crypto/modes/asm/ghash-x86_64.pl

   1 #!/usr/bin/env perl
   2 #
   3 # ====================================================================
   4 # Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
   5 # project. The module is, however, dual licensed under OpenSSL and
   6 # CRYPTOGAMS licenses depending on where you obtain it. For further
   7 # details see http://www.openssl.org/~appro/cryptogams/.
   8 # ====================================================================
   9 #
  10 # March 2010
  11 #
  12 # The module implements "4-bit" GCM GHASH function and underlying
  13 # single multiplication operation in GF(2^128). "4-bit" means that it
  14 # uses 256 bytes per-key table [+128 bytes shared table]. Performance
  15 # results are for streamed GHASH subroutine and are expressed in
  16 # cycles per processed byte, less is better:
  17 #
  18 #               gcc 3.4.x       assembler
  19 #
  20 # Opteron       18.5            10.2            +80%
  21 # Core2         17.5            11.0            +59%
  22
  23 # May 2010
  24 #
  25 # Add PCLMULQDQ version performing at 2.07 cycles per processed byte.
  26 # See ghash-x86.pl for background information and details about coding
  27 # techniques.
  28 #
  29 # Special thanks to David Woodhouse <dwmw2@infradead.org> for
  30 # providing access to a Westmere-based system on behalf of Intel
  31 # Open Source Technology Centre.
  32
  33 $flavour = shift;
  34 $output  = shift;
  35 if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
  36
  37 $win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);
  38
  39 $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
  40 ( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
  41 ( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or
  42 die "can't locate x86_64-xlate.pl";
  43
  44 open STDOUT,"| $^X $xlate $flavour $output";
  45
  46 # common register layout
  47 $nlo="%rax";
  48 $nhi="%rbx";
  49 $Zlo="%r8";
  50 $Zhi="%r9";
  51 $tmp="%r10";
  52 $rem_4bit = "%r11";
  53
  54 $Xi="%rdi";
  55 $Htbl="%rsi";
  56
  57 # per-function register layout
  58 $cnt="%rcx";
  59 $rem="%rdx";
  60
  61 sub lo() { my $r=shift; $r =~ s/%[er]([a-d])x/%\1l/;
  62                         $r =~ s/%[er]([sd]i)/%\1l/;
  63                         $r =~ s/%(r[0-9]+)[d]?/%\1b/;   $r; }
  64 \f
  65 { my $N;
  66   sub loop() {
  67   my $inp = shift;
  68
  69         $N++;
  70 $code.=<<___;
  71         xor     $nlo,$nlo
  72         xor     $nhi,$nhi
  73         mov     `&lo("$Zlo")`,`&lo("$nlo")`
  74         mov     `&lo("$Zlo")`,`&lo("$nhi")`
  75         shl     \$4,`&lo("$nlo")`
  76         mov     \$14,$cnt
  77         mov     8($Htbl,$nlo),$Zlo
  78         mov     ($Htbl,$nlo),$Zhi
  79         and     \$0xf0,`&lo("$nhi")`
  80         mov     $Zlo,$rem
  81         jmp     .Loop$N
  82
  83 .align  16
  84 .Loop$N:
  85         shr     \$4,$Zlo
  86         and     \$0xf,$rem
  87         mov     $Zhi,$tmp
  88         mov     ($inp,$cnt),`&lo("$nlo")`
  89         shr     \$4,$Zhi
  90         xor     8($Htbl,$nhi),$Zlo
  91         shl     \$60,$tmp
  92         xor     ($Htbl,$nhi),$Zhi
  93         mov     `&lo("$nlo")`,`&lo("$nhi")`
  94         xor     ($rem_4bit,$rem,8),$Zhi
  95         mov     $Zlo,$rem
  96         shl     \$4,`&lo("$nlo")`
