),%rax                     # save stack pointer
 674 .cfi_def_cfa_register   %rax
 675         push    %rbx
 676 .cfi_push       %rbx
 677         push    %rbp
 678 .cfi_push       %rbp
 679         push    %r12
 680 .cfi_push       %r12
 681         push    %r13
 682 .cfi_push       %r13
 683         push    %r14
 684 .cfi_push       %r14
 685         push    %r15
 686 .cfi_push       %r15
 687 ___
 688 $code.=<<___ if ($win64);
 689         lea     -0xa8(%rsp),%rsp
 690         movaps  %xmm6,-0xd8(%rax)
 691         movaps  %xmm7,-0xc8(%rax)
 692         movaps  %xmm8,-0xb8(%rax)
 693         movaps  %xmm9,-0xa8(%rax)
 694         movaps  %xmm10,-0x98(%rax)
 695         movaps  %xmm11,-0x88(%rax)
 696         movaps  %xmm12,-0x78(%rax)
 697         movaps  %xmm13,-0x68(%rax)
 698         movaps  %xmm14,-0x58(%rax)
 699         movaps  %xmm15,-0x48(%rax)
 700 .Lgcm_enc_body:
 701 ___
 702 $code.=<<___;
 703         vzeroupper
 704
 705         vmovdqu         ($ivp),$T1              # input counter value
 706         add             \$-128,%rsp
 707         mov             12($ivp),$counter
 708         lea             .Lbswap_mask(%rip),$const
 709         lea             -0x80($key),$in0        # borrow $in0
 710         mov             \$0xf80,$end0           # borrow $end0
 711         lea             0x80($key),$key         # size optimization
 712         vmovdqu         ($const),$Ii            # borrow $Ii for .Lbswap_mask
 713         and             \$-128,%rsp             # ensure stack alignment
 714         mov             0xf0-0x80($key),$rounds
 715
 716         and             $end0,$in0
 717         and             %rsp,$end0
 718         sub             $in0,$end0
 719         jc              .Lenc_no_key_aliasing
 720         cmp             \$768,$end0
 721         jnc             .Lenc_no_key_aliasing
 722         sub             $end0,%rsp              # avoid aliasing with key
 723 .Lenc_no_key_aliasing:
 724
 725         lea             ($out),$in0
 726
 727         # |_aesni_ctr32_ghash_6x| requires |$end0| to point to 2*96 (0xc0)
 728         # bytes before the end of the input. Note, in particular, that this is
 729         # correct even if |$len| is not an even multiple of 96 or 16. Unlike in
 730         # the decryption case, there's no caveat that |$out| must not be near
 731         # the very beginning of the address space, because we know that
 732         # |$len| >= 3*96 from the check above, and so we know
 733         # |$out| + |$len| >= 2*96 (0xc0).
