xref: /freebsd/crypto/openssl/crypto/sha/asm/sha1-mb-x86_64.pl (revision bdd1243df58e60e85101c09001d9812a789b6bc4)
1#! /usr/bin/env perl
2# Copyright 2013-2020 The OpenSSL Project Authors. All Rights Reserved.
3#
4# Licensed under the OpenSSL license (the "License").  You may not use
5# this file except in compliance with the License.  You can obtain a copy
6# in the file LICENSE in the source distribution or at
7# https://www.openssl.org/source/license.html
8
9
10# ====================================================================
11# Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
12# project. The module is, however, dual licensed under OpenSSL and
13# CRYPTOGAMS licenses depending on where you obtain it. For further
14# details see http://www.openssl.org/~appro/cryptogams/.
15# ====================================================================
16
17# Multi-buffer SHA1 procedure processes n buffers in parallel by
18# placing buffer data to designated lane of SIMD register. n is
19# naturally limited to 4 on pre-AVX2 processors and to 8 on
20# AVX2-capable processors such as Haswell.
21#
22#		this	+aesni(i)	sha1	aesni-sha1	gain(iv)
23# -------------------------------------------------------------------
24# Westmere(ii)	10.7/n	+1.28=3.96(n=4)	5.30	6.66		+68%
25# Atom(ii)	18.1/n	+3.93=8.46(n=4)	9.37	12.8		+51%
26# Sandy Bridge	(8.16	+5.15=13.3)/n	4.99	5.98		+80%
27# Ivy Bridge	(8.08	+5.14=13.2)/n	4.60	5.54		+68%
28# Haswell(iii)	(8.96	+5.00=14.0)/n	3.57	4.55		+160%
29# Skylake	(8.70	+5.00=13.7)/n	3.64	4.20		+145%
30# Bulldozer	(9.76	+5.76=15.5)/n	5.95	6.37		+64%
31#
32# (i)	multi-block CBC encrypt with 128-bit key;
33# (ii)	(HASH+AES)/n does not apply to Westmere for n>3 and Atom,
34#	because of lower AES-NI instruction throughput;
35# (iii)	"this" is for n=8, when we gather twice as much data, result
36#	for n=4 is 8.00+4.44=12.4;
37# (iv)	presented improvement coefficients are asymptotic limits and
38#	in real-life application are somewhat lower, e.g. for 2KB
39#	fragments they range from 30% to 100% (on Haswell);
40
41$flavour = shift;
42$output  = shift;
43if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
44
45$win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);
46
47$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
48( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
49( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or
50die "can't locate x86_64-xlate.pl";
