x86_64 assembly pack: make Windows build more robust.
[openssl.git] / 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, June 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
14 # it uses 256 bytes per-key table [+128 bytes shared table]. GHASH
15 # function features so called "528B" variant utilizing additional
16 # 256+16 bytes of per-key storage [+512 bytes shared table].
17 # Performance results are for this streamed GHASH subroutine and are
18 # expressed in cycles per processed byte, less is better:
19 #
20 #               gcc 3.4.x(*)    assembler
21 #
22 # P4            28.6            14.0            +100%
23 # Opteron       19.3            7.7             +150%
24 # Core2         17.8            8.1(**)         +120%
25 # Atom          31.6            16.8            +88%
26 # VIA Nano      21.8            10.1            +115%
27 #
28 # (*)   comparison is not completely fair, because C results are
29 #       for vanilla "256B" implementation, while assembler results
30 #       are for "528B";-)
31 # (**)  it's mystery [to me] why Core2 result is not same as for
32 #       Opteron;
33
34 # May 2010
35 #
36 # Add PCLMULQDQ version performing at 2.02 cycles per processed byte.
37 # See ghash-x86.pl for background information and details about coding
38 # techniques.
39 #
40 # Special thanks to David Woodhouse <dwmw2@infradead.org> for
41 # providing access to a Westmere-based system on behalf of Intel
42 # Open Source Technology Centre.
43
44 $flavour = shift;
45 $output  = shift;
46 if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
47
48 $win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);
49
50 $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
51 ( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
52 ( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or
53 die "can't locate x86_64-xlate.pl";
54
55 open OUT,"| \"$^X\" $xlate $flavour $output";
56 *STDOUT=*OUT;
57
58 # common register layout
59 $nlo="%rax";
60 $nhi="%rbx";
61 $Zlo="%r8";
62 $Zhi="%r9";
63 $tmp="%r10";
64 $rem_4bit = "%r11";
65
66 $Xi="%rdi";
67 $Htbl="%rsi";
68
69 # per-function register layout
70 $cnt="%rcx";
71 $rem="%rdx";
72
73 sub LB() { my $r=shift; $r =~ s/%[er]([a-d])x/%\1l/     or
74                         $r =~ s/%[er]([sd]i)/%\1l/      or
75                         $r =~ s/%[er](bp)/%\1l/         or
76                         $r =~ s/%(r[0-9]+)[d]?/%\1b/;   $r; }
77
78 sub AUTOLOAD()          # thunk [simplified] 32-bit style perlasm
79 { my $opcode = $AUTOLOAD; $opcode =~ s/.*:://;
80   my $arg = pop;
81     $arg = "\$$arg" if ($arg*1 eq $arg);
82     $code .= "\t$opcode\t".join(',',$arg,reverse @_)."\n";
83 }
84 \f
85 { my $N;
86   sub loop() {
87   my $inp = shift;
88
89         $N++;
90 $code.=<<___;
91         xor     $nlo,$nlo
92         xor     $nhi,$nhi
93         mov     `&LB("$Zlo")`,`&LB("$nlo")`
94         mov     `&LB("$Zlo")`,`&LB("$nhi")`
95         shl     \$4,`&LB("$nlo")`
96         mov     \$14,$cnt
97         mov     8($Htbl,$nlo),$Zlo
98         mov     ($Htbl,$nlo),$Zhi
99         and     \$0xf0,`&LB("$nhi")`
100         mov     $Zlo,$rem
101         jmp     .Loop$N
102
103 .align  16
104 .Loop$N:
105         shr     \$4,$Zlo
106         and     \$0xf,$rem
107         mov     $Zhi,$tmp
108         mov     ($inp,$cnt),`&LB("$nlo")`
109         shr     \$4,$Zhi
110         xor     8($Htbl,$nhi),$Zlo
111         shl     \$60,$tmp
112         xor     ($Htbl,$nhi),$Zhi
113         mov     `&LB("$nlo")`,`&LB("$nhi")`
114         xor     ($rem_4bit,$rem,8),$Zhi
115         mov     $Zlo,$rem
116         shl     \$4,`&LB("$nlo")`
117         xor     $tmp,$Zlo
118         dec     $cnt
119         js      .Lbreak$N
120
121         shr     \$4,$Zlo
122         and     \$0xf,$rem
123         mov     $Zhi,$tmp
124         shr     \$4,$Zhi