  97         xor     $tmp,$Zlo
  98         dec     $cnt
  99         js      .Lbreak$N
 100
 101         shr     \$4,$Zlo
 102         and     \$0xf,$rem
 103         mov     $Zhi,$tmp
 104         shr     \$4,$Zhi
 105         xor     8($Htbl,$nlo),$Zlo
 106         shl     \$60,$tmp
 107         xor     ($Htbl,$nlo),$Zhi
 108         and     \$0xf0,`&lo("$nhi")`
 109         xor     ($rem_4bit,$rem,8),$Zhi
 110         mov     $Zlo,$rem
 111         xor     $tmp,$Zlo
 112         jmp     .Loop$N
 113
 114 .align  16
 115 .Lbreak$N:
 116         shr     \$4,$Zlo
 117         and     \$0xf,$rem
 118         mov     $Zhi,$tmp
 119         shr     \$4,$Zhi
 120         xor     8($Htbl,$nlo),$Zlo
 121         shl     \$60,$tmp
 122         xor     ($Htbl,$nlo),$Zhi
 123         and     \$0xf0,`&lo("$nhi")`
 124         xor     ($rem_4bit,$rem,8),$Zhi
 125         mov     $Zlo,$rem
 126         xor     $tmp,$Zlo
 127
 128         shr     \$4,$Zlo
 129         and     \$0xf,$rem
 130         mov     $Zhi,$tmp
 131         shr     \$4,$Zhi
 132         xor     8($Htbl,$nhi),$Zlo
 133         shl     \$60,$tmp
 134         xor     ($Htbl,$nhi),$Zhi
 135         xor     $tmp,$Zlo
 136         xor     ($rem_4bit,$rem,8),$Zhi
 137
 138         bswap   $Zlo
 139         bswap   $Zhi
 140 ___
 141 }}
 142
 143 $code=<<___;
 144 .text
 145
 146 .globl  gcm_gmult_4bit
 147 .type   gcm_gmult_4bit,\@function,2
 148 .align  16
 149 gcm_gmult_4bit:
 150         push    %rbx
 151         push    %rbp            # %rbp and %r12 are pushed exclusively in
 152         push    %r12            # order to reuse Win64 exception handler...
 153 .Lgmult_prologue:
 154
 155         movzb   15($Xi),$Zlo
 156         lea     .Lrem_4bit(%rip),$rem_4bit
 157 ___
 158         &loop   ($Xi);
 159 $code.=<<___;
 160         mov     $Zlo,8($Xi)
 161         mov     $Zhi,($Xi)
 162
 163         mov     16(%rsp),%rbx
 164         lea     24(%rsp),%rsp
 165 .Lgmult_epilogue:
 166         ret
 167 .size   gcm_gmult_4bit,.-gcm_gmult_4bit
 168 ___
 169 \f
 170 # per-function register layout
 171 $inp="%rdx";
 172 $len="%rcx";
 173
 174 $cnt="%rbp";
 175 $rem="%r12";
 176
 177 $code.=<<___;
 178 .globl  gcm_ghash_4bit
 179 .type   gcm_ghash_4bit,\@function,4
 180 .align  16
 181 gcm_ghash_4bit:
 182         push    %rbx
 183         push    %rbp
 184         push    %r12
 185 .Lghash_prologue:
 186
 187         mov     8($Xi),$Zlo
 188         mov     ($Xi),$Zhi
 189         add     $inp,$len
 190         lea     .Lrem_4bit(%rip),$rem_4bit
 191 .align  4
 192 .Louter_loop:
 193         xor     8($inp),$Zlo
 194         xor     ($inp),$Zhi
 195         lea     16($inp),$inp
 196         mov     $Zlo,8($Xi)
 197         mov     $Zhi,($Xi)
 198         shr     \$56,$Zlo
 199 ___
 200         &loop   ($Xi);
 201 $code.=<<___;
 202         cmp     $len,$inp
 203         jb      .Louter_loop
 204
 205         mov     $Zlo,8($Xi)
 206         mov     $Zhi,($Xi)
 207
 208         mov     0(%rsp),%r12
 209         mov     8(%rsp),%rbp
 210         mov     16(%rsp),%rbx
 211         lea     24(%rsp),%rsp
 212 .Lghash_epilogue:
 213         ret
 214 .size   gcm_ghash_4bit,.-gcm_ghash_4bit
 215 ___
 216 \f
 217 ######################################################################
 218 # PCLMULQDQ version.
