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#
26# (*)	comparison is not completely fair, because C results are
27#	for vanilla "256B" implementation, while assembler results
28#	are for "528B";-)
29# (**)	it's mystery [to me] why Core2 result is not same as for
30#	Opteron;
31
32# May 2010
33#
34# Add PCLMULQDQ version performing at 2.02 cycles per processed byte.
35# See ghash-x86.pl for background information and details about coding
36# techniques.
37#
38# Special thanks to David Woodhouse <dwmw2@infradead.org> for
39# providing access to a Westmere-based system on behalf of Intel
40# Open Source Technology Centre.
41
42$flavour = shift;
43$output  = shift;
44if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
45
46$win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);
47
48$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
49( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
50( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or
51die "can't locate x86_64-xlate.pl";
52
53open OUT,"| \"$^X\" $xlate $flavour $output";
54*STDOUT=*OUT;
55
56# common register layout
57$nlo="%rax";
58$nhi="%rbx";
59$Zlo="%r8";
60$Zhi="%r9";
61$tmp="%r10";
62$rem_4bit = "%r11";
63
64$Xi="%rdi";
65$Htbl="%rsi";
66
67# per-function register layout
68$cnt="%rcx";
69$rem="%rdx";
70
71sub LB() { my $r=shift; $r =~ s/%[er]([a-d])x/%\1l/	or
72			$r =~ s/%[er]([sd]i)/%\1l/	or
73			$r =~ s/%[er](bp)/%\1l/		or
74			$r =~ s/%(r[0-9]+)[d]?/%\1b/;   $r; }
75
76sub AUTOLOAD()		# thunk [simplified] 32-bit style perlasm
77{ my $opcode = $AUTOLOAD; $opcode =~ s/.*:://;
78  my $arg = pop;
79    $arg = "\$$arg" if ($arg*1 eq $arg);
80    $code .= "\t$opcode\t".join(',',$arg,reverse @_)."\n";
81}
82
83{ my $N;
84  sub loop() {
85  my $inp = shift;
86
87	$N++;
88$code.=<<___;
89	xor	$nlo,$nlo
90	xor	$nhi,$nhi
91	mov	`&LB("$Zlo")`,`&LB("$nlo")`
92	mov	`&LB("$Zlo")`,`&LB("$nhi")`
93	shl	\$4,`&LB("$nlo")`
94	mov	\$14,$cnt
95	mov	8($Htbl,$nlo),$Zlo
96	mov	($Htbl,$nlo),$Zhi
97	and	\$0xf0,`&LB("$nhi")`
98	mov	$Zlo,$rem
99	jmp	.Loop$N
100
101.align	16
102.Loop$N:
103	shr	\$4,$Zlo
104	and	\$0xf,$rem
105	mov	$Zhi,$tmp
106	mov	($inp,$cnt),`&LB("$nlo")`
107	shr	\$4,$Zhi
108	xor	8($Htbl,$nhi),$Zlo
109	shl	\$60,$tmp
110	xor	($Htbl,$nhi),$Zhi
111	mov	`&LB("$nlo")`,`&LB("$nhi")`
112	xor	($rem_4bit,$rem,8),$Zhi
113	mov	$Zlo,$rem
114	shl	\$4,`&LB("$nlo")`
115	xor	$tmp,$Zlo
116	dec	$cnt
117	js	.Lbreak$N
118
119	shr	\$4,$Zlo
120	and	\$0xf,$rem
121	mov	$Zhi,$tmp
122	shr	\$4,$Zhi
123	xor	8($Htbl,$nlo),$Zlo
124	shl	\$60,$tmp
125	xor	($Htbl,$nlo),$Zhi
126	and	\$0xf0,`&LB("$nhi")`
127	xor	($rem_4bit,$rem,8),$Zhi
128	mov	$Zlo,$rem
129	xor	$tmp,$Zlo
130	jmp	.Loop$N
131
132.align	16
133.Lbreak$N:
134	shr	\$4,$Zlo
135	and	\$0xf,$rem
136	mov	$Zhi,$tmp
137	shr	\$4,$Zhi
138	xor	8($Htbl,$nlo),$Zlo
139	shl	\$60,$tmp
140	xor	($Htbl,$nlo),$Zhi
141	and	\$0xf0,`&LB("$nhi")`
142	xor	($rem_4bit,$rem,8),$Zhi
143	mov	$Zlo,$rem
144	xor	$tmp,$Zlo
145
146	shr	\$4,$Zlo
147	and	\$0xf,$rem
148	mov	$Zhi,$tmp
149	shr	\$4,$Zhi
150	xor	8($Htbl,$nhi),$Zlo
151	shl	\$60,$tmp
152	xor	($Htbl,$nhi),$Zhi
153	xor	$tmp,$Zlo
154	xor	($rem_4bit,$rem,8),$Zhi
155
156	bswap	$Zlo
157	bswap	$Zhi
158___
159}}
160
161$code=<<___;
162.text
163
164.globl	gcm_gmult_4bit
165.type	gcm_gmult_4bit,\@function,2
166.align	16
167gcm_gmult_4bit:
168	push	%rbx
169	push	%rbp		# %rbp and %r12 are pushed exclusively in
170	push	%r12		# order to reuse Win64 exception handler...