125         xor     8($Htbl,$nlo),$Zlo
126         shl     \$60,$tmp
127         xor     ($Htbl,$nlo),$Zhi
128         and     \$0xf0,`&LB("$nhi")`
129         xor     ($rem_4bit,$rem,8),$Zhi
130         mov     $Zlo,$rem
131         xor     $tmp,$Zlo
132         jmp     .Loop$N
133
134 .align  16
135 .Lbreak$N:
136         shr     \$4,$Zlo
137         and     \$0xf,$rem
138         mov     $Zhi,$tmp
139         shr     \$4,$Zhi
140         xor     8($Htbl,$nlo),$Zlo
141         shl     \$60,$tmp
142         xor     ($Htbl,$nlo),$Zhi
143         and     \$0xf0,`&LB("$nhi")`
144         xor     ($rem_4bit,$rem,8),$Zhi
145         mov     $Zlo,$rem
146         xor     $tmp,$Zlo
147
148         shr     \$4,$Zlo
149         and     \$0xf,$rem
150         mov     $Zhi,$tmp
151         shr     \$4,$Zhi
152         xor     8($Htbl,$nhi),$Zlo
153         shl     \$60,$tmp
154         xor     ($Htbl,$nhi),$Zhi
155         xor     $tmp,$Zlo
156         xor     ($rem_4bit,$rem,8),$Zhi
157
158         bswap   $Zlo
159         bswap   $Zhi
160 ___
161 }}
162
163 $code=<<___;
164 .text
165
166 .globl  gcm_gmult_4bit
167 .type   gcm_gmult_4bit,\@function,2
168 .align  16
169 gcm_gmult_4bit:
170         push    %rbx
171         push    %rbp            # %rbp and %r12 are pushed exclusively in
172         push    %r12            # order to reuse Win64 exception handler...
173 .Lgmult_prologue:
174
175         movzb   15($Xi),$Zlo
176         lea     .Lrem_4bit(%rip),$rem_4bit
177 ___
178         &loop   ($Xi);
179 $code.=<<___;
180         mov     $Zlo,8($Xi)
181         mov     $Zhi,($Xi)
182
183         mov     16(%rsp),%rbx
184         lea     24(%rsp),%rsp
185 .Lgmult_epilogue:
186         ret
187 .size   gcm_gmult_4bit,.-gcm_gmult_4bit
188 ___
189 \f
190 # per-function register layout
191 $inp="%rdx";
192 $len="%rcx";
193 $rem_8bit=$rem_4bit;
194
195 $code.=<<___;
196 .globl  gcm_ghash_4bit
197 .type   gcm_ghash_4bit,\@function,4
198 .align  16
199 gcm_ghash_4bit:
200         push    %rbx
201         push    %rbp
202         push    %r12
203         push    %r13
204         push    %r14
205         push    %r15
206         sub     \$280,%rsp
207 .Lghash_prologue:
208         mov     $inp,%r14               # reassign couple of args
209         mov     $len,%r15
210 ___
211 { my $inp="%r14";
212   my $dat="%edx";
213   my $len="%r15";
214   my @nhi=("%ebx","%ecx");
215   my @rem=("%r12","%r13");
216   my $Hshr4="%rbp";
217
218         &sub    ($Htbl,-128);           # size optimization
219         &lea    ($Hshr4,"16+128(%rsp)");
220         { my @lo =($nlo,$nhi);
221           my @hi =($Zlo,$Zhi);
222
223           &xor  ($dat,$dat);
224           for ($i=0,$j=-2;$i<18;$i++,$j++) {
225             &mov        ("$j(%rsp)",&LB($dat))          if ($i>1);
226             &or         ($lo[0],$tmp)                   if ($i>1);
227             &mov        (&LB($dat),&LB($lo[1]))         if ($i>0 && $i<17);
228             &shr        ($lo[1],4)                      if ($i>0 && $i<17);
229             &mov        ($tmp,$hi[1])                   if ($i>0 && $i<17);
230             &shr        ($hi[1],4)                      if ($i>0 && $i<17);
231             &mov        ("8*$j($Hshr4)",$hi[0])         if ($i>1);
232             &mov        ($hi[0],"16*$i+0-128($Htbl)")   if ($i<16);
233             &shl        (&LB($dat),4)                   if ($i>0 && $i<17);
234             &mov        ("8*$j-128($Hshr4)",$lo[0])     if ($i>1);
235             &mov        ($lo[0],"16*$i+8-128($Htbl)")   if ($i<16);
236             &shl        ($tmp,60)                       if ($i>0 && $i<17);
237
238             push        (@lo,shift(@lo));
239             push        (@hi,shift(@hi));
240           }
241         }
242         &add    ($Htbl,-128);
243         &mov    ($Zlo,"8($Xi)");