 734         lea             -0xc0($out,$len),$end0
 735
 736         shr             \$4,$len
 737
 738         call            _aesni_ctr32_6x
 739         vpshufb         $Ii,$inout0,$Xi         # save bswapped output on stack
 740         vpshufb         $Ii,$inout1,$T2
 741         vmovdqu         $Xi,0x70(%rsp)
 742         vpshufb         $Ii,$inout2,$Z0
 743         vmovdqu         $T2,0x60(%rsp)
 744         vpshufb         $Ii,$inout3,$Z1
 745         vmovdqu         $Z0,0x50(%rsp)
 746         vpshufb         $Ii,$inout4,$Z2
 747         vmovdqu         $Z1,0x40(%rsp)
 748         vpshufb         $Ii,$inout5,$Z3         # passed to _aesni_ctr32_ghash_6x
 749         vmovdqu         $Z2,0x30(%rsp)
 750
 751         call            _aesni_ctr32_6x
 752
 753         vmovdqu         ($Xip),$Xi              # load Xi
 754         lea             0x20+0x20($Xip),$Xip    # size optimization
 755         sub             \$12,$len
 756         mov             \$0x60*2,$ret
 757         vpshufb         $Ii,$Xi,$Xi
 758
 759         call            _aesni_ctr32_ghash_6x
 760         vmovdqu         0x20(%rsp),$Z3          # I[5]
 761          vmovdqu        ($const),$Ii            # borrow $Ii for .Lbswap_mask
 762         vmovdqu         0x00-0x20($Xip),$Hkey   # $Hkey^1
 763         vpunpckhqdq     $Z3,$Z3,$T1
 764         vmovdqu         0x20-0x20($Xip),$rndkey # borrow $rndkey for $HK
 765          vmovups        $inout0,-0x60($out)     # save output
 766          vpshufb        $Ii,$inout0,$inout0     # but keep bswapped copy
 767         vpxor           $Z3,$T1,$T1
 768          vmovups        $inout1,-0x50($out)
 769          vpshufb        $Ii,$inout1,$inout1
 770          vmovups        $inout2,-0x40($out)
 771          vpshufb        $Ii,$inout2,$inout2
 772          vmovups        $inout3,-0x30($out)
 773          vpshufb        $Ii,$inout3,$inout3
 774          vmovups        $inout4,-0x20($out)
 775          vpshufb        $Ii,$inout4,$inout4
 776          vmovups        $inout5,-0x10($out)
 777          vpshufb        $Ii,$inout5,$inout5
 778          vmovdqu        $inout0,0x10(%rsp)      # free $inout0
 779 ___
 780 { my ($HK,$T3)=($rndkey,$inout0);
 781
 782 $code.=<<___;
 783          vmovdqu        0x30(%rsp),$Z2          # I[4]
 784          vmovdqu        0x10-0x20($Xip),$Ii     # borrow $Ii for $Hkey^2
 785          vpunpckhqdq    $Z2,$Z2,$T2
 786         vpclmulqdq      \$0x00,$Hkey,$Z3,$Z1
 787          vpxor          $Z2,$T2,$T2
 788         vpclmulqdq      \$0x11,$Hkey,$Z3,$Z3
 789         vpclmulqdq      \$0x00,$HK,$T1,$T1
 790
 791          vmovdqu        0x40(%rsp),$T3          # I[3]
 792         vpclmulqdq      \$0x00,$Ii,$Z2,$Z0
 793          vmovdqu        0x30-0x20($Xip),$Hkey   # $Hkey^3
 794         vpxor           $Z1,$Z0,$Z0
 795          vpunpckhqdq    $T3,$T3,$Z1
 796         vpclmulqdq      \$0x11,$Ii,$Z2,$Z2
 797          vpxor          $T3,$Z1,$Z1
 798         vpxor           $Z3,$Z2,$Z2
 799         vpclmulqdq      \$0x10,$HK,$T2,$T2
 800          vmovdqu        0x50-0x20($Xip),$HK
 801         vpxor           $T1,$T2,$T2
 802
 803          vmovdqu        0x50(%rsp),$T1          # I[2]
 804         vpclmulqdq      \$0x00,$Hkey,$T3,$Z3