171.Lgmult_prologue:
172
173	movzb	15($Xi),$Zlo
174	lea	.Lrem_4bit(%rip),$rem_4bit
175___
176	&loop	($Xi);
177$code.=<<___;
178	mov	$Zlo,8($Xi)
179	mov	$Zhi,($Xi)
180
181	mov	16(%rsp),%rbx
182	lea	24(%rsp),%rsp
183.Lgmult_epilogue:
184	ret
185.size	gcm_gmult_4bit,.-gcm_gmult_4bit
186___
187
188# per-function register layout
189$inp="%rdx";
190$len="%rcx";
191$rem_8bit=$rem_4bit;
192
193$code.=<<___;
194.globl	gcm_ghash_4bit
195.type	gcm_ghash_4bit,\@function,4
196.align	16
197gcm_ghash_4bit:
198	push	%rbx
199	push	%rbp
200	push	%r12
201	push	%r13
202	push	%r14
203	push	%r15
204	sub	\$280,%rsp
205.Lghash_prologue:
206	mov	$inp,%r14		# reassign couple of args
207	mov	$len,%r15
208___
209{ my $inp="%r14";
210  my $dat="%edx";
211  my $len="%r15";
212  my @nhi=("%ebx","%ecx");
213  my @rem=("%r12","%r13");
214  my $Hshr4="%rbp";
215
216	&sub	($Htbl,-128);		# size optimization
217	&lea	($Hshr4,"16+128(%rsp)");
218	{ my @lo =($nlo,$nhi);
219          my @hi =($Zlo,$Zhi);
220
221	  &xor	($dat,$dat);
222	  for ($i=0,$j=-2;$i<18;$i++,$j++) {
223	    &mov	("$j(%rsp)",&LB($dat))		if ($i>1);
224	    &or		($lo[0],$tmp)			if ($i>1);
225	    &mov	(&LB($dat),&LB($lo[1]))		if ($i>0 && $i<17);
226	    &shr	($lo[1],4)			if ($i>0 && $i<17);
227	    &mov	($tmp,$hi[1])			if ($i>0 && $i<17);
228	    &shr	($hi[1],4)			if ($i>0 && $i<17);
229	    &mov	("8*$j($Hshr4)",$hi[0])		if ($i>1);
230	    &mov	($hi[0],"16*$i+0-128($Htbl)")	if ($i<16);
231	    &shl	(&LB($dat),4)			if ($i>0 && $i<17);
232	    &mov	("8*$j-128($Hshr4)",$lo[0])	if ($i>1);
233	    &mov	($lo[0],"16*$i+8-128($Htbl)")	if ($i<16);
234	    &shl	($tmp,60)			if ($i>0 && $i<17);
235
236	    push	(@lo,shift(@lo));
237	    push	(@hi,shift(@hi));
238	  }
239	}
240	&add	($Htbl,-128);
241	&mov	($Zlo,"8($Xi)");
242	&mov	($Zhi,"0($Xi)");
243	&add	($len,$inp);		# pointer to the end of data
244	&lea	($rem_8bit,".Lrem_8bit(%rip)");
245	&jmp	(".Louter_loop");
246
247$code.=".align	16\n.Louter_loop:\n";
248	&xor	($Zhi,"($inp)");
249	&mov	("%rdx","8($inp)");
250	&lea	($inp,"16($inp)");
251	&xor	("%rdx",$Zlo);
252	&mov	("($Xi)",$Zhi);
253	&mov	("8($Xi)","%rdx");
254	&shr	("%rdx",32);
255
256	&xor	($nlo,$nlo);
257	&rol	($dat,8);
258	&mov	(&LB($nlo),&LB($dat));