244         &mov    ($Zhi,"0($Xi)");
245         &add    ($len,$inp);            # pointer to the end of data
246         &lea    ($rem_8bit,".Lrem_8bit(%rip)");
247         &jmp    (".Louter_loop");
248
249 $code.=".align  16\n.Louter_loop:\n";
250         &xor    ($Zhi,"($inp)");
251         &mov    ("%rdx","8($inp)");
252         &lea    ($inp,"16($inp)");
253         &xor    ("%rdx",$Zlo);
254         &mov    ("($Xi)",$Zhi);
255         &mov    ("8($Xi)","%rdx");
256         &shr    ("%rdx",32);
257
258         &xor    ($nlo,$nlo);
259         &rol    ($dat,8);
260         &mov    (&LB($nlo),&LB($dat));
261         &movz   ($nhi[0],&LB($dat));
262         &shl    (&LB($nlo),4);
263         &shr    ($nhi[0],4);
264
265         for ($j=11,$i=0;$i<15;$i++) {
266             &rol        ($dat,8);
267             &xor        ($Zlo,"8($Htbl,$nlo)")                  if ($i>0);
268             &xor        ($Zhi,"($Htbl,$nlo)")                   if ($i>0);
269             &mov        ($Zlo,"8($Htbl,$nlo)")                  if ($i==0);
270             &mov        ($Zhi,"($Htbl,$nlo)")                   if ($i==0);
271
272             &mov        (&LB($nlo),&LB($dat));
273             &xor        ($Zlo,$tmp)                             if ($i>0);
274             &movzw      ($rem[1],"($rem_8bit,$rem[1],2)")       if ($i>0);
275
276             &movz       ($nhi[1],&LB($dat));
277             &shl        (&LB($nlo),4);
278             &movzb      ($rem[0],"(%rsp,$nhi[0])");
279
280             &shr        ($nhi[1],4)                             if ($i<14);
281             &and        ($nhi[1],0xf0)                          if ($i==14);
282             &shl        ($rem[1],48)                            if ($i>0);
283             &xor        ($rem[0],$Zlo);
284
285             &mov        ($tmp,$Zhi);
286             &xor        ($Zhi,$rem[1])                          if ($i>0);
287             &shr        ($Zlo,8);
288
289             &movz       ($rem[0],&LB($rem[0]));
290             &mov        ($dat,"$j($Xi)")                        if (--$j%4==0);
291             &shr        ($Zhi,8);
292
293             &xor        ($Zlo,"-128($Hshr4,$nhi[0],8)");
294             &shl        ($tmp,56);
295             &xor        ($Zhi,"($Hshr4,$nhi[0],8)");
296
297             unshift     (@nhi,pop(@nhi));               # "rotate" registers
298             unshift     (@rem,pop(@rem));
299         }
300         &movzw  ($rem[1],"($rem_8bit,$rem[1],2)");
301         &xor    ($Zlo,"8($Htbl,$nlo)");
302         &xor    ($Zhi,"($Htbl,$nlo)");
303
304         &shl    ($rem[1],48);
305         &xor    ($Zlo,$tmp);
306
307         &xor    ($Zhi,$rem[1]);
308         &movz   ($rem[0],&LB($Zlo));
309         &shr    ($Zlo,4);
310
311         &mov    ($tmp,$Zhi);
312         &shl    (&LB($rem[0]),4);
313         &shr    ($Zhi,4);
314
315         &xor    ($Zlo,"8($Htbl,$nhi[0])");
316         &movzw  ($rem[0],"($rem_8bit,$rem[0],2)");
317         &shl    ($tmp,60);
318
319         &xor    ($Zhi,"($Htbl,$nhi[0])");
320         &xor    ($Zlo,$tmp);
321         &shl    ($rem[0],48);
322
323         &bswap  ($Zlo);
324         &xor    ($Zhi,$rem[0]);
325
326         &bswap  ($Zhi);
327         &cmp    ($inp,$len);
328         &jb     (".Louter_loop");
329 }
330 $code.=<<___;
331         mov     $Zlo,8($Xi)
332         mov     $Zhi,($Xi)
333
334         lea     280(%rsp),%rsi
335         mov     0(%rsi),%r15
336         mov     8(%rsi),%r14
337         mov     16(%rsi),%r13
338         mov     24(%rsi),%r12
339         mov     32(%rsi),%rbp
340         mov     40(%rsi),%rbx
341         lea     48(%rsi),%rsp
342 .Lghash_epilogue:
343         ret
344 .size   gcm_ghash_4bit,.-gcm_ghash_4bit
345 ___
346 \f
347 ######################################################################
348 # PCLMULQDQ version.