 805          vmovdqu        0x40-0x20($Xip),$Ii     # borrow $Ii for $Hkey^4
 806         vpxor           $Z0,$Z3,$Z3
 807          vpunpckhqdq    $T1,$T1,$Z0
 808         vpclmulqdq      \$0x11,$Hkey,$T3,$T3
 809          vpxor          $T1,$Z0,$Z0
 810         vpxor           $Z2,$T3,$T3
 811         vpclmulqdq      \$0x00,$HK,$Z1,$Z1
 812         vpxor           $T2,$Z1,$Z1
 813
 814          vmovdqu        0x60(%rsp),$T2          # I[1]
 815         vpclmulqdq      \$0x00,$Ii,$T1,$Z2
 816          vmovdqu        0x60-0x20($Xip),$Hkey   # $Hkey^5
 817         vpxor           $Z3,$Z2,$Z2
 818          vpunpckhqdq    $T2,$T2,$Z3
 819         vpclmulqdq      \$0x11,$Ii,$T1,$T1
 820          vpxor          $T2,$Z3,$Z3
 821         vpxor           $T3,$T1,$T1
 822         vpclmulqdq      \$0x10,$HK,$Z0,$Z0
 823          vmovdqu        0x80-0x20($Xip),$HK
 824         vpxor           $Z1,$Z0,$Z0
 825
 826          vpxor          0x70(%rsp),$Xi,$Xi      # accumulate I[0]
 827         vpclmulqdq      \$0x00,$Hkey,$T2,$Z1
 828          vmovdqu        0x70-0x20($Xip),$Ii     # borrow $Ii for $Hkey^6
 829          vpunpckhqdq    $Xi,$Xi,$T3
 830         vpxor           $Z2,$Z1,$Z1
 831         vpclmulqdq      \$0x11,$Hkey,$T2,$T2
 832          vpxor          $Xi,$T3,$T3
 833         vpxor           $T1,$T2,$T2
 834         vpclmulqdq      \$0x00,$HK,$Z3,$Z3
 835         vpxor           $Z0,$Z3,$Z0
 836
 837         vpclmulqdq      \$0x00,$Ii,$Xi,$Z2
 838          vmovdqu        0x00-0x20($Xip),$Hkey   # $Hkey^1
 839          vpunpckhqdq    $inout5,$inout5,$T1
 840         vpclmulqdq      \$0x11,$Ii,$Xi,$Xi
 841          vpxor          $inout5,$T1,$T1
 842         vpxor           $Z1,$Z2,$Z1
 843         vpclmulqdq      \$0x10,$HK,$T3,$T3
 844          vmovdqu        0x20-0x20($Xip),$HK
 845         vpxor           $T2,$Xi,$Z3
 846         vpxor           $Z0,$T3,$Z2
 847
 848          vmovdqu        0x10-0x20($Xip),$Ii     # borrow $Ii for $Hkey^2
 849           vpxor         $Z1,$Z3,$T3             # aggregated Karatsuba post-processing
 850         vpclmulqdq      \$0x00,$Hkey,$inout5,$Z0
 851           vpxor         $T3,$Z2,$Z2
 852          vpunpckhqdq    $inout4,$inout4,$T2
 853         vpclmulqdq      \$0x11,$Hkey,$inout5,$inout5
 854          vpxor          $inout4,$T2,$T2
 855           vpslldq       \$8,$Z2,$T3
 856         vpclmulqdq      \$0x00,$HK,$T1,$T1
 857           vpxor         $T3,$Z1,$Xi
 858           vpsrldq       \$8,$Z2,$Z2
 859           vpxor         $Z2,$Z3,$Z3
 860
 861         vpclmulqdq      \$0x00,$Ii,$inout4,$Z1
 862          vmovdqu        0x30-0x20($Xip),$Hkey   # $Hkey^3
 863         vpxor           $Z0,$Z1,$Z1
 864          vpunpckhqdq    $inout3,$inout3,$T3
 865         vpclmulqdq      \$0x11,$Ii,$inout4,$inout4
 866          vpxor          $inout3,$T3,$T3
 867         vpxor           $inout5,$inout4,$inout4
 868           vpalignr      \$8,$Xi,$Xi,$inout5     # 1st phase
 869         vpclmulqdq      \$0x10,$HK,$T2,$T2
 870          vmovdqu        0x50-0x20($Xip),$HK
 871         vpxor           $T1,$T2,$T2
 872
 873         vpclmulqdq      \$0x00,$Hkey,$inout3,$Z0
 874          vmovdqu        0x40-0x20($Xip),$Ii     # borrow $Ii for $Hkey^4