259	&movz	($nhi[0],&LB($dat));
260	&shl	(&LB($nlo),4);
261	&shr	($nhi[0],4);
262
263	for ($j=11,$i=0;$i<15;$i++) {
264	    &rol	($dat,8);
265	    &xor	($Zlo,"8($Htbl,$nlo)")			if ($i>0);
266	    &xor	($Zhi,"($Htbl,$nlo)")			if ($i>0);
267	    &mov	($Zlo,"8($Htbl,$nlo)")			if ($i==0);
268	    &mov	($Zhi,"($Htbl,$nlo)")			if ($i==0);
269
270	    &mov	(&LB($nlo),&LB($dat));
271	    &xor	($Zlo,$tmp)				if ($i>0);
272	    &movzw	($rem[1],"($rem_8bit,$rem[1],2)")	if ($i>0);
273
274	    &movz	($nhi[1],&LB($dat));
275	    &shl	(&LB($nlo),4);
276	    &movzb	($rem[0],"(%rsp,$nhi[0])");
277
278	    &shr	($nhi[1],4)				if ($i<14);
279	    &and	($nhi[1],0xf0)				if ($i==14);
280	    &shl	($rem[1],48)				if ($i>0);
281	    &xor	($rem[0],$Zlo);
282
283	    &mov	($tmp,$Zhi);
284	    &xor	($Zhi,$rem[1])				if ($i>0);
285	    &shr	($Zlo,8);
286
287	    &movz	($rem[0],&LB($rem[0]));
288	    &mov	($dat,"$j($Xi)")			if (--$j%4==0);
289	    &shr	($Zhi,8);
290
291	    &xor	($Zlo,"-128($Hshr4,$nhi[0],8)");
292	    &shl	($tmp,56);
293	    &xor	($Zhi,"($Hshr4,$nhi[0],8)");
294
295	    unshift	(@nhi,pop(@nhi));		# "rotate" registers
296	    unshift	(@rem,pop(@rem));
297	}
298	&movzw	($rem[1],"($rem_8bit,$rem[1],2)");
299	&xor	($Zlo,"8($Htbl,$nlo)");
300	&xor	($Zhi,"($Htbl,$nlo)");
301
302	&shl	($rem[1],48);
303	&xor	($Zlo,$tmp);
304
305	&xor	($Zhi,$rem[1]);
306	&movz	($rem[0],&LB($Zlo));
307	&shr	($Zlo,4);
308
309	&mov	($tmp,$Zhi);
310	&shl	(&LB($rem[0]),4);
311	&shr	($Zhi,4);
312
313	&xor	($Zlo,"8($Htbl,$nhi[0])");
314	&movzw	($rem[0],"($rem_8bit,$rem[0],2)");
315	&shl	($tmp,60);
316
317	&xor	($Zhi,"($Htbl,$nhi[0])");
318	&xor	($Zlo,$tmp);
319	&shl	($rem[0],48);
320
321	&bswap	($Zlo);
322	&xor	($Zhi,$rem[0]);
323
324	&bswap	($Zhi);
325	&cmp	($inp,$len);
326	&jb	(".Louter_loop");
327}
328$code.=<<___;
329	mov	$Zlo,8($Xi)
330	mov	$Zhi,($Xi)
331
332	lea	280(%rsp),%rsi
333	mov	0(%rsi),%r15
334	mov	8(%rsi),%r14
335	mov	16(%rsi),%r13
336	mov	24(%rsi),%r12
337	mov	32(%rsi),%rbp
338	mov	40(%rsi),%rbx
339	lea	48(%rsi),%rsp
340.Lghash_epilogue:
341	ret
342.size	gcm_ghash_4bit,.-gcm_ghash_4bit
343___
344
345######################################################################
346# PCLMULQDQ version.