 219
 220 @_4args=$win64? ("%rcx","%rdx","%r8", "%r9") :  # Win64 order
 221                 ("%rdi","%rsi","%rdx","%rcx");  # Unix order
 222
 223 ($Xi,$Xhi)=("%xmm0","%xmm1");   $Hkey="%xmm2";
 224 ($T1,$T2,$T3)=("%xmm3","%xmm4","%xmm5");
 225
 226 sub clmul64x64_T2 {     # minimal register pressure
 227 my ($Xhi,$Xi,$Hkey,$modulo)=@_;
 228
 229 $code.=<<___ if (!defined($modulo));
 230         movdqa          $Xi,$Xhi                #
 231         pshufd          \$0b01001110,$Xi,$T1
 232         pshufd          \$0b01001110,$Hkey,$T2
 233         pxor            $Xi,$T1                 #
 234         pxor            $Hkey,$T2
 235 ___
 236 $code.=<<___;
 237         pclmulqdq       \$0x00,$Hkey,$Xi        #######
 238         pclmulqdq       \$0x11,$Hkey,$Xhi       #######
 239         pclmulqdq       \$0x00,$T2,$T1          #######
 240         pxor            $Xi,$T1                 #
 241         pxor            $Xhi,$T1                #
 242
 243         movdqa          $T1,$T2                 #
 244         psrldq          \$8,$T1
 245         pslldq          \$8,$T2                 #
 246         pxor            $T1,$Xhi
 247         pxor            $T2,$Xi                 #
 248 ___
 249 }
 250
 251 sub reduction_alg9 {    # 17/13 times faster than Intel version
 252 my ($Xhi,$Xi) = @_;
 253
 254 $code.=<<___;
 255         # 1st phase
 256         movdqa          $Xi,$T1                 #
 257         psllq           \$1,$Xi
 258         pxor            $T1,$Xi                 #
 259         psllq           \$5,$Xi                 #
 260         pxor            $T1,$Xi                 #
 261         psllq           \$57,$Xi                #
 262         movdqa          $Xi,$T2                 #
 263         pslldq          \$8,$Xi
 264         psrldq          \$8,$T2                 #
 265         pxor            $T1,$Xi
 266         pxor            $T2,$Xhi                #
 267
 268         # 2nd phase
 269         movdqa          $Xi,$T2
 270         psrlq           \$5,$Xi
 271         pxor            $T2,$Xi                 #
 272         psrlq           \$1,$Xi                 #
 273         pxor            $T2,$Xi                 #
 274         pxor            $Xhi,$T2
 275         psrlq           \$1,$Xi                 #
 276         pxor            $T2,$Xi                 #
 277 ___
 278 }
 279 \f
 280 { my ($Htbl,$Xip)=@_4args;
 281
 282 $code.=<<___;
 283 .globl  gcm_init_clmul
 284 .type   gcm_init_clmul,\@abi-omnipotent
 285 .align  16
 286 gcm_init_clmul:
 287         movdqu          ($Xip),$Hkey
 288         pshufd          \$0b01001110,$Hkey,$Hkey        # dword swap
 289
 290         # <<1 twist
 291         pshufd          \$0b11111111,$Hkey,$T2  # broadcast uppermost dword
 292         movdqa          $Hkey,$T1
 293         psllq           \$1,$Hkey
 294         pxor            $T3,$T3                 #
 295         psrlq           \$63,$T1
 296         pcmpgtd         $T2,$T3                 # broadcast carry bit
 297         pslldq          \$8,$T1
 298         por             $T1,$Hkey               # H<<=1
 299
 300         # magic reduction
 301         pand            .L0x1c2_polynomial(%rip),$T3
 302         pxor            $T3,$Hkey               # if(carry) H^=0x1c2_polynomial
 303
 304         # calculate H^2
 305         movdqa          $Hkey,$Xi
 306 ___