 875         vpxor           $Z1,$Z0,$Z0
 876          vpunpckhqdq    $inout2,$inout2,$T1
 877         vpclmulqdq      \$0x11,$Hkey,$inout3,$inout3
 878          vpxor          $inout2,$T1,$T1
 879         vpxor           $inout4,$inout3,$inout3
 880           vxorps        0x10(%rsp),$Z3,$Z3      # accumulate $inout0
 881         vpclmulqdq      \$0x00,$HK,$T3,$T3
 882         vpxor           $T2,$T3,$T3
 883
 884           vpclmulqdq    \$0x10,0x10($const),$Xi,$Xi
 885           vxorps        $inout5,$Xi,$Xi
 886
 887         vpclmulqdq      \$0x00,$Ii,$inout2,$Z1
 888          vmovdqu        0x60-0x20($Xip),$Hkey   # $Hkey^5
 889         vpxor           $Z0,$Z1,$Z1
 890          vpunpckhqdq    $inout1,$inout1,$T2
 891         vpclmulqdq      \$0x11,$Ii,$inout2,$inout2
 892          vpxor          $inout1,$T2,$T2
 893           vpalignr      \$8,$Xi,$Xi,$inout5     # 2nd phase
 894         vpxor           $inout3,$inout2,$inout2
 895         vpclmulqdq      \$0x10,$HK,$T1,$T1
 896          vmovdqu        0x80-0x20($Xip),$HK
 897         vpxor           $T3,$T1,$T1
 898
 899           vxorps        $Z3,$inout5,$inout5
 900           vpclmulqdq    \$0x10,0x10($const),$Xi,$Xi
 901           vxorps        $inout5,$Xi,$Xi
 902
 903         vpclmulqdq      \$0x00,$Hkey,$inout1,$Z0
 904          vmovdqu        0x70-0x20($Xip),$Ii     # borrow $Ii for $Hkey^6
 905         vpxor           $Z1,$Z0,$Z0
 906          vpunpckhqdq    $Xi,$Xi,$T3
 907         vpclmulqdq      \$0x11,$Hkey,$inout1,$inout1
 908          vpxor          $Xi,$T3,$T3
 909         vpxor           $inout2,$inout1,$inout1
 910         vpclmulqdq      \$0x00,$HK,$T2,$T2
 911         vpxor           $T1,$T2,$T2
 912
 913         vpclmulqdq      \$0x00,$Ii,$Xi,$Z1
 914         vpclmulqdq      \$0x11,$Ii,$Xi,$Z3
 915         vpxor           $Z0,$Z1,$Z1
 916         vpclmulqdq      \$0x10,$HK,$T3,$Z2
 917         vpxor           $inout1,$Z3,$Z3
 918         vpxor           $T2,$Z2,$Z2
 919
 920         vpxor           $Z1,$Z3,$Z0             # aggregated Karatsuba post-processing
 921         vpxor           $Z0,$Z2,$Z2
 922         vpslldq         \$8,$Z2,$T1
 923         vmovdqu         0x10($const),$Hkey      # .Lpoly
 924         vpsrldq         \$8,$Z2,$Z2
 925         vpxor           $T1,$Z1,$Xi
 926         vpxor           $Z2,$Z3,$Z3
 927
 928         vpalignr        \$8,$Xi,$Xi,$T2         # 1st phase
 929         vpclmulqdq      \$0x10,$Hkey,$Xi,$Xi
 930         vpxor           $T2,$Xi,$Xi
 931
 932         vpalignr        \$8,$Xi,$Xi,$T2         # 2nd phase
 933         vpclmulqdq      \$0x10,$Hkey,$Xi,$Xi
 934         vpxor           $Z3,$T2,$T2
 935         vpxor           $T2,$Xi,$Xi
 936 ___
 937 }
 938 $code.=<<___;
 939         vpshufb         ($const),$Xi,$Xi        # .Lbswap_mask
 940         vmovdqu         $Xi,-0x40($Xip)         # output Xi
 941
 942         vzeroupper
 943 ___
 944 $code.=<<___ if ($win64);
 945         movaps  -0xd8(%rax),%xmm6
 946         movaps  -0xc8(%rax),%xmm7
 947         movaps  -0xb8(%rax),%xmm8
 948         movaps  -0xa8(%rax),%xmm9
 949         movaps  -0x98(%rax),%xmm10
 950         movaps  -0x88(%rax),%xmm11
 951         movaps  -0x78(%rax),%xmm12
 952         movaps  -0x68(%rax),%xmm13