347
348@_4args=$win64?	("%rcx","%rdx","%r8", "%r9") :	# Win64 order
349		("%rdi","%rsi","%rdx","%rcx");	# Unix order
350
351($Xi,$Xhi)=("%xmm0","%xmm1");	$Hkey="%xmm2";
352($T1,$T2,$T3)=("%xmm3","%xmm4","%xmm5");
353
354sub clmul64x64_T2 {	# minimal register pressure
355my ($Xhi,$Xi,$Hkey,$modulo)=@_;
356
357$code.=<<___ if (!defined($modulo));
358	movdqa		$Xi,$Xhi		#
359	pshufd		\$0b01001110,$Xi,$T1
360	pshufd		\$0b01001110,$Hkey,$T2
361	pxor		$Xi,$T1			#
362	pxor		$Hkey,$T2
363___
364$code.=<<___;
365	pclmulqdq	\$0x00,$Hkey,$Xi	#######
366	pclmulqdq	\$0x11,$Hkey,$Xhi	#######
367	pclmulqdq	\$0x00,$T2,$T1		#######
368	pxor		$Xi,$T1			#
369	pxor		$Xhi,$T1		#
370
371	movdqa		$T1,$T2			#
372	psrldq		\$8,$T1
373	pslldq		\$8,$T2			#
374	pxor		$T1,$Xhi
375	pxor		$T2,$Xi			#
376___
377}
378
379sub reduction_alg9 {	# 17/13 times faster than Intel version
380my ($Xhi,$Xi) = @_;
381
382$code.=<<___;
383	# 1st phase
384	movdqa		$Xi,$T1			#
385	psllq		\$1,$Xi
386	pxor		$T1,$Xi			#
387	psllq		\$5,$Xi			#
388	pxor		$T1,$Xi			#
389	psllq		\$57,$Xi		#
390	movdqa		$Xi,$T2			#
391	pslldq		\$8,$Xi
392	psrldq		\$8,$T2			#
393	pxor		$T1,$Xi
394	pxor		$T2,$Xhi		#
395
396	# 2nd phase
397	movdqa		$Xi,$T2
398	psrlq		\$5,$Xi
399	pxor		$T2,$Xi			#
400	psrlq		\$1,$Xi			#
401	pxor		$T2,$Xi			#
402	pxor		$Xhi,$T2
403	psrlq		\$1,$Xi			#
404	pxor		$T2,$Xi			#
405___
406}
407
408{ my ($Htbl,$Xip)=@_4args;
409
410$code.=<<___;
411.globl	gcm_init_clmul
412.type	gcm_init_clmul,\@abi-omnipotent
413.align	16
414gcm_init_clmul:
415	endbr64
416	movdqu		($Xip),$Hkey
417	pshufd		\$0b01001110,$Hkey,$Hkey	# dword swap
418
419	# <<1 twist
420	pshufd		\$0b11111111,$Hkey,$T2	# broadcast uppermost dword
421	movdqa		$Hkey,$T1
422	psllq		\$1,$Hkey
423	pxor		$T3,$T3			#
424	psrlq		\$63,$T1
425	pcmpgtd		$T2,$T3			# broadcast carry bit
426	pslldq		\$8,$T1
427	por		$T1,$Hkey		# H<<=1
428
429	# magic reduction
430	pand		.L0x1c2_polynomial(%rip),$T3
431	pxor		$T3,$Hkey		# if(carry) H^=0x1c2_polynomial
432
433	# calculate H^2
434	movdqa		$Hkey,$Xi
435___
436	&clmul64x64_T2	($Xhi,$Xi,$Hkey);
437	&reduction_alg9	($Xhi,$Xi);
438$code.=<<___;
439	movdqu		$Hkey,($Htbl)		# save H
440	movdqu		$Xi,16($Htbl)		# save H^2
441	ret
442.size	gcm_init_clmul,.-gcm_init_clmul
443___
444}
445
446{ my ($Xip,$Htbl)=@_4args;
447
448$code.=<<___;
449.globl	gcm_gmult_clmul
450.type	gcm_gmult_clmul,\@abi-omnipotent
451.align	16
452gcm_gmult_clmul:
453	endbr64
454	movdqu		($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
