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	_CET_ENDBR
169	push	%rbx
170	push	%rbp		# %rbp and %r12 are pushed exclusively in
171	push	%r12		# order to reuse Win64 exception handler...
172.Lgmult_prologue:
173
174	movzb	15($Xi),$Zlo
175	lea	.Lrem_4bit(%rip),$rem_4bit
176___
177	&loop	($Xi);
178$code.=<<___;
179	mov	$Zlo,8($Xi)
180	mov	$Zhi,($Xi)
181
182	mov	16(%rsp),%rbx
183	lea	24(%rsp),%rsp
184.Lgmult_epilogue:
185	ret
186.size	gcm_gmult_4bit,.-gcm_gmult_4bit
187___
188
189# per-function register layout
190$inp="%rdx";
191$len="%rcx";
192$rem_8bit=$rem_4bit;
193
194$code.=<<___;
195.globl	gcm_ghash_4bit
196.type	gcm_ghash_4bit,\@function,4
197.align	16
198gcm_ghash_4bit:
199	_CET_ENDBR
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 && $j>=0);
291	    &shr	($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
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
356sub clmul64x64_T2 {	# minimal register pressure
357my ($Xhi,$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
381sub reduction_alg9 {	# 17/13 times faster than Intel version
382my ($Xhi,$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
410{ my ($Htbl,$Xip)=@_4args;
411
412$code.=<<___;
413.globl	gcm_init_clmul
414.type	gcm_init_clmul,\@abi-omnipotent
415.align	16
416gcm_init_clmul:
417	_CET_ENDBR
418	movdqu		($Xip),$Hkey
419	pshufd		\$0b01001110,$Hkey,$Hkey	# dword swap
420
421	# <<1 twist
422	pshufd		\$0b11111111,$Hkey,$T2	# broadcast uppermost dword
423	movdqa		$Hkey,$T1
424	psllq		\$1,$Hkey
425	pxor		$T3,$T3			#
426	psrlq		\$63,$T1
427	pcmpgtd		$T2,$T3			# broadcast carry bit
428	pslldq		\$8,$T1
429	por		$T1,$Hkey		# H<<=1
430
431	# magic reduction
432	pand		.L0x1c2_polynomial(%rip),$T3
433	pxor		$T3,$Hkey		# if(carry) H^=0x1c2_polynomial
434
435	# calculate H^2
436	movdqa		$Hkey,$Xi
437___
438	&clmul64x64_T2	($Xhi,$Xi,$Hkey);
439	&reduction_alg9	($Xhi,$Xi);
440$code.=<<___;
441	movdqu		$Hkey,($Htbl)		# save H
442	movdqu		$Xi,16($Htbl)		# save H^2
443	ret
444.size	gcm_init_clmul,.-gcm_init_clmul
445___
446}
447
448{ my ($Xip,$Htbl)=@_4args;
449
450$code.=<<___;
451.globl	gcm_gmult_clmul
452.type	gcm_gmult_clmul,\@abi-omnipotent
453.align	16
454gcm_gmult_clmul:
455	_CET_ENDBR
456	movdqu		($Xip),$Xi
457	movdqa		.Lbswap_mask(%rip),$T3
458	movdqu		($Htbl),$Hkey
459	pshufb		$T3,$Xi
460___
461	&clmul64x64_T2	($Xhi,$Xi,$Hkey);
462	&reduction_alg9	($Xhi,$Xi);
463$code.=<<___;
464	pshufb		$T3,$Xi
465	movdqu		$Xi,($Xip)
466	ret
467.size	gcm_gmult_clmul,.-gcm_gmult_clmul
468___
469}
470