 307         &clmul64x64_T2  ($Xhi,$Xi,$Hkey);
 308         &reduction_alg9 ($Xhi,$Xi);
 309 $code.=<<___;
 310         movdqu          $Hkey,($Htbl)           # save H
 311         movdqu          $Xi,16($Htbl)           # save H^2
 312         ret
 313 .size   gcm_init_clmul,.-gcm_init_clmul
 314 ___
 315 }
 316
 317 { my ($Xip,$Htbl)=@_4args;
 318
 319 $code.=<<___;
 320 .globl  gcm_gmult_clmul
 321 .type   gcm_gmult_clmul,\@abi-omnipotent
 322 .align  16
 323 gcm_gmult_clmul:
 324         movdqu          ($Xip),$Xi
 325         movdqa          .Lbswap_mask(%rip),$T3
 326         movdqu          ($Htbl),$Hkey
 327         pshufb          $T3,$Xi
 328 ___
 329         &clmul64x64_T2  ($Xhi,$Xi,$Hkey);
 330         &reduction_alg9 ($Xhi,$Xi);
 331 $code.=<<___;
 332         pshufb          $T3,$Xi
 333         movdqu          $Xi,($Xip)
 334         ret
 335 .size   gcm_gmult_clmul,.-gcm_gmult_clmul
 336 ___
 337 }
 338 \f
 339 { my ($Xip,$Htbl,$inp,$len)=@_4args;
 340   my $Xn="%xmm6";
 341   my $Xhn="%xmm7";
 342   my $Hkey2="%xmm8";
 343   my $T1n="%xmm9";
 344   my $T2n="%xmm10";
 345
 346 $code.=<<___;
 347 .globl  gcm_ghash_clmul
 348 .type   gcm_ghash_clmul,\@abi-omnipotent
 349 .align  16
 350 gcm_ghash_clmul:
 351 ___
 352 $code.=<<___ if ($win64);
 353 .LSEH_begin_gcm_ghash_clmul:
 354         # I can't trust assembler to use specific encoding:-(
 355         .byte   0x48,0x83,0xec,0x58             #sub    \$0x58,%rsp
 356         .byte   0x0f,0x29,0x34,0x24             #movaps %xmm6,(%rsp)
 357         .byte   0x0f,0x29,0x7c,0x24,0x10        #movdqa %xmm7,0x10(%rsp)
 358         .byte   0x44,0x0f,0x29,0x44,0x24,0x20   #movaps %xmm8,0x20(%rsp)
 359         .byte   0x44,0x0f,0x29,0x4c,0x24,0x30   #movaps %xmm9,0x30(%rsp)
 360         .byte   0x44,0x0f,0x29,0x54,0x24,0x40   #movaps %xmm10,0x40(%rsp)
 361 ___
 362 $code.=<<___;
 363         movdqa          .Lbswap_mask(%rip),$T3
 364
 365         movdqu          ($Xip),$Xi
 366         movdqu          ($Htbl),$Hkey
 367         pshufb          $T3,$Xi
 368
 369         sub             \$0x10,$len
 370         jz              .Lodd_tail
 371
 372         movdqu          16($Htbl),$Hkey2
 373         #######
 374         # Xi+2 =[H*(Ii+1 + Xi+1)] mod P =
 375         #       [(H*Ii+1) + (H*Xi+1)] mod P =
 376         #       [(H*Ii+1) + H^2*(Ii+Xi)] mod P
 377         #
 378         movdqu          ($inp),$T1              # Ii
 379         movdqu          16($inp),$Xn            # Ii+1
 380         pshufb          $T3,$T1
 381         pshufb          $T3,$Xn
 382         pxor            $T1,$Xi                 # Ii+Xi
 383 ___
 384         &clmul64x64_T2  ($Xhn,$Xn,$Hkey);       # H*Ii+1
 385 $code.=<<___;
 386         movdqa          $Xi,$Xhi                #
 387         pshufd          \$0b01001110,$Xi,$T1
 388         pshufd          \$0b01001110,$Hkey2,$T2
 389         pxor            $Xi,$T1                 #
 390         pxor            $Hkey2,$T2
 391
 392         lea             32($inp),$inp           # i+=2
 393         sub             \$0x20,$len
 394         jbe             .Leven_tail
 395
 396 .Lmod_loop:
 397 ___
 398         &clmul64x64_T2  ($Xhi,$Xi,$Hkey2,1);    # H^2*(Ii+Xi)
 399 $code.=<<___;
 400         movdqu          ($inp),$T1              # Ii
 401         pxor            $Xn,$Xi                 # (H*Ii+1) + H^2*(Ii+Xi)