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
480gcm_ghash_clmul:
481	endbr64
482___
483$code.=<<___ if ($win64);
484.LSEH_begin_gcm_ghash_clmul:
485	# I can't trust assembler to use specific encoding:-(
486	.byte	0x48,0x83,0xec,0x58		#sub	\$0x58,%rsp
487	.byte	0x0f,0x29,0x34,0x24		#movaps	%xmm6,(%rsp)
488	.byte	0x0f,0x29,0x7c,0x24,0x10	#movdqa	%xmm7,0x10(%rsp)
489	.byte	0x44,0x0f,0x29,0x44,0x24,0x20	#movaps	%xmm8,0x20(%rsp)
490	.byte	0x44,0x0f,0x29,0x4c,0x24,0x30	#movaps	%xmm9,0x30(%rsp)
491	.byte	0x44,0x0f,0x29,0x54,0x24,0x40	#movaps	%xmm10,0x40(%rsp)
492___
493$code.=<<___;
494	movdqa		.Lbswap_mask(%rip),$T3
495
496	movdqu		($Xip),$Xi
497	movdqu		($Htbl),$Hkey
498	pshufb		$T3,$Xi
499
500	sub		\$0x10,$len
501	jz		.Lodd_tail
502
503	movdqu		16($Htbl),$Hkey2
504	#######
505	# Xi+2 =[H*(Ii+1 + Xi+1)] mod P =
506	#	[(H*Ii+1) + (H*Xi+1)] mod P =
507	#	[(H*Ii+1) + H^2*(Ii+Xi)] mod P
508	#
509	movdqu		($inp),$T1		# Ii
510	movdqu		16($inp),$Xn		# Ii+1
511	pshufb		$T3,$T1
512	pshufb		$T3,$Xn
513	pxor		$T1,$Xi			# Ii+Xi
514___
515	&clmul64x64_T2	($Xhn,$Xn,$Hkey);	# H*Ii+1
516$code.=<<___;
517	movdqa		$Xi,$Xhi		#
518	pshufd		\$0b01001110,$Xi,$T1
519	pshufd		\$0b01001110,$Hkey2,$T2
520	pxor		$Xi,$T1			#
521	pxor		$Hkey2,$T2
522
523	lea		32($inp),$inp		# i+=2
524	sub		\$0x20,$len
525	jbe		.Leven_tail
526
527.Lmod_loop:
528___
529	&clmul64x64_T2	($Xhi,$Xi,$Hkey2,1);	# H^2*(Ii+Xi)
530$code.=<<___;
531	movdqu		($inp),$T1		# Ii
532	pxor		$Xn,$Xi			# (H*Ii+1) + H^2*(Ii+Xi)
533	pxor		$Xhn,$Xhi
534
535	movdqu		16($inp),$Xn		# Ii+1
536	pshufb		$T3,$T1
537	pshufb		$T3,$Xn
538
539	movdqa		$Xn,$Xhn		#
540	pshufd		\$0b01001110,$Xn,$T1n
541	pshufd		\$0b01001110,$Hkey,$T2n
542	pxor		$Xn,$T1n		#
543	pxor		$Hkey,$T2n
544	 pxor		$T1,$Xhi		# "Ii+Xi", consume early
545
546	  movdqa	$Xi,$T1			# 1st phase
547	  psllq		\$1,$Xi
548	  pxor		$T1,$Xi			#
549	  psllq		\$5,$Xi			#
550	  pxor		$T1,$Xi			#
551	pclmulqdq	\$0x00,$Hkey,$Xn	#######
552	  psllq		\$57,$Xi		#
553	  movdqa	$Xi,$T2			#
554	  pslldq	\$8,$Xi
555	  psrldq	\$8,$T2			#
556	  pxor		$T1,$Xi
557	  pxor		$T2,$Xhi		#
558
559	pclmulqdq	\$0x11,$Hkey,$Xhn	#######
560	  movdqa	$Xi,$T2			# 2nd phase
561	  psrlq		\$5,$Xi
562	  pxor		$T2,$Xi			#
563	  psrlq		\$1,$Xi			#
564	  pxor		$T2,$Xi			#
565	  pxor		$Xhi,$T2
566	  psrlq		\$1,$Xi			#
567	  pxor		$T2,$Xi			#
568
569	pclmulqdq	\$0x00,$T2n,$T1n	#######
570	 movdqa		$Xi,$Xhi		#
571	 pshufd		\$0b01001110,$Xi,$T1