349
350 @_4args=$win64? ("%rcx","%rdx","%r8", "%r9") :  # Win64 order
351                 ("%rdi","%rsi","%rdx","%rcx");  # Unix order
352
353 ($Xi,$Xhi)=("%xmm0","%xmm1");   $Hkey="%xmm2";
354 ($T1,$T2,$T3)=("%xmm3","%xmm4","%xmm5");
355
356 sub clmul64x64_T2 {     # minimal register pressure
357 my ($Xhi,$Xi,$Hkey,$modulo)=@_;
358
359 $code.=<<___ if (!defined($modulo));
360         movdqa          $Xi,$Xhi                #
361         pshufd          \$0b01001110,$Xi,$T1
362         pshufd          \$0b01001110,$Hkey,$T2
363         pxor            $Xi,$T1                 #
364         pxor            $Hkey,$T2
365 ___
366 $code.=<<___;
367         pclmulqdq       \$0x00,$Hkey,$Xi        #######
368         pclmulqdq       \$0x11,$Hkey,$Xhi       #######
369         pclmulqdq       \$0x00,$T2,$T1          #######
370         pxor            $Xi,$T1                 #
371         pxor            $Xhi,$T1                #
372
373         movdqa          $T1,$T2                 #
374         psrldq          \$8,$T1
375         pslldq          \$8,$T2                 #
376         pxor            $T1,$Xhi
377         pxor            $T2,$Xi                 #
378 ___
379 }
380
381 sub reduction_alg9 {    # 17/13 times faster than Intel version
382 my ($Xhi,$Xi) = @_;
383
384 $code.=<<___;
385         # 1st phase
386         movdqa          $Xi,$T1                 #
387         psllq           \$1,$Xi
388         pxor            $T1,$Xi                 #
389         psllq           \$5,$Xi                 #
390         pxor            $T1,$Xi                 #
391         psllq           \$57,$Xi                #
392         movdqa          $Xi,$T2                 #
393         pslldq          \$8,$Xi
394         psrldq          \$8,$T2                 #       
395         pxor            $T1,$Xi
396         pxor            $T2,$Xhi                #
397
398         # 2nd phase
399         movdqa          $Xi,$T2
400         psrlq           \$5,$Xi
401         pxor            $T2,$Xi                 #
402         psrlq           \$1,$Xi                 #
403         pxor            $T2,$Xi                 #
404         pxor            $Xhi,$T2
405         psrlq           \$1,$Xi                 #
406         pxor            $T2,$Xi                 #
407 ___
408 }
409 \f
410 { my ($Htbl,$Xip)=@_4args;
411
412 $code.=<<___;
413 .globl  gcm_init_clmul
414 .type   gcm_init_clmul,\@abi-omnipotent
415 .align  16
416 gcm_init_clmul:
417         movdqu          ($Xip),$Hkey
418         pshufd          \$0b01001110,$Hkey,$Hkey        # dword swap
419
420         # <<1 twist
421         pshufd          \$0b11111111,$Hkey,$T2  # broadcast uppermost dword
422         movdqa          $Hkey,$T1
423         psllq           \$1,$Hkey
424         pxor            $T3,$T3                 #
425         psrlq           \$63,$T1
426         pcmpgtd         $T2,$T3                 # broadcast carry bit
427         pslldq          \$8,$T1
428         por             $T1,$Hkey               # H<<=1
429
430         # magic reduction
431         pand            .L0x1c2_polynomial(%rip),$T3
432         pxor            $T3,$Hkey               # if(carry) H^=0x1c2_polynomial
433
434         # calculate H^2
435         movdqa          $Hkey,$Xi
436 ___
437         &clmul64x64_T2  ($Xhi,$Xi,$Hkey);
438         &reduction_alg9 ($Xhi,$Xi);
439 $code.=<<___;
440         movdqu          $Hkey,($Htbl)           # save H
441         movdqu          $Xi,16($Htbl)           # save H^2
442         ret
443 .size   gcm_init_clmul,.-gcm_init_clmul
444 ___
445 }
446
447 { my ($Xip,$Htbl)=@_4args;
448