 953         movaps  -0x58(%rax),%xmm14
 954         movaps  -0x48(%rax),%xmm15
 955 ___
 956 $code.=<<___;
 957         mov     -48(%rax),%r15
 958 .cfi_restore    %r15
 959         mov     -40(%rax),%r14
 960 .cfi_restore    %r14
 961         mov     -32(%rax),%r13
 962 .cfi_restore    %r13
 963         mov     -24(%rax),%r12
 964 .cfi_restore    %r12
 965         mov     -16(%rax),%rbp
 966 .cfi_restore    %rbp
 967         mov     -8(%rax),%rbx
 968 .cfi_restore    %rbx
 969         lea     (%rax),%rsp             # restore %rsp
 970 .cfi_def_cfa_register   %rsp
 971 .Lgcm_enc_abort:
 972         mov     $ret,%rax               # return value
 973         ret
 974 .cfi_endproc
 975 .size   aesni_gcm_encrypt,.-aesni_gcm_encrypt
 976 ___
 977
 978 $code.=<<___;
 979 .align  64
 980 .Lbswap_mask:
 981         .byte   15,14,13,12,11,10,9,8,7,6,5,4,3,2,1,0
 982 .Lpoly:
 983         .byte   0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0xc2
 984 .Lone_msb:
 985         .byte   0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1
 986 .Ltwo_lsb:
 987         .byte   2,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0
 988 .Lone_lsb:
 989         .byte   1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0
 990 .asciz  "AES-NI GCM module for x86_64, CRYPTOGAMS by <appro\@openssl.org>"
 991 .align  64
 992 ___
 993 if ($win64) {
 994 $rec="%rcx";
 995 $frame="%rdx";
 996 $context="%r8";
 997 $disp="%r9";
 998
 999 $code.=<<___
1000 .extern __imp_RtlVirtualUnwind
1001 .type   gcm_se_handler,\@abi-omnipotent
1002 .align  16
1003 gcm_se_handler:
1004         push    %rsi
1005         push    %rdi
1006         push    %rbx
1007         push    %rbp
1008         push    %r12
1009         push    %r13
1010         push    %r14
1011         push    %r15
1012         pushfq
1013         sub     \$64,%rsp
1014
1015         mov     120($context),%rax      # pull context->Rax
1016         mov     248($context),%rbx      # pull context->Rip
1017
1018         mov     8($disp),%rsi           # disp->ImageBase
1019         mov     56($disp),%r11          # disp->HandlerData
1020
1021         mov     0(%r11),%r10d           # HandlerData[0]
1022         lea     (%rsi,%r10),%r10        # prologue label
1023         cmp     %r10,%rbx               # context->Rip<prologue label
1024         jb      .Lcommon_seh_tail
1025
1026         mov     152($context),%rax      # pull context->Rsp
1027
1028         mov     4(%r11),%r10d           # HandlerData[1]
1029         lea     (%rsi,%r10),%r10        # epilogue label
1030         cmp     %r10,%rbx               # context->Rip>=epilogue label
1031         jae     .Lcommon_seh_tail
1032
1033         mov     120($context),%rax      # pull context->Rax
1034
1035         mov     -48(%rax),%r15
1036         mov     -40(%rax),%r14
1037         mov     -32(%rax),%r13
1038         mov     -24(%rax),%r12
1039         mov     -16(%rax),%rbp
1040         mov     -8(%rax),%rbx
1041         mov     %r15,240($context)
1042         mov     %r14,232($context)
1043         mov     %r13,224($context)
1044         mov     %r12,216($context)
1045         mov     %rbp,160($context)
1046         mov     %rbx,144($context)
1047
1048         lea     -0xd8(%rax),%rsi        # %xmm save area