572	 pshufd		\$0b01001110,$Hkey2,$T2
573	 pxor		$Xi,$T1			#
574	 pxor		$Hkey2,$T2
575
576	pxor		$Xn,$T1n		#
577	pxor		$Xhn,$T1n		#
578	movdqa		$T1n,$T2n		#
579	psrldq		\$8,$T1n
580	pslldq		\$8,$T2n		#
581	pxor		$T1n,$Xhn
582	pxor		$T2n,$Xn		#
583
584	lea		32($inp),$inp
585	sub		\$0x20,$len
586	ja		.Lmod_loop
587
588.Leven_tail:
589___
590	&clmul64x64_T2	($Xhi,$Xi,$Hkey2,1);	# H^2*(Ii+Xi)
591$code.=<<___;
592	pxor		$Xn,$Xi			# (H*Ii+1) + H^2*(Ii+Xi)
593	pxor		$Xhn,$Xhi
594___
595	&reduction_alg9	($Xhi,$Xi);
596$code.=<<___;
597	test		$len,$len
598	jnz		.Ldone
599
600.Lodd_tail:
601	movdqu		($inp),$T1		# Ii
602	pshufb		$T3,$T1
603	pxor		$T1,$Xi			# Ii+Xi
604___
605	&clmul64x64_T2	($Xhi,$Xi,$Hkey);	# H*(Ii+Xi)
606	&reduction_alg9	($Xhi,$Xi);
607$code.=<<___;
608.Ldone:
609	pshufb		$T3,$Xi
610	movdqu		$Xi,($Xip)
611___
612$code.=<<___ if ($win64);
613	movaps	(%rsp),%xmm6
614	movaps	0x10(%rsp),%xmm7
615	movaps	0x20(%rsp),%xmm8
616	movaps	0x30(%rsp),%xmm9
617	movaps	0x40(%rsp),%xmm10
618	add	\$0x58,%rsp
619___
620$code.=<<___;
621	ret
622.LSEH_end_gcm_ghash_clmul:
623.size	gcm_ghash_clmul,.-gcm_ghash_clmul
624___
625}
626
627$code.=<<___;
628.section .rodata
629.align	64
630.Lbswap_mask:
631	.byte	15,14,13,12,11,10,9,8,7,6,5,4,3,2,1,0
632.L0x1c2_polynomial:
633	.byte	1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0xc2
634.align	64
635.type	.Lrem_4bit,\@object
636.Lrem_4bit:
637	.long	0,`0x0000<<16`,0,`0x1C20<<16`,0,`0x3840<<16`,0,`0x2460<<16`
638	.long	0,`0x7080<<16`,0,`0x6CA0<<16`,0,`0x48C0<<16`,0,`0x54E0<<16`
639	.long	0,`0xE100<<16`,0,`0xFD20<<16`,0,`0xD940<<16`,0,`0xC560<<16`
640	.long	0,`0x9180<<16`,0,`0x8DA0<<16`,0,`0xA9C0<<16`,0,`0xB5E0<<16`
641.type	.Lrem_8bit,\@object
642.Lrem_8bit:
643	.value	0x0000,0x01C2,0x0384,0x0246,0x0708,0x06CA,0x048C,0x054E
644	.value	0x0E10,0x0FD2,0x0D94,0x0C56,0x0918,0x08DA,0x0A9C,0x0B5E
645	.value	0x1C20,0x1DE2,0x1FA4,0x1E66,0x1B28,0x1AEA,0x18AC,0x196E
646	.value	0x1230,0x13F2,0x11B4,0x1076,0x1538,0x14FA,0x16BC,0x177E
647	.value	0x3840,0x3982,0x3BC4,0x3A06,0x3F48,0x3E8A,0x3CCC,0x3D0E
648	.value	0x3650,0x3792,0x35D4,0x3416,0x3158,0x309A,0x32DC,0x331E
649	.value	0x2460,0x25A2,0x27E4,0x2626,0x2368,0x22AA,0x20EC,0x212E
650	.value	0x2A70,0x2BB2,0x29F4,0x2836,0x2D78,0x2CBA,0x2EFC,0x2F3E
651	.value	0x7080,0x7142,0x7304,0x72C6,0x7788,0x764A,0x740C,0x75CE
652	.value	0x7E90,0x7F52,0x7D14,0x7CD6,0x7998,0x785A,0x7A1C,0x7BDE
653	.value	0x6CA0,0x6D62,0x6F24,0x6EE6,0x6BA8,0x6A6A,0x682C,0x69EE