471{ my ($Xip,$Htbl,$inp,$len)=@_4args;
472  my $Xn="%xmm6";
473  my $Xhn="%xmm7";
474  my $Hkey2="%xmm8";
475  my $T1n="%xmm9";
476  my $T2n="%xmm10";
477
478$code.=<<___;
479.globl	gcm_ghash_clmul
480.type	gcm_ghash_clmul,\@abi-omnipotent
481.align	16
482gcm_ghash_clmul:
483	_CET_ENDBR
484___
485$code.=<<___ if ($win64);
486.LSEH_begin_gcm_ghash_clmul:
487	# I can't trust assembler to use specific encoding:-(
488	.byte	0x48,0x83,0xec,0x58		#sub	\$0x58,%rsp
489	.byte	0x0f,0x29,0x34,0x24		#movaps	%xmm6,(%rsp)
490	.byte	0x0f,0x29,0x7c,0x24,0x10	#movdqa	%xmm7,0x10(%rsp)
491	.byte	0x44,0x0f,0x29,0x44,0x24,0x20	#movaps	%xmm8,0x20(%rsp)
492	.byte	0x44,0x0f,0x29,0x4c,0x24,0x30	#movaps	%xmm9,0x30(%rsp)
493	.byte	0x44,0x0f,0x29,0x54,0x24,0x40	#movaps	%xmm10,0x40(%rsp)
494___
495$code.=<<___;
496	movdqa		.Lbswap_mask(%rip),$T3
497
498	movdqu		($Xip),$Xi
499	movdqu		($Htbl),$Hkey
500	pshufb		$T3,$Xi
501
502	sub		\$0x10,$len
503	jz		.Lodd_tail
504
505	movdqu		16($Htbl),$Hkey2
506	#######
507	# Xi+2 =[H*(Ii+1 + Xi+1)] mod P =
508	#	[(H*Ii+1) + (H*Xi+1)] mod P =
509	#	[(H*Ii+1) + H^2*(Ii+Xi)] mod P
510	#
511	movdqu		($inp),$T1		# Ii
512	movdqu		16($inp),$Xn		# Ii+1
513	pshufb		$T3,$T1
514	pshufb		$T3,$Xn
515	pxor		$T1,$Xi			# Ii+Xi
516___
517	&clmul64x64_T2	($Xhn,$Xn,$Hkey);	# H*Ii+1
518$code.=<<___;
519	movdqa		$Xi,$Xhi		#
520	pshufd		\$0b01001110,$Xi,$T1
521	pshufd		\$0b01001110,$Hkey2,$T2
522	pxor		$Xi,$T1			#
523	pxor		$Hkey2,$T2
524
525	lea		32($inp),$inp		# i+=2
526	sub		\$0x20,$len
527	jbe		.Leven_tail
528
529.Lmod_loop:
530___
531	&clmul64x64_T2	($Xhi,$Xi,$Hkey2,1);	# H^2*(Ii+Xi)
532$code.=<<___;
533	movdqu		($inp),$T1		# Ii
534	pxor		$Xn,$Xi			# (H*Ii+1) + H^2*(Ii+Xi)
535	pxor		$Xhn,$Xhi
536
537	movdqu		16($inp),$Xn		# Ii+1
538	pshufb		$T3,$T1
539	pshufb		$T3,$Xn
540
541	movdqa		$Xn,$Xhn		#
542	pshufd		\$0b01001110,$Xn,$T1n
543	pshufd		\$0b01001110,$Hkey,$T2n
544	pxor		$Xn,$T1n		#
545	pxor		$Hkey,$T2n
546	 pxor		$T1,$Xhi		# "Ii+Xi", consume early
547
548	  movdqa	$Xi,$T1			# 1st phase
549	  psllq		\$1,$Xi
550	  pxor		$T1,$Xi			#
551	  psllq		\$5,$Xi			#
552	  pxor		$T1,$Xi			#
553	pclmulqdq	\$0x00,$Hkey,$Xn	#######
554	  psllq		\$57,$Xi		#
555	  movdqa	$Xi,$T2			#
556	  pslldq	\$8,$Xi
557	  psrldq	\$8,$T2			#
558	  pxor		$T1,$Xi
559	  pxor		$T2,$Xhi		#
560
561	pclmulqdq	\$0x11,$Hkey,$Xhn	#######
562	  movdqa	$Xi,$T2			# 2nd phase
563	  psrlq		\$5,$Xi
564	  pxor		$T2,$Xi			#
565	  psrlq		\$1,$Xi			#
566	  pxor		$T2,$Xi			#
567	  pxor		$Xhi,$T2
568	  psrlq		\$1,$Xi			#