1049         lea     512($context),%rdi      # & context.Xmm6
1050         mov     \$20,%ecx               # 10*sizeof(%xmm0)/sizeof(%rax)
1051         .long   0xa548f3fc              # cld; rep movsq
1052
1053 .Lcommon_seh_tail:
1054         mov     8(%rax),%rdi
1055         mov     16(%rax),%rsi
1056         mov     %rax,152($context)      # restore context->Rsp
1057         mov     %rsi,168($context)      # restore context->Rsi
1058         mov     %rdi,176($context)      # restore context->Rdi
1059
1060         mov     40($disp),%rdi          # disp->ContextRecord
1061         mov     $context,%rsi           # context
1062         mov     \$154,%ecx              # sizeof(CONTEXT)
1063         .long   0xa548f3fc              # cld; rep movsq
1064
1065         mov     $disp,%rsi
1066         xor     %rcx,%rcx               # arg1, UNW_FLAG_NHANDLER
1067         mov     8(%rsi),%rdx            # arg2, disp->ImageBase
1068         mov     0(%rsi),%r8             # arg3, disp->ControlPc
1069         mov     16(%rsi),%r9            # arg4, disp->FunctionEntry
1070         mov     40(%rsi),%r10           # disp->ContextRecord
1071         lea     56(%rsi),%r11           # &disp->HandlerData
1072         lea     24(%rsi),%r12           # &disp->EstablisherFrame
1073         mov     %r10,32(%rsp)           # arg5
1074         mov     %r11,40(%rsp)           # arg6
1075         mov     %r12,48(%rsp)           # arg7
1076         mov     %rcx,56(%rsp)           # arg8, (NULL)
1077         call    *__imp_RtlVirtualUnwind(%rip)
1078
1079         mov     \$1,%eax                # ExceptionContinueSearch
1080         add     \$64,%rsp
1081         popfq
1082         pop     %r15
1083         pop     %r14
1084         pop     %r13
1085         pop     %r12
1086         pop     %rbp
1087         pop     %rbx
1088         pop     %rdi
1089         pop     %rsi
1090         ret
1091 .size   gcm_se_handler,.-gcm_se_handler
1092
1093 .section        .pdata
1094 .align  4
1095         .rva    .LSEH_begin_aesni_gcm_decrypt
1096         .rva    .LSEH_end_aesni_gcm_decrypt
1097         .rva    .LSEH_gcm_dec_info
1098
1099         .rva    .LSEH_begin_aesni_gcm_encrypt
1100         .rva    .LSEH_end_aesni_gcm_encrypt
1101         .rva    .LSEH_gcm_enc_info
1102 .section        .xdata
1103 .align  8
1104 .LSEH_gcm_dec_info:
1105         .byte   9,0,0,0
1106         .rva    gcm_se_handler
1107         .rva    .Lgcm_dec_body,.Lgcm_dec_abort
1108 .LSEH_gcm_enc_info:
1109         .byte   9,0,0,0
1110         .rva    gcm_se_handler
1111         .rva    .Lgcm_enc_body,.Lgcm_enc_abort
1112 ___
1113 }
1114 }}} else {{{
1115 $code=<<___;    # assembler is too old
1116 .text
1117
1118 .globl  aesni_gcm_encrypt
1119 .type   aesni_gcm_encrypt,\@abi-omnipotent
1120 aesni_gcm_encrypt:
1121         xor     %eax,%eax
1122         ret
1123 .size   aesni_gcm_encrypt,.-aesni_gcm_encrypt
1124
1125 .globl  aesni_gcm_decrypt
1126 .type   aesni_gcm_decrypt,\@abi-omnipotent
1127 aesni_gcm_decrypt:
1128         xor     %eax,%eax
1129         ret
1130 .size   aesni_gcm_decrypt,.-aesni_gcm_decrypt
1131 ___
1132 }}}
1133
1134 $code =~ s/\`([^\`]*)\`/eval($1)/gem;
1135
1136 print $code;
1137
1138 close STDOUT;