449 $code.=<<___;
450 .globl  gcm_gmult_clmul
451 .type   gcm_gmult_clmul,\@abi-omnipotent
452 .align  16
453 gcm_gmult_clmul:
454         movdqu          ($Xip),$Xi
455         movdqa          .Lbswap_mask(%rip),$T3
456         movdqu          ($Htbl),$Hkey
457         pshufb          $T3,$Xi
458 ___
459         &clmul64x64_T2  ($Xhi,$Xi,$Hkey);
460         &reduction_alg9 ($Xhi,$Xi);
461 $code.=<<___;
462         pshufb          $T3,$Xi
463         movdqu          $Xi,($Xip)
464         ret
465 .size   gcm_gmult_clmul,.-gcm_gmult_clmul
466 ___
467 }
468 \f
469 { my ($Xip,$Htbl,$inp,$len)=@_4args;
470   my $Xn="%xmm6";
471   my $Xhn="%xmm7";
472   my $Hkey2="%xmm8";
473   my $T1n="%xmm9";
474   my $T2n="%xmm10";
475
476 $code.=<<___;
477 .globl  gcm_ghash_clmul
478 .type   gcm_ghash_clmul,\@abi-omnipotent
479 .align  16
480 gcm_ghash_clmul:
481 ___
482 $code.=<<___ if ($win64);
483 .LSEH_begin_gcm_ghash_clmul:
484         # I can't trust assembler to use specific encoding:-(
485         .byte   0x48,0x83,0xec,0x58             #sub    \$0x58,%rsp
486         .byte   0x0f,0x29,0x34,0x24             #movaps %xmm6,(%rsp)
487         .byte   0x0f,0x29,0x7c,0x24,0x10        #movdqa %xmm7,0x10(%rsp)
488         .byte   0x44,0x0f,0x29,0x44,0x24,0x20   #movaps %xmm8,0x20(%rsp)
489         .byte   0x44,0x0f,0x29,0x4c,0x24,0x30   #movaps %xmm9,0x30(%rsp)
490         .byte   0x44,0x0f,0x29,0x54,0x24,0x40   #movaps %xmm10,0x40(%rsp)
491 ___
492 $code.=<<___;
493         movdqa          .Lbswap_mask(%rip),$T3
494
495         movdqu          ($Xip),$Xi
496         movdqu          ($Htbl),$Hkey
497         pshufb          $T3,$Xi
498
499         sub             \$0x10,$len
500         jz              .Lodd_tail
501
502         movdqu          16($Htbl),$Hkey2
503         #######
504         # Xi+2 =[H*(Ii+1 + Xi+1)] mod P =
505         #       [(H*Ii+1) + (H*Xi+1)] mod P =
506         #       [(H*Ii+1) + H^2*(Ii+Xi)] mod P
507         #
508         movdqu          ($inp),$T1              # Ii
509         movdqu          16($inp),$Xn            # Ii+1
510         pshufb          $T3,$T1
511         pshufb          $T3,$Xn
512         pxor            $T1,$Xi                 # Ii+Xi
513 ___
514         &clmul64x64_T2  ($Xhn,$Xn,$Hkey);       # H*Ii+1
515 $code.=<<___;
516         movdqa          $Xi,$Xhi                #
517         pshufd          \$0b01001110,$Xi,$T1
518         pshufd          \$0b01001110,$Hkey2,$T2
519         pxor            $Xi,$T1                 #
520         pxor            $Hkey2,$T2
521
522         lea             32($inp),$inp           # i+=2
523         sub             \$0x20,$len
524         jbe             .Leven_tail
525
526 .Lmod_loop:
527 ___
528         &clmul64x64_T2  ($Xhi,$Xi,$Hkey2,1);    # H^2*(Ii+Xi)
529 $code.=<<___;
530         movdqu          ($inp),$T1              # Ii
531         pxor            $Xn,$Xi                 # (H*Ii+1) + H^2*(Ii+Xi)
532         pxor            $Xhn,$Xhi
533
534         movdqu          16($inp),$Xn            # Ii+1
535         pshufb          $T3,$T1
536         pshufb          $T3,$Xn
537
538         movdqa          $Xn,$Xhn                #
539         pshufd          \$0b01001110,$Xn,$T1n
540         pshufd          \$0b01001110,$Hkey,$T2n
541         pxor            $Xn,$T1n                #
542         pxor            $Hkey,$T2n
543          pxor           $T1,$Xhi                # "Ii+Xi", consume early
544
545           movdqa        $Xi,$T1                 # 1st phase
546           psllq         \$1,$Xi
547           pxor          $T1,$Xi                 #
548           psllq         \$5,$Xi                 #