569	  pxor		$T2,$Xi			#
570
571	pclmulqdq	\$0x00,$T2n,$T1n	#######
572	 movdqa		$Xi,$Xhi		#
573	 pshufd		\$0b01001110,$Xi,$T1
574	 pshufd		\$0b01001110,$Hkey2,$T2
575	 pxor		$Xi,$T1			#
576	 pxor		$Hkey2,$T2
577
578	pxor		$Xn,$T1n		#
579	pxor		$Xhn,$T1n		#
580	movdqa		$T1n,$T2n		#
581	psrldq		\$8,$T1n
582	pslldq		\$8,$T2n		#
583	pxor		$T1n,$Xhn
584	pxor		$T2n,$Xn		#
585
586	lea		32($inp),$inp
587	sub		\$0x20,$len
588	ja		.Lmod_loop
589
590.Leven_tail:
591___
592	&clmul64x64_T2	($Xhi,$Xi,$Hkey2,1);	# H^2*(Ii+Xi)
593$code.=<<___;
594	pxor		$Xn,$Xi			# (H*Ii+1) + H^2*(Ii+Xi)
595	pxor		$Xhn,$Xhi
596___
597	&reduction_alg9	($Xhi,$Xi);
598$code.=<<___;
599	test		$len,$len
600	jnz		.Ldone
601
602.Lodd_tail:
603	movdqu		($inp),$T1		# Ii
604	pshufb		$T3,$T1
605	pxor		$T1,$Xi			# Ii+Xi
606___
607	&clmul64x64_T2	($Xhi,$Xi,$Hkey);	# H*(Ii+Xi)
608	&reduction_alg9	($Xhi,$Xi);
609$code.=<<___;
610.Ldone:
611	pshufb		$T3,$Xi
612	movdqu		$Xi,($Xip)
613___
614$code.=<<___ if ($win64);
615	movaps	(%rsp),%xmm6
616	movaps	0x10(%rsp),%xmm7
617	movaps	0x20(%rsp),%xmm8
618	movaps	0x30(%rsp),%xmm9
619	movaps	0x40(%rsp),%xmm10
620	add	\$0x58,%rsp
621___
622$code.=<<___;
623	ret
624.LSEH_end_gcm_ghash_clmul:
625.size	gcm_ghash_clmul,.-gcm_ghash_clmul
626___
627}
628
629$code.=<<___;
630.section .rodata
631.align	64
632.Lbswap_mask:
633	.byte	15,14,13,12,11,10,9,8,7,6,5,4,3,2,1,0
634.L0x1c2_polynomial:
635	.byte	1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0xc2
636.align	64
637.type	.Lrem_4bit,\@object
638.Lrem_4bit:
639	.long	0,`0x0000<<16`,0,`0x1C20<<16`,0,`0x3840<<16`,0,`0x2460<<16`
640	.long	0,`0x7080<<16`,0,`0x6CA0<<16`,0,`0x48C0<<16`,0,`0x54E0<<16`
641	.long	0,`0xE100<<16`,0,`0xFD20<<16`,0,`0xD940<<16`,0,`0xC560<<16`
642	.long	0,`0x9180<<16`,0,`0x8DA0<<16`,0,`0xA9C0<<16`,0,`0xB5E0<<16`
643.type	.Lrem_8bit,\@object
644.Lrem_8bit:
645	.value	0x0000,0x01C2,0x0384,0x0246,0x0708,0x06CA,0x048C,0x054E
646	.value	0x0E10,0x0FD2,0x0D94,0x0C56,0x0918,0x08DA,0x0A9C,0x0B5E
647	.value	0x1C20,0x1DE2,0x1FA4,0x1E66,0x1B28,0x1AEA,0x18AC,0x196E
648	.value	0x1230,0x13F2,0x11B4,0x1076,0x1538,0x14FA,0x16BC,0x177E
649	.value	0x3840,0x3982,0x3BC4,0x3A06,0x3F48,0x3E8A,0x3CCC,0x3D0E
650	.value	0x3650,0x3792,0x35D4,0x3416,0x3158,0x309A,0x32DC,0x331E
651	.value	0x2460,0x25A2,0x27E4,0x2626,0x2368,0x22AA,0x20EC,0x212E
652	.value	0x2A70,0x2BB2,0x29F4,0x2836,0x2D78,0x2CBA,0x2EFC,0x2F3E
653	.value	0x7080,0x7142,0x7304,0x72C6,0x7788,0x764A,0x740C,0x75CE
654	.value	0x7E90,0x7F52,0x7D14,0x7CD6,0x7998,0x785A,0x7A1C,0x7BDE