51
52$avx=0;
53
54if (`$ENV{CC} -Wa,-v -c -o /dev/null -x assembler /dev/null 2>&1`
55		=~ /GNU assembler version ([2-9]\.[0-9]+)/) {
56	$avx = ($1>=2.19) + ($1>=2.22);
57}
58
59if (!$avx && $win64 && ($flavour =~ /nasm/ || $ENV{ASM} =~ /nasm/) &&
60	   `nasm -v 2>&1` =~ /NASM version ([2-9]\.[0-9]+)/) {
61	$avx = ($1>=2.09) + ($1>=2.10);
62}
63
64if (!$avx && $win64 && ($flavour =~ /masm/ || $ENV{ASM} =~ /ml64/) &&
65	   `ml64 2>&1` =~ /Version ([0-9]+)\./) {
66	$avx = ($1>=10) + ($1>=11);
67}
68
69if (!$avx && `$ENV{CC} -v 2>&1` =~ /((?:clang|LLVM) version|.*based on LLVM) ([0-9]+\.[0-9]+)/) {
70	$avx = ($2>=3.0) + ($2>3.0);
71}
72
73open OUT,"| \"$^X\" \"$xlate\" $flavour \"$output\"";
74*STDOUT=*OUT;
75
76# void sha1_multi_block (
77#     struct {	unsigned int A[8];
78#		unsigned int B[8];
79#		unsigned int C[8];
80#		unsigned int D[8];
81#		unsigned int E[8];	} *ctx,
82#     struct {	void *ptr; int blocks;	} inp[8],
83#     int num);		/* 1 or 2 */
84#
85$ctx="%rdi";	# 1st arg
86$inp="%rsi";	# 2nd arg
87$num="%edx";
88@ptr=map("%r$_",(8..11));
89$Tbl="%rbp";
90
91@V=($A,$B,$C,$D,$E)=map("%xmm$_",(0..4));
92($t0,$t1,$t2,$t3,$tx)=map("%xmm$_",(5..9));
93@Xi=map("%xmm$_",(10..14));
94$K="%xmm15";
95
96if (1) {
97    # Atom-specific optimization aiming to eliminate pshufb with high
98    # registers [and thus get rid of 48 cycles accumulated penalty]
99    @Xi=map("%xmm$_",(0..4));
100    ($tx,$t0,$t1,$t2,$t3)=map("%xmm$_",(5..9));
101    @V=($A,$B,$C,$D,$E)=map("%xmm$_",(10..14));
102}
103
104$REG_SZ=16;
105
106sub Xi_off {
107my $off = shift;
108
109    $off %= 16; $off *= $REG_SZ;
110    $off<256 ? "$off-128(%rax)" : "$off-256-128(%rbx)";
111}
112
113sub BODY_00_19 {
114my ($i,$a,$b,$c,$d,$e)=@_;
115my $j=$i+1;
116my $k=$i+2;
117
118# Loads are performed 2+3/4 iterations in advance. 3/4 means that out
119# of 4 words you would expect to be loaded per given iteration one is
120# spilled to next iteration. In other words indices in four input
121# streams are distributed as following:
122#
123# $i==0:	0,0,0,0,1,1,1,1,2,2,2,
124# $i==1:	2,3,3,3,
125# $i==2:	3,4,4,4,
126# ...
127# $i==13:	14,15,15,15,
128# $i==14:	15
129#
130# Then at $i==15 Xupdate is applied one iteration in advance...
131$code.=<<___ if ($i==0);
132	movd		(@ptr[0]),@Xi[0]
133	 lea		`16*4`(@ptr[0]),@ptr[0]
134	movd		(@ptr[1]),@Xi[2]	# borrow @Xi[2]
135	 lea		`16*4`(@ptr[1]),@ptr[1]