549           pxor          $T1,$Xi                 #
550         pclmulqdq       \$0x00,$Hkey,$Xn        #######
551           psllq         \$57,$Xi                #
552           movdqa        $Xi,$T2                 #
553           pslldq        \$8,$Xi
554           psrldq        \$8,$T2                 #       
555           pxor          $T1,$Xi
556           pxor          $T2,$Xhi                #
557
558         pclmulqdq       \$0x11,$Hkey,$Xhn       #######
559           movdqa        $Xi,$T2                 # 2nd phase
560           psrlq         \$5,$Xi
561           pxor          $T2,$Xi                 #
562           psrlq         \$1,$Xi                 #
563           pxor          $T2,$Xi                 #
564           pxor          $Xhi,$T2
565           psrlq         \$1,$Xi                 #
566           pxor          $T2,$Xi                 #
567
568         pclmulqdq       \$0x00,$T2n,$T1n        #######
569          movdqa         $Xi,$Xhi                #
570          pshufd         \$0b01001110,$Xi,$T1
571          pshufd         \$0b01001110,$Hkey2,$T2
572          pxor           $Xi,$T1                 #
573          pxor           $Hkey2,$T2
574
575         pxor            $Xn,$T1n                #
576         pxor            $Xhn,$T1n               #
577         movdqa          $T1n,$T2n               #
578         psrldq          \$8,$T1n
579         pslldq          \$8,$T2n                #
580         pxor            $T1n,$Xhn
581         pxor            $T2n,$Xn                #
582
583         lea             32($inp),$inp
584         sub             \$0x20,$len
585         ja              .Lmod_loop
586
587 .Leven_tail:
588 ___
589         &clmul64x64_T2  ($Xhi,$Xi,$Hkey2,1);    # H^2*(Ii+Xi)
590 $code.=<<___;
591         pxor            $Xn,$Xi                 # (H*Ii+1) + H^2*(Ii+Xi)
592         pxor            $Xhn,$Xhi
593 ___
594         &reduction_alg9 ($Xhi,$Xi);
595 $code.=<<___;
596         test            $len,$len
597         jnz             .Ldone
598
599 .Lodd_tail:
600         movdqu          ($inp),$T1              # Ii
601         pshufb          $T3,$T1
602         pxor            $T1,$Xi                 # Ii+Xi
603 ___
604         &clmul64x64_T2  ($Xhi,$Xi,$Hkey);       # H*(Ii+Xi)
605         &reduction_alg9 ($Xhi,$Xi);
606 $code.=<<___;
607 .Ldone:
608         pshufb          $T3,$Xi
609         movdqu          $Xi,($Xip)
610 ___
611 $code.=<<___ if ($win64);
612         movaps  (%rsp),%xmm6
613         movaps  0x10(%rsp),%xmm7
614         movaps  0x20(%rsp),%xmm8
615         movaps  0x30(%rsp),%xmm9
616         movaps  0x40(%rsp),%xmm10
617         add     \$0x58,%rsp
618 ___
619 $code.=<<___;
620         ret
621 .LSEH_end_gcm_ghash_clmul:
622 .size   gcm_ghash_clmul,.-gcm_ghash_clmul
623 ___
624 }
625
626 $code.=<<___;
627 .align  64
628 .Lbswap_mask:
629         .byte   15,14,13,12,11,10,9,8,7,6,5,4,3,2,1,0
630 .L0x1c2_polynomial:
631         .byte   1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0xc2
632 .align  64
633 .type   .Lrem_4bit,\@object
634 .Lrem_4bit:
635         .long   0,`0x0000<<16`,0,`0x1C20<<16`,0,`0x3840<<16`,0,`0x2460<<16`
636         .long   0,`0x7080<<16`,0,`0x6CA0<<16`,0,`0x48C0<<16`,0,`0x54E0<<16`
637         .long   0,`0xE100<<16`,0,`0xFD20<<16`,0,`0xD940<<16`,0,`0xC560<<16`
638         .long   0,`0x9180<<16`,0,`0x8DA0<<16`,0,`0xA9C0<<16`,0,`0xB5E0<<16`
639 .type   .Lrem_8bit,\@object
640 .Lrem_8bit:
641         .value  0x0000,0x01C2,0x0384,0x0246,0x0708,0x06CA,0x048C,0x054E
642         .value  0x0E10,0x0FD2,0x0D94,0x0C56,0x0918,0x08DA,0x0A9C,0x0B5E