655	.value	0x6CA0,0x6D62,0x6F24,0x6EE6,0x6BA8,0x6A6A,0x682C,0x69EE
656	.value	0x62B0,0x6372,0x6134,0x60F6,0x65B8,0x647A,0x663C,0x67FE
657	.value	0x48C0,0x4902,0x4B44,0x4A86,0x4FC8,0x4E0A,0x4C4C,0x4D8E
658	.value	0x46D0,0x4712,0x4554,0x4496,0x41D8,0x401A,0x425C,0x439E
659	.value	0x54E0,0x5522,0x5764,0x56A6,0x53E8,0x522A,0x506C,0x51AE
660	.value	0x5AF0,0x5B32,0x5974,0x58B6,0x5DF8,0x5C3A,0x5E7C,0x5FBE
661	.value	0xE100,0xE0C2,0xE284,0xE346,0xE608,0xE7CA,0xE58C,0xE44E
662	.value	0xEF10,0xEED2,0xEC94,0xED56,0xE818,0xE9DA,0xEB9C,0xEA5E
663	.value	0xFD20,0xFCE2,0xFEA4,0xFF66,0xFA28,0xFBEA,0xF9AC,0xF86E
664	.value	0xF330,0xF2F2,0xF0B4,0xF176,0xF438,0xF5FA,0xF7BC,0xF67E
665	.value	0xD940,0xD882,0xDAC4,0xDB06,0xDE48,0xDF8A,0xDDCC,0xDC0E
666	.value	0xD750,0xD692,0xD4D4,0xD516,0xD058,0xD19A,0xD3DC,0xD21E
667	.value	0xC560,0xC4A2,0xC6E4,0xC726,0xC268,0xC3AA,0xC1EC,0xC02E
668	.value	0xCB70,0xCAB2,0xC8F4,0xC936,0xCC78,0xCDBA,0xCFFC,0xCE3E
669	.value	0x9180,0x9042,0x9204,0x93C6,0x9688,0x974A,0x950C,0x94CE
670	.value	0x9F90,0x9E52,0x9C14,0x9DD6,0x9898,0x995A,0x9B1C,0x9ADE
671	.value	0x8DA0,0x8C62,0x8E24,0x8FE6,0x8AA8,0x8B6A,0x892C,0x88EE
672	.value	0x83B0,0x8272,0x8034,0x81F6,0x84B8,0x857A,0x873C,0x86FE
673	.value	0xA9C0,0xA802,0xAA44,0xAB86,0xAEC8,0xAF0A,0xAD4C,0xAC8E
674	.value	0xA7D0,0xA612,0xA454,0xA596,0xA0D8,0xA11A,0xA35C,0xA29E
675	.value	0xB5E0,0xB422,0xB664,0xB7A6,0xB2E8,0xB32A,0xB16C,0xB0AE
676	.value	0xBBF0,0xBA32,0xB874,0xB9B6,0xBCF8,0xBD3A,0xBF7C,0xBEBE
677.align	64
678.text
679___
680
681# EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
682#		CONTEXT *context,DISPATCHER_CONTEXT *disp)
683if ($win64) {
684$rec="%rcx";
685$frame="%rdx";
686$context="%r8";
687$disp="%r9";
688
689$code.=<<___;
690.extern	__imp_RtlVirtualUnwind
691.type	se_handler,\@abi-omnipotent
692.align	16
693se_handler:
694	_CET_ENDBR
695	push	%rsi
696	push	%rdi
697	push	%rbx
698	push	%rbp
699	push	%r12
700	push	%r13
701	push	%r14
702	push	%r15
703	pushfq
704	sub	\$64,%rsp
705
706	mov	120($context),%rax	# pull context->Rax
707	mov	248($context),%rbx	# pull context->Rip
708
709	mov	8($disp),%rsi		# disp->ImageBase
710	mov	56($disp),%r11		# disp->HandlerData
711
712	mov	0(%r11),%r10d		# HandlerData[0]
713	lea	(%rsi,%r10),%r10	# prologue label
714	cmp	%r10,%rbx		# context->Rip<prologue label
715	jb	.Lin_prologue
716
717	mov	152($context),%rax	# pull context->Rsp
718
719	mov	4(%r11),%r10d		# HandlerData[1]
720	lea	(%rsi,%r10),%r10	# epilogue label
721	cmp	%r10,%rbx		# context->Rip>=epilogue label