654	.value	0x62B0,0x6372,0x6134,0x60F6,0x65B8,0x647A,0x663C,0x67FE
655	.value	0x48C0,0x4902,0x4B44,0x4A86,0x4FC8,0x4E0A,0x4C4C,0x4D8E
656	.value	0x46D0,0x4712,0x4554,0x4496,0x41D8,0x401A,0x425C,0x439E
657	.value	0x54E0,0x5522,0x5764,0x56A6,0x53E8,0x522A,0x506C,0x51AE
658	.value	0x5AF0,0x5B32,0x5974,0x58B6,0x5DF8,0x5C3A,0x5E7C,0x5FBE
659	.value	0xE100,0xE0C2,0xE284,0xE346,0xE608,0xE7CA,0xE58C,0xE44E
660	.value	0xEF10,0xEED2,0xEC94,0xED56,0xE818,0xE9DA,0xEB9C,0xEA5E
661	.value	0xFD20,0xFCE2,0xFEA4,0xFF66,0xFA28,0xFBEA,0xF9AC,0xF86E
662	.value	0xF330,0xF2F2,0xF0B4,0xF176,0xF438,0xF5FA,0xF7BC,0xF67E
663	.value	0xD940,0xD882,0xDAC4,0xDB06,0xDE48,0xDF8A,0xDDCC,0xDC0E
664	.value	0xD750,0xD692,0xD4D4,0xD516,0xD058,0xD19A,0xD3DC,0xD21E
665	.value	0xC560,0xC4A2,0xC6E4,0xC726,0xC268,0xC3AA,0xC1EC,0xC02E
666	.value	0xCB70,0xCAB2,0xC8F4,0xC936,0xCC78,0xCDBA,0xCFFC,0xCE3E
667	.value	0x9180,0x9042,0x9204,0x93C6,0x9688,0x974A,0x950C,0x94CE
668	.value	0x9F90,0x9E52,0x9C14,0x9DD6,0x9898,0x995A,0x9B1C,0x9ADE
669	.value	0x8DA0,0x8C62,0x8E24,0x8FE6,0x8AA8,0x8B6A,0x892C,0x88EE
670	.value	0x83B0,0x8272,0x8034,0x81F6,0x84B8,0x857A,0x873C,0x86FE
671	.value	0xA9C0,0xA802,0xAA44,0xAB86,0xAEC8,0xAF0A,0xAD4C,0xAC8E
672	.value	0xA7D0,0xA612,0xA454,0xA596,0xA0D8,0xA11A,0xA35C,0xA29E
673	.value	0xB5E0,0xB422,0xB664,0xB7A6,0xB2E8,0xB32A,0xB16C,0xB0AE
674	.value	0xBBF0,0xBA32,0xB874,0xB9B6,0xBCF8,0xBD3A,0xBF7C,0xBEBE
675.align	64
676.text
677___
678
679# EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
680#		CONTEXT *context,DISPATCHER_CONTEXT *disp)
681if ($win64) {
682$rec="%rcx";
683$frame="%rdx";
684$context="%r8";
685$disp="%r9";
686
687$code.=<<___;
688.extern	__imp_RtlVirtualUnwind
689.type	se_handler,\@abi-omnipotent
690.align	16
691se_handler:
692	endbr64
693	push	%rsi
694	push	%rdi
695	push	%rbx
696	push	%rbp
697	push	%r12
698	push	%r13
699	push	%r14
700	push	%r15
701	pushfq
702	sub	\$64,%rsp
703
704	mov	120($context),%rax	# pull context->Rax
705	mov	248($context),%rbx	# pull context->Rip
706
707	mov	8($disp),%rsi		# disp->ImageBase
708	mov	56($disp),%r11		# disp->HandlerData
709
710	mov	0(%r11),%r10d		# HandlerData[0]
711	lea	(%rsi,%r10),%r10	# prologue label
712	cmp	%r10,%rbx		# context->Rip<prologue label
713	jb	.Lin_prologue
714
715	mov	152($context),%rax	# pull context->Rsp
716
717	mov	4(%r11),%r10d		# HandlerData[1]
718	lea	(%rsi,%r10),%r10	# epilogue label
719	cmp	%r10,%rbx		# context->Rip>=epilogue label
720	jae	.Lin_prologue
721
722	lea	24(%rax),%rax		# adjust "rsp"