643         .value  0x1C20,0x1DE2,0x1FA4,0x1E66,0x1B28,0x1AEA,0x18AC,0x196E
644         .value  0x1230,0x13F2,0x11B4,0x1076,0x1538,0x14FA,0x16BC,0x177E
645         .value  0x3840,0x3982,0x3BC4,0x3A06,0x3F48,0x3E8A,0x3CCC,0x3D0E
646         .value  0x3650,0x3792,0x35D4,0x3416,0x3158,0x309A,0x32DC,0x331E
647         .value  0x2460,0x25A2,0x27E4,0x2626,0x2368,0x22AA,0x20EC,0x212E
648         .value  0x2A70,0x2BB2,0x29F4,0x2836,0x2D78,0x2CBA,0x2EFC,0x2F3E
649         .value  0x7080,0x7142,0x7304,0x72C6,0x7788,0x764A,0x740C,0x75CE
650         .value  0x7E90,0x7F52,0x7D14,0x7CD6,0x7998,0x785A,0x7A1C,0x7BDE
651         .value  0x6CA0,0x6D62,0x6F24,0x6EE6,0x6BA8,0x6A6A,0x682C,0x69EE
652         .value  0x62B0,0x6372,0x6134,0x60F6,0x65B8,0x647A,0x663C,0x67FE
653         .value  0x48C0,0x4902,0x4B44,0x4A86,0x4FC8,0x4E0A,0x4C4C,0x4D8E
654         .value  0x46D0,0x4712,0x4554,0x4496,0x41D8,0x401A,0x425C,0x439E
655         .value  0x54E0,0x5522,0x5764,0x56A6,0x53E8,0x522A,0x506C,0x51AE
656         .value  0x5AF0,0x5B32,0x5974,0x58B6,0x5DF8,0x5C3A,0x5E7C,0x5FBE
657         .value  0xE100,0xE0C2,0xE284,0xE346,0xE608,0xE7CA,0xE58C,0xE44E
658         .value  0xEF10,0xEED2,0xEC94,0xED56,0xE818,0xE9DA,0xEB9C,0xEA5E
659         .value  0xFD20,0xFCE2,0xFEA4,0xFF66,0xFA28,0xFBEA,0xF9AC,0xF86E
660         .value  0xF330,0xF2F2,0xF0B4,0xF176,0xF438,0xF5FA,0xF7BC,0xF67E
661         .value  0xD940,0xD882,0xDAC4,0xDB06,0xDE48,0xDF8A,0xDDCC,0xDC0E
662         .value  0xD750,0xD692,0xD4D4,0xD516,0xD058,0xD19A,0xD3DC,0xD21E
663         .value  0xC560,0xC4A2,0xC6E4,0xC726,0xC268,0xC3AA,0xC1EC,0xC02E
664         .value  0xCB70,0xCAB2,0xC8F4,0xC936,0xCC78,0xCDBA,0xCFFC,0xCE3E
665         .value  0x9180,0x9042,0x9204,0x93C6,0x9688,0x974A,0x950C,0x94CE
666         .value  0x9F90,0x9E52,0x9C14,0x9DD6,0x9898,0x995A,0x9B1C,0x9ADE
667         .value  0x8DA0,0x8C62,0x8E24,0x8FE6,0x8AA8,0x8B6A,0x892C,0x88EE
668         .value  0x83B0,0x8272,0x8034,0x81F6,0x84B8,0x857A,0x873C,0x86FE
669         .value  0xA9C0,0xA802,0xAA44,0xAB86,0xAEC8,0xAF0A,0xAD4C,0xAC8E
670         .value  0xA7D0,0xA612,0xA454,0xA596,0xA0D8,0xA11A,0xA35C,0xA29E
671         .value  0xB5E0,0xB422,0xB664,0xB7A6,0xB2E8,0xB32A,0xB16C,0xB0AE
672         .value  0xBBF0,0xBA32,0xB874,0xB9B6,0xBCF8,0xBD3A,0xBF7C,0xBEBE
673
674 .asciz  "GHASH for x86_64, CRYPTOGAMS by <appro\@openssl.org>"
675 .align  64
676 ___
677 \f
678 # EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
679 #               CONTEXT *context,DISPATCHER_CONTEXT *disp)
680 if ($win64) {
681 $rec="%rcx";
682 $frame="%rdx";
683 $context="%r8";
684 $disp="%r9";
685
686 $code.=<<___;
687 .extern __imp_RtlVirtualUnwind
688 .type   se_handler,\@abi-omnipotent
689 .align  16
690 se_handler:
691         push    %rsi
692         push    %rdi
693         push    %rbx
694         push    %rbp
695         push    %r12
696         push    %r13
697         push    %r14
698         push    %r15
699         pushfq
700         sub     \$64,%rsp
701
702         mov     120($context),%rax      # pull context->Rax
703         mov     248($context),%rbx      # pull context->Rip
704
705         mov     8($disp),%rsi           # disp->ImageBase
706         mov     56($disp),%r11          # disp->HandlerData
707
708         mov     0(%r11),%r10d           # HandlerData[0]
709         lea     (%rsi,%r10),%r10        # prologue label
710         cmp     %r10,%rbx               # context->Rip<prologue label
711         jb      .Lin_prologue
712
713         mov     152($context),%rax      # pull context->Rsp
714
715         mov     4(%r11),%r10d           # HandlerData[1]