 402         pxor            $Xhn,$Xhi
 403
 404         movdqu          16($inp),$Xn            # Ii+1
 405         pshufb          $T3,$T1
 406         pshufb          $T3,$Xn
 407
 408         movdqa          $Xn,$Xhn                #
 409         pshufd          \$0b01001110,$Xn,$T1n
 410         pshufd          \$0b01001110,$Hkey,$T2n
 411         pxor            $Xn,$T1n                #
 412         pxor            $Hkey,$T2n
 413          pxor           $T1,$Xhi                # "Ii+Xi", consume early
 414
 415           movdqa        $Xi,$T1                 # 1st phase
 416           psllq         \$1,$Xi
 417           pxor          $T1,$Xi                 #
 418           psllq         \$5,$Xi                 #
 419           pxor          $T1,$Xi                 #
 420         pclmulqdq       \$0x00,$Hkey,$Xn        #######
 421           psllq         \$57,$Xi                #
 422           movdqa        $Xi,$T2                 #
 423           pslldq        \$8,$Xi
 424           psrldq        \$8,$T2                 #
 425           pxor          $T1,$Xi
 426           pxor          $T2,$Xhi                #
 427
 428         pclmulqdq       \$0x11,$Hkey,$Xhn       #######
 429           movdqa        $Xi,$T2                 # 2nd phase
 430           psrlq         \$5,$Xi
 431           pxor          $T2,$Xi                 #
 432           psrlq         \$1,$Xi                 #
 433           pxor          $T2,$Xi                 #
 434           pxor          $Xhi,$T2
 435           psrlq         \$1,$Xi                 #
 436           pxor          $T2,$Xi                 #
 437
 438         pclmulqdq       \$0x00,$T2n,$T1n        #######
 439          movdqa         $Xi,$Xhi                #
 440          pshufd         \$0b01001110,$Xi,$T1
 441          pshufd         \$0b01001110,$Hkey2,$T2
 442          pxor           $Xi,$T1                 #
 443          pxor           $Hkey2,$T2
 444
 445         pxor            $Xn,$T1n                #
 446         pxor            $Xhn,$T1n               #
 447         movdqa          $T1n,$T2n               #
 448         psrldq          \$8,$T1n
 449         pslldq          \$8,$T2n                #
 450         pxor            $T1n,$Xhn
 451         pxor            $T2n,$Xn                #
 452
 453         lea             32($inp),$inp
 454         sub             \$0x20,$len
 455         ja              .Lmod_loop
 456
 457 .Leven_tail:
 458 ___
 459         &clmul64x64_T2  ($Xhi,$Xi,$Hkey2,1);    # H^2*(Ii+Xi)
 460 $code.=<<___;
 461         pxor            $Xn,$Xi                 # (H*Ii+1) + H^2*(Ii+Xi)
 462         pxor            $Xhn,$Xhi
 463 ___
 464         &reduction_alg9 ($Xhi,$Xi);
 465 $code.=<<___;
 466         test            $len,$len
 467         jnz             .Ldone
 468
 469 .Lodd_tail:
 470         movdqu          ($inp),$T1              # Ii
 471         pshufb          $T3,$T1
 472         pxor            $T1,$Xi                 # Ii+Xi
 473 ___
 474         &clmul64x64_T2  ($Xhi,$Xi,$Hkey);       # H*(Ii+Xi)
 475         &reduction_alg9 ($Xhi,$Xi);
 476 $code.=<<___;
 477 .Ldone:
 478         pshufb          $T3,$Xi
 479         movdqu          $Xi,($Xip)
 480 ___
 481 $code.=<<___ if ($win64);
 482         movaps  (%rsp),%xmm6
 483         movaps  0x10(%rsp),%xmm7
 484         movaps  0x20(%rsp),%xmm8
 485         movaps  0x30(%rsp),%xmm9