722	jae	.Lin_prologue
723
724	lea	24(%rax),%rax		# adjust "rsp"
725
726	mov	-8(%rax),%rbx
727	mov	-16(%rax),%rbp
728	mov	-24(%rax),%r12
729	mov	%rbx,144($context)	# restore context->Rbx
730	mov	%rbp,160($context)	# restore context->Rbp
731	mov	%r12,216($context)	# restore context->R12
732
733.Lin_prologue:
734	mov	8(%rax),%rdi
735	mov	16(%rax),%rsi
736	mov	%rax,152($context)	# restore context->Rsp
737	mov	%rsi,168($context)	# restore context->Rsi
738	mov	%rdi,176($context)	# restore context->Rdi
739
740	mov	40($disp),%rdi		# disp->ContextRecord
741	mov	$context,%rsi		# context
742	mov	\$`1232/8`,%ecx		# sizeof(CONTEXT)
743	.long	0xa548f3fc		# cld; rep movsq
744
745	mov	$disp,%rsi
746	xor	%rcx,%rcx		# arg1, UNW_FLAG_NHANDLER
747	mov	8(%rsi),%rdx		# arg2, disp->ImageBase
748	mov	0(%rsi),%r8		# arg3, disp->ControlPc
749	mov	16(%rsi),%r9		# arg4, disp->FunctionEntry
750	mov	40(%rsi),%r10		# disp->ContextRecord
751	lea	56(%rsi),%r11		# &disp->HandlerData
752	lea	24(%rsi),%r12		# &disp->EstablisherFrame
753	mov	%r10,32(%rsp)		# arg5
754	mov	%r11,40(%rsp)		# arg6
755	mov	%r12,48(%rsp)		# arg7
756	mov	%rcx,56(%rsp)		# arg8, (NULL)
757	call	*__imp_RtlVirtualUnwind(%rip)
758
759	mov	\$1,%eax		# ExceptionContinueSearch
760	add	\$64,%rsp
761	popfq
762	pop	%r15
763	pop	%r14
764	pop	%r13
765	pop	%r12
766	pop	%rbp
767	pop	%rbx
768	pop	%rdi
769	pop	%rsi
770	ret
771.size	se_handler,.-se_handler
772
773.section	.pdata
774.align	4
775	.rva	.LSEH_begin_gcm_gmult_4bit
776	.rva	.LSEH_end_gcm_gmult_4bit
777	.rva	.LSEH_info_gcm_gmult_4bit
778
779	.rva	.LSEH_begin_gcm_ghash_4bit
780	.rva	.LSEH_end_gcm_ghash_4bit
781	.rva	.LSEH_info_gcm_ghash_4bit
782
783	.rva	.LSEH_begin_gcm_ghash_clmul
784	.rva	.LSEH_end_gcm_ghash_clmul
785	.rva	.LSEH_info_gcm_ghash_clmul
786
787.section	.xdata
788.align	8
789.LSEH_info_gcm_gmult_4bit:
790	.byte	9,0,0,0
791	.rva	se_handler
792	.rva	.Lgmult_prologue,.Lgmult_epilogue	# HandlerData
793.LSEH_info_gcm_ghash_4bit:
794	.byte	9,0,0,0
795	.rva	se_handler
796	.rva	.Lghash_prologue,.Lghash_epilogue	# HandlerData
797.LSEH_info_gcm_ghash_clmul:
798	.byte	0x01,0x1f,0x0b,0x00
799	.byte	0x1f,0xa8,0x04,0x00	#movaps 0x40(rsp),xmm10
800	.byte	0x19,0x98,0x03,0x00	#movaps 0x30(rsp),xmm9
801	.byte	0x13,0x88,0x02,0x00	#movaps 0x20(rsp),xmm8
802	.byte	0x0d,0x78,0x01,0x00	#movaps 0x10(rsp),xmm7
803	.byte	0x08,0x68,0x00,0x00	#movaps (rsp),xmm6
804	.byte	0x04,0xa2,0x00,0x00	#sub	rsp,0x58
805___
806}
807
808$code =~ s/\`([^\`]*)\`/eval($1)/gem;
809
810print $code;
811
812close STDOUT;
813