716         lea     (%rsi,%r10),%r10        # epilogue label
717         cmp     %r10,%rbx               # context->Rip>=epilogue label
718         jae     .Lin_prologue
719
720         lea     24(%rax),%rax           # adjust "rsp"
721
722         mov     -8(%rax),%rbx
723         mov     -16(%rax),%rbp
724         mov     -24(%rax),%r12
725         mov     %rbx,144($context)      # restore context->Rbx
726         mov     %rbp,160($context)      # restore context->Rbp
727         mov     %r12,216($context)      # restore context->R12
728
729 .Lin_prologue:
730         mov     8(%rax),%rdi
731         mov     16(%rax),%rsi
732         mov     %rax,152($context)      # restore context->Rsp
733         mov     %rsi,168($context)      # restore context->Rsi
734         mov     %rdi,176($context)      # restore context->Rdi
735
736         mov     40($disp),%rdi          # disp->ContextRecord
737         mov     $context,%rsi           # context
738         mov     \$`1232/8`,%ecx         # sizeof(CONTEXT)
739         .long   0xa548f3fc              # cld; rep movsq
740
741         mov     $disp,%rsi
742         xor     %rcx,%rcx               # arg1, UNW_FLAG_NHANDLER
743         mov     8(%rsi),%rdx            # arg2, disp->ImageBase
744         mov     0(%rsi),%r8             # arg3, disp->ControlPc
745         mov     16(%rsi),%r9            # arg4, disp->FunctionEntry
746         mov     40(%rsi),%r10           # disp->ContextRecord
747         lea     56(%rsi),%r11           # &disp->HandlerData
748         lea     24(%rsi),%r12           # &disp->EstablisherFrame
749         mov     %r10,32(%rsp)           # arg5
750         mov     %r11,40(%rsp)           # arg6
751         mov     %r12,48(%rsp)           # arg7
752         mov     %rcx,56(%rsp)           # arg8, (NULL)
753         call    *__imp_RtlVirtualUnwind(%rip)
754
755         mov     \$1,%eax                # ExceptionContinueSearch
756         add     \$64,%rsp
757         popfq
758         pop     %r15
759         pop     %r14
760         pop     %r13
761         pop     %r12
762         pop     %rbp
763         pop     %rbx
764         pop     %rdi
765         pop     %rsi
766         ret
767 .size   se_handler,.-se_handler
768
769 .section        .pdata
770 .align  4
771         .rva    .LSEH_begin_gcm_gmult_4bit
772         .rva    .LSEH_end_gcm_gmult_4bit
773         .rva    .LSEH_info_gcm_gmult_4bit
774
775         .rva    .LSEH_begin_gcm_ghash_4bit
776         .rva    .LSEH_end_gcm_ghash_4bit
777         .rva    .LSEH_info_gcm_ghash_4bit
778
779         .rva    .LSEH_begin_gcm_ghash_clmul
780         .rva    .LSEH_end_gcm_ghash_clmul
781         .rva    .LSEH_info_gcm_ghash_clmul
782
783 .section        .xdata
784 .align  8
785 .LSEH_info_gcm_gmult_4bit:
786         .byte   9,0,0,0
787         .rva    se_handler
788         .rva    .Lgmult_prologue,.Lgmult_epilogue       # HandlerData
789 .LSEH_info_gcm_ghash_4bit:
790         .byte   9,0,0,0
791         .rva    se_handler
792         .rva    .Lghash_prologue,.Lghash_epilogue       # HandlerData
793 .LSEH_info_gcm_ghash_clmul:
794         .byte   0x01,0x1f,0x0b,0x00
795         .byte   0x1f,0xa8,0x04,0x00     #movaps 0x40(rsp),xmm10
796         .byte   0x19,0x98,0x03,0x00     #movaps 0x30(rsp),xmm9
797         .byte   0x13,0x88,0x02,0x00     #movaps 0x20(rsp),xmm8
798         .byte   0x0d,0x78,0x01,0x00     #movaps 0x10(rsp),xmm7
799         .byte   0x08,0x68,0x00,0x00     #movaps (rsp),xmm6
800         .byte   0x04,0xa2,0x00,0x00     #sub    rsp,0x58
801 ___
802 }
803 \f
804 $code =~ s/\`([^\`]*)\`/eval($1)/gem;
805
806 print $code;
807
808 close STDOUT;