 486         movaps  0x40(%rsp),%xmm10
 487         add     \$0x58,%rsp
 488 ___
 489 $code.=<<___;
 490         ret
 491 .LSEH_end_gcm_ghash_clmul:
 492 .size   gcm_ghash_clmul,.-gcm_ghash_clmul
 493 ___
 494 }
 495
 496 $code.=<<___;
 497 .align  64
 498 .Lbswap_mask:
 499         .byte   15,14,13,12,11,10,9,8,7,6,5,4,3,2,1,0
 500 .L0x1c2_polynomial:
 501         .byte   1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0xc2
 502 .align  64
 503 .type   .Lrem_4bit,\@object
 504 .Lrem_4bit:
 505         .long   0,`0x0000<<16`,0,`0x1C20<<16`,0,`0x3840<<16`,0,`0x2460<<16`
 506         .long   0,`0x7080<<16`,0,`0x6CA0<<16`,0,`0x48C0<<16`,0,`0x54E0<<16`
 507         .long   0,`0xE100<<16`,0,`0xFD20<<16`,0,`0xD940<<16`,0,`0xC560<<16`
 508         .long   0,`0x9180<<16`,0,`0x8DA0<<16`,0,`0xA9C0<<16`,0,`0xB5E0<<16`
 509 .asciz  "GHASH for x86_64, CRYPTOGAMS by <appro\@openssl.org>"
 510 .align  64
 511 ___
 512 \f
 513 # EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
 514 #               CONTEXT *context,DISPATCHER_CONTEXT *disp)
 515 if ($win64) {
 516 $rec="%rcx";
 517 $frame="%rdx";
 518 $context="%r8";
 519 $disp="%r9";
 520
 521 $code.=<<___;
 522 .extern __imp_RtlVirtualUnwind
 523 .type   se_handler,\@abi-omnipotent
 524 .align  16
 525 se_handler:
 526         push    %rsi
 527         push    %rdi
 528         push    %rbx
 529         push    %rbp
 530         push    %r12
 531         push    %r13
 532         push    %r14
 533         push    %r15
 534         pushfq
 535         sub     \$64,%rsp
 536
 537         mov     120($context),%rax      # pull context->Rax
 538         mov     248($context),%rbx      # pull context->Rip
 539
 540         mov     8($disp),%rsi           # disp->ImageBase
 541         mov     56($disp),%r11          # disp->HandlerData
 542
 543         mov     0(%r11),%r10d           # HandlerData[0]
 544         lea     (%rsi,%r10),%r10        # prologue label
 545         cmp     %r10,%rbx               # context->Rip<prologue label
 546         jb      .Lin_prologue
 547
 548         mov     152($context),%rax      # pull context->Rsp
 549
 550         mov     4(%r11),%r10d           # HandlerData[1]
 551         lea     (%rsi,%r10),%r10        # epilogue label
 552         cmp     %r10,%rbx               # context->Rip>=epilogue label
 553         jae     .Lin_prologue
 554
 555         lea     24(%rax),%rax           # adjust "rsp"
 556
 557         mov     -8(%rax),%rbx
 558         mov     -16(%rax),%rbp
 559         mov     -24(%rax),%r12
 560         mov     %rbx,144($context)      # restore context->Rbx
 561         mov     %rbp,160($context)      # restore context->Rbp
 562         mov     %r12,216($context)      # restore context->R12
 563
 564 .Lin_prologue:
 565         mov     8(%rax),%rdi
 566         mov     16(%rax),%rsi
 567         mov     %rax,152($context)      # restore context->Rsp
 568         mov     %rsi,168($context)      # restore context->Rsi
 569         mov     %rdi,176($context)      # restore context->Rdi
 570
 571         mov     40($disp),%rdi          # disp->ContextRecord
 572         mov     $context,%rsi           # context
 573         mov     \$`1232/8`,%ecx         # sizeof(CONTEXT)
 574         .long   0xa548f3fc              # cld; rep movsq
 575
 576         mov     $disp,%rsi