723
724	mov	-8(%rax),%rbx
725	mov	-16(%rax),%rbp
726	mov	-24(%rax),%r12
727	mov	%rbx,144($context)	# restore context->Rbx
728	mov	%rbp,160($context)	# restore context->Rbp
729	mov	%r12,216($context)	# restore context->R12
730
731.Lin_prologue:
732	mov	8(%rax),%rdi
733	mov	16(%rax),%rsi
734	mov	%rax,152($context)	# restore context->Rsp
735	mov	%rsi,168($context)	# restore context->Rsi
736	mov	%rdi,176($context)	# restore context->Rdi
737
738	mov	40($disp),%rdi		# disp->ContextRecord
739	mov	$context,%rsi		# context
740	mov	\$`1232/8`,%ecx		# sizeof(CONTEXT)
741	.long	0xa548f3fc		# cld; rep movsq
742
743	mov	$disp,%rsi
744	xor	%rcx,%rcx		# arg1, UNW_FLAG_NHANDLER
745	mov	8(%rsi),%rdx		# arg2, disp->ImageBase
746	mov	0(%rsi),%r8		# arg3, disp->ControlPc
747	mov	16(%rsi),%r9		# arg4, disp->FunctionEntry
748	mov	40(%rsi),%r10		# disp->ContextRecord
749	lea	56(%rsi),%r11		# &disp->HandlerData
750	lea	24(%rsi),%r12		# &disp->EstablisherFrame
751	mov	%r10,32(%rsp)		# arg5
752	mov	%r11,40(%rsp)		# arg6
753	mov	%r12,48(%rsp)		# arg7
754	mov	%rcx,56(%rsp)		# arg8, (NULL)
755	call	*__imp_RtlVirtualUnwind(%rip)
756
757	mov	\$1,%eax		# ExceptionContinueSearch
758	add	\$64,%rsp
759	popfq
760	pop	%r15
761	pop	%r14
762	pop	%r13
763	pop	%r12
764	pop	%rbp
765	pop	%rbx
766	pop	%rdi
767	pop	%rsi
768	ret
769.size	se_handler,.-se_handler
770
771.section	.pdata
772.align	4
773	.rva	.LSEH_begin_gcm_gmult_4bit
774	.rva	.LSEH_end_gcm_gmult_4bit
775	.rva	.LSEH_info_gcm_gmult_4bit
776
777	.rva	.LSEH_begin_gcm_ghash_4bit
778	.rva	.LSEH_end_gcm_ghash_4bit
779	.rva	.LSEH_info_gcm_ghash_4bit
780
781	.rva	.LSEH_begin_gcm_ghash_clmul
782	.rva	.LSEH_end_gcm_ghash_clmul
783	.rva	.LSEH_info_gcm_ghash_clmul
784
785.section	.xdata
786.align	8
787.LSEH_info_gcm_gmult_4bit:
788	.byte	9,0,0,0
789	.rva	se_handler
790	.rva	.Lgmult_prologue,.Lgmult_epilogue	# HandlerData
791.LSEH_info_gcm_ghash_4bit:
792	.byte	9,0,0,0
793	.rva	se_handler
794	.rva	.Lghash_prologue,.Lghash_epilogue	# HandlerData
795.LSEH_info_gcm_ghash_clmul:
796	.byte	0x01,0x1f,0x0b,0x00
797	.byte	0x1f,0xa8,0x04,0x00	#movaps 0x40(rsp),xmm10
798	.byte	0x19,0x98,0x03,0x00	#movaps 0x30(rsp),xmm9
799	.byte	0x13,0x88,0x02,0x00	#movaps 0x20(rsp),xmm8
800	.byte	0x0d,0x78,0x01,0x00	#movaps 0x10(rsp),xmm7
801	.byte	0x08,0x68,0x00,0x00	#movaps (rsp),xmm6
802	.byte	0x04,0xa2,0x00,0x00	#sub	rsp,0x58
803___
804}
805
806$code =~ s/\`([^\`]*)\`/eval($1)/gem;
807
808print $code;
809
810close STDOUT;
811