136	movd		(@ptr[2]),@Xi[3]	# borrow @Xi[3]
137	 lea		`16*4`(@ptr[2]),@ptr[2]
138	movd		(@ptr[3]),@Xi[4]	# borrow @Xi[4]
139	 lea		`16*4`(@ptr[3]),@ptr[3]
140	punpckldq	@Xi[3],@Xi[0]
141	 movd		`4*$j-16*4`(@ptr[0]),@Xi[1]
142	punpckldq	@Xi[4],@Xi[2]
143	 movd		`4*$j-16*4`(@ptr[1]),$t3
144	punpckldq	@Xi[2],@Xi[0]
145	 movd		`4*$j-16*4`(@ptr[2]),$t2
146	pshufb		$tx,@Xi[0]
147___
148$code.=<<___ if ($i<14);			# just load input
149	 movd		`4*$j-16*4`(@ptr[3]),$t1
150	 punpckldq	$t2,@Xi[1]
151	movdqa	$a,$t2
152	paddd	$K,$e				# e+=K_00_19
153	 punpckldq	$t1,$t3
154	movdqa	$b,$t1
155	movdqa	$b,$t0
156	pslld	\$5,$t2
157	pandn	$d,$t1
158	pand	$c,$t0
159	 punpckldq	$t3,@Xi[1]
160	movdqa	$a,$t3
161
162	movdqa	@Xi[0],`&Xi_off($i)`
163	paddd	@Xi[0],$e			# e+=X[i]
164	 movd		`4*$k-16*4`(@ptr[0]),@Xi[2]
165	psrld	\$27,$t3
166	pxor	$t1,$t0				# Ch(b,c,d)
167	movdqa	$b,$t1
168
169	por	$t3,$t2				# rol(a,5)
170	 movd		`4*$k-16*4`(@ptr[1]),$t3
171	pslld	\$30,$t1
172	paddd	$t0,$e				# e+=Ch(b,c,d)
173
174	psrld	\$2,$b
175	paddd	$t2,$e				# e+=rol(a,5)
176	 pshufb	$tx,@Xi[1]
177	 movd		`4*$k-16*4`(@ptr[2]),$t2
178	por	$t1,$b				# b=rol(b,30)
179___
180$code.=<<___ if ($i==14);			# just load input
181	 movd		`4*$j-16*4`(@ptr[3]),$t1
182	 punpckldq	$t2,@Xi[1]
183	movdqa	$a,$t2
184	paddd	$K,$e				# e+=K_00_19
185	 punpckldq	$t1,$t3
186	movdqa	$b,$t1
187	movdqa	$b,$t0
188	pslld	\$5,$t2
189	 prefetcht0	63(@ptr[0])
190	pandn	$d,$t1
191	pand	$c,$t0
192	 punpckldq	$t3,@Xi[1]
193	movdqa	$a,$t3
194
195	movdqa	@Xi[0],`&Xi_off($i)`
196	paddd	@Xi[0],$e			# e+=X[i]
197	psrld	\$27,$t3
198	pxor	$t1,$t0				# Ch(b,c,d)
199	movdqa	$b,$t1
200	 prefetcht0	63(@ptr[1])
201
202	por	$t3,$t2				# rol(a,5)
203	pslld	\$30,$t1
204	paddd	$t0,$e				# e+=Ch(b,c,d)
205	 prefetcht0	63(@ptr[2])
206
207	psrld	\$2,$b
208	paddd	$t2,$e				# e+=rol(a,5)
209	 pshufb	$tx,@Xi[1]
210	 prefetcht0	63(@ptr[3])
211	por	$t1,$b				# b=rol(b,30)
212___
213$code.=<<___ if ($i>=13 && $i<15);
214	movdqa	`&Xi_off($j+2)`,@Xi[3]		# preload "X[2]"
215___
216$code.=<<___ if ($i>=15);			# apply Xupdate
217	pxor	@Xi[-2],@Xi[1]			# "X[13]"
218	movdqa	`&Xi_off($j+2)`,@Xi[3]		# "X[2]"
219
220	movdqa	$a,$t2
221	 pxor	`&Xi_off($j+8)`,@Xi[1]
222	paddd	$K,$e				# e+=K_00_19
223	movdqa	$b,$t1
224	pslld	\$5,$t2
225	 pxor	@Xi[3],@Xi[1]
226	movdqa	$b,$t0
227	pandn	$d,$t1
228	 movdqa	@Xi[1],$tx
229	pand	$c,$t0
230	movdqa	$a,$t3
231	 psrld	\$31,$tx
232	 paddd	@Xi[1],@Xi[1]
233