 577         xor     %rcx,%rcx               # arg1, UNW_FLAG_NHANDLER
 578         mov     8(%rsi),%rdx            # arg2, disp->ImageBase
 579         mov     0(%rsi),%r8             # arg3, disp->ControlPc
 580         mov     16(%rsi),%r9            # arg4, disp->FunctionEntry
 581         mov     40(%rsi),%r10           # disp->ContextRecord
 582         lea     56(%rsi),%r11           # &disp->HandlerData
 583         lea     24(%rsi),%r12           # &disp->EstablisherFrame
 584         mov     %r10,32(%rsp)           # arg5
 585         mov     %r11,40(%rsp)           # arg6
 586         mov     %r12,48(%rsp)           # arg7
 587         mov     %rcx,56(%rsp)           # arg8, (NULL)
 588         call    *__imp_RtlVirtualUnwind(%rip)
 589
 590         mov     \$1,%eax                # ExceptionContinueSearch
 591         add     \$64,%rsp
 592         popfq
 593         pop     %r15
 594         pop     %r14
 595         pop     %r13
 596         pop     %r12
 597         pop     %rbp
 598         pop     %rbx
 599         pop     %rdi
 600         pop     %rsi
 601         ret
 602 .size   se_handler,.-se_handler
 603
 604 .section        .pdata
 605 .align  4
 606         .rva    .LSEH_begin_gcm_gmult_4bit
 607         .rva    .LSEH_end_gcm_gmult_4bit
 608         .rva    .LSEH_info_gcm_gmult_4bit
 609
 610         .rva    .LSEH_begin_gcm_ghash_4bit
 611         .rva    .LSEH_end_gcm_ghash_4bit
 612         .rva    .LSEH_info_gcm_ghash_4bit
 613
 614         .rva    .LSEH_begin_gcm_ghash_clmul
 615         .rva    .LSEH_end_gcm_ghash_clmul
 616         .rva    .LSEH_info_gcm_ghash_clmul
 617
 618 .section        .xdata
 619 .align  8
 620 .LSEH_info_gcm_gmult_4bit:
 621         .byte   9,0,0,0
 622         .rva    se_handler
 623         .rva    .Lgmult_prologue,.Lgmult_epilogue       # HandlerData
 624 .LSEH_info_gcm_ghash_4bit:
 625         .byte   9,0,0,0
 626         .rva    se_handler
 627         .rva    .Lghash_prologue,.Lghash_epilogue       # HandlerData
 628 .LSEH_info_gcm_ghash_clmul:
 629         .byte   0x01,0x1f,0x0b,0x00
 630         .byte   0x1f,0xa8,0x04,0x00     #movaps 0x40(rsp),xmm10
 631         .byte   0x19,0x98,0x03,0x00     #movaps 0x30(rsp),xmm9
 632         .byte   0x13,0x88,0x02,0x00     #movaps 0x20(rsp),xmm8
 633         .byte   0x0d,0x78,0x01,0x00     #movaps 0x10(rsp),xmm7
 634         .byte   0x08,0x68,0x00,0x00     #movaps (rsp),xmm6
 635         .byte   0x04,0xa2,0x00,0x00     #sub    rsp,0x58
 636 ___
 637 }
 638 \f
 639 sub rex {
 640  local *opcode=shift;
 641  my ($dst,$src)=@_;
 642
 643    if ($dst>=8 || $src>=8) {
 644         $rex=0x40;
 645         $rex|=0x04 if($dst>=8);
 646         $rex|=0x01 if($src>=8);
 647         push @opcode,$rex;
 648    }
 649 }
 650
 651 sub pclmulqdq {
 652   my $arg=shift;
 653   my @opcode=(0x66);
 654
 655     if ($arg=~/\$([x0-9a-f]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) {
 656         rex(\@opcode,$3,$2);
 657         push @opcode,0x0f,0x3a,0x44;
 658         push @opcode,0xc0|($2&7)|(($3&7)<<3);   # ModR/M
 659         my $c=$1;
 660         push @opcode,$c=~/^0/?oct($c):$c;
 661         return ".byte\t".join(',',@opcode);
 662     }
 663     return "pclmulqdq\t".$arg;
 664 }
 665
 666 $code =~ s/\`([^\`]*)\`/eval($1)/gem;
 667 $code =~ s/\bpclmulqdq\s+(\$.*%xmm[0-9]+).*$/pclmulqdq($1)/gem;
 668
 669 print $code;
 670
 671 close STDOUT;