234	movdqa	@Xi[0],`&Xi_off($i)`
235	paddd	@Xi[0],$e			# e+=X[i]
236	psrld	\$27,$t3
237	pxor	$t1,$t0				# Ch(b,c,d)
238
239	movdqa	$b,$t1
240	por	$t3,$t2				# rol(a,5)
241	pslld	\$30,$t1
242	paddd	$t0,$e				# e+=Ch(b,c,d)
243
244	psrld	\$2,$b
245	paddd	$t2,$e				# e+=rol(a,5)
246	 por	$tx,@Xi[1]			# rol	\$1,@Xi[1]
247	por	$t1,$b				# b=rol(b,30)
248___
249push(@Xi,shift(@Xi));
250}
251
252sub BODY_20_39 {
253my ($i,$a,$b,$c,$d,$e)=@_;
254my $j=$i+1;
255
256$code.=<<___ if ($i<79);
257	pxor	@Xi[-2],@Xi[1]			# "X[13]"
258	movdqa	`&Xi_off($j+2)`,@Xi[3]		# "X[2]"
259
260	movdqa	$a,$t2
261	movdqa	$d,$t0
262	 pxor	`&Xi_off($j+8)`,@Xi[1]
263	paddd	$K,$e				# e+=K_20_39
264	pslld	\$5,$t2
265	pxor	$b,$t0
266
267	movdqa	$a,$t3
268___
269$code.=<<___ if ($i<72);
270	movdqa	@Xi[0],`&Xi_off($i)`
271___
272$code.=<<___ if ($i<79);
273	paddd	@Xi[0],$e			# e+=X[i]
274	 pxor	@Xi[3],@Xi[1]
275	psrld	\$27,$t3
276	pxor	$c,$t0				# Parity(b,c,d)
277	movdqa	$b,$t1
278
279	pslld	\$30,$t1
280	 movdqa	@Xi[1],$tx
281	por	$t3,$t2				# rol(a,5)
282	 psrld	\$31,$tx
283	paddd	$t0,$e				# e+=Parity(b,c,d)
284	 paddd	@Xi[1],@Xi[1]
285
286	psrld	\$2,$b
287	paddd	$t2,$e				# e+=rol(a,5)
288	 por	$tx,@Xi[1]			# rol(@Xi[1],1)
289	por	$t1,$b				# b=rol(b,30)
290___
291$code.=<<___ if ($i==79);
292	movdqa	$a,$t2
293	paddd	$K,$e				# e+=K_20_39
294	movdqa	$d,$t0
295	pslld	\$5,$t2
296	pxor	$b,$t0
297
298	movdqa	$a,$t3
299	paddd	@Xi[0],$e			# e+=X[i]
300	psrld	\$27,$t3
301	movdqa	$b,$t1
302	pxor	$c,$t0				# Parity(b,c,d)
303
304	pslld	\$30,$t1
305	por	$t3,$t2				# rol(a,5)
306	paddd	$t0,$e				# e+=Parity(b,c,d)
307
308	psrld	\$2,$b
309	paddd	$t2,$e				# e+=rol(a,5)
310	por	$t1,$b				# b=rol(b,30)
311___
312push(@Xi,shift(@Xi));
313}
314
315sub BODY_40_59 {
316my ($i,$a,$b,$c,$d,$e)=@_;
317my $j=$i+1;
318
319$code.=<<___;
320	pxor	@Xi[-2],@Xi[1]			# "X[13]"
321	movdqa	`&Xi_off($j+2)`,@Xi[3]		# "X[2]"
322
323	movdqa	$a,$t2
324	movdqa	$d,$t1
325	 pxor	`&Xi_off($j+8)`,@Xi[1]
326	pxor	@Xi[3],@Xi[1]
327	paddd	$K,$e				# e+=K_40_59
328	pslld	\$5,$t2
329	movdqa	$a,$t3
330	pand	$c,$t1
331
332	movdqa	$d,$t0
333	 movdqa	@Xi[1],$tx
334	psrld	\$27,$t3
335	paddd	$t1,$e
336	pxor	$c,$t0
337
338	movdqa	@Xi[0],`&Xi_off($i)`
339	paddd	@Xi[0],$e			# e+=X[i]
340	por	$t3,$t2				# rol(a,5)
341	 psrld	\$31,$tx
342	pand	$b,$t0
343	movdqa	$b,$t1
344
345	pslld	\$30,$t1
346	 paddd	@Xi[1],@Xi[1]
347	paddd	$t0,$e				# e+=Maj(b,d,c)
348
349	psrld	\$2,$b
350	paddd	$t2,$e				# e+=rol(a,5)
351	 por	$tx,@Xi[1]			# rol(@X[1],1)
352	por	$t1,$b				# b=rol(b,30)
353___
354push(@Xi,shift(@Xi));
355}
356
357$code.=<<___;
358.text
359
360.extern	OPENSSL_ia32cap_P
361
362.globl	sha1_multi_block
363.type	sha1_multi_block,\@function,3
364.align	32
365sha1_multi_block:
366.cfi_startproc
367	mov	OPENSSL_ia32cap_P+4(%rip),%rcx
368	bt	\$61,%rcx			# check SHA bit
369	jc	_shaext_shortcut
370___
371$code.=<<___ if ($avx);
372	test	\$`1<<28`,%ecx
373	jnz	_avx_shortcut
374___
375$code.=<<___;
376	mov	%rsp,%rax
377.cfi_def_cfa_register	%rax
378	push	%rbx
379.cfi_push	%rbx
380	push	%rbp
381.cfi_push	%rbx
382___
383$code.=<<___ if ($win64);
384	lea	-0xa8(%rsp),%rsp
385	movaps	%xmm6,(%rsp)
386	movaps	%xmm7,0x10(%rsp)
387	movaps	%xmm8,0x20(%rsp)
388	movaps	%xmm9,0x30(%rsp)
389	movaps	%xmm10,-0x78(%rax)
390	movaps	%xmm11,-0x68(%rax)
391	movaps	%xmm12,-0x58(%rax)
392	movaps	%xmm13,-0x48(%rax)
393	movaps	%xmm14,-0x38(%rax)
394	movaps	%xmm15,-0x28(%rax)
395___
396$code.=<<___;
397	sub	\$`$REG_SZ*18`,%rsp
398	and	\$-256,%rsp
399	mov	%rax,`$REG_SZ*17`(%rsp)		# original %rsp
400.cfi_cfa_expression	%rsp+`$REG_SZ*17`,deref,+8
401.Lbody:
402	lea	K_XX_XX(%rip),$Tbl
403	lea	`$REG_SZ*16`(%rsp),%rbx
404
405.Loop_grande:
406	mov	$num,`$REG_SZ*17+8`(%rsp)	# original $num
407	xor	$num,$num
408___
409for($i=0;$i<4;$i++) {
410    $code.=<<___;
411	mov	`16*$i+0`($inp),@ptr[$i]	# input pointer
412	mov	`16*$i+8`($inp),%ecx		# number of blocks
413	cmp	$num,%ecx
414	cmovg	%ecx,$num			# find maximum
415	test	%ecx,%ecx
416	mov	%ecx,`4*$i`(%rbx)		# initialize counters
417	cmovle	$Tbl,@ptr[$i]			# cancel input
418___
419}
420$code.=<<___;
421	test	$num,$num
422	jz	.Ldone
423
424	movdqu	0x00($ctx),$A			# load context
425	 lea	128(%rsp),%rax
426	movdqu	0x20($ctx),$B
427	movdqu	0x40($ctx),$C
428	movdqu	0x60($ctx),$D
429	movdqu	0x80($ctx),$E
430	movdqa	0x60($Tbl),$tx			# pbswap_mask
431	movdqa	-0x20($Tbl),$K			# K_00_19
432	jmp	.Loop
433
434.align	32
435.Loop:
436___
437for($i=0;$i<20;$i++)	{ &BODY_00_19($i,@V); unshift(@V,pop(@V)); }
438$code.="	movdqa	0x00($Tbl),$K\n";	# K_20_39
439for(;$i<40;$i++)	{ &BODY_20_39($i,@V); unshift(@V,pop(@V)); }
440$code.="	movdqa	0x20($Tbl),$K\n";	# K_40_59
441for(;$i<60;$i++)	{ &BODY_40_59($i,@V); unshift(@V,pop(@V)); }
442$code.="	movdqa	0x40($Tbl),$K\n";	# K_60_79
443for(;$i<80;$i++)	{ &BODY_20_39($i,@V); unshift(@V,pop(@V)); }
444$code.=<<___;
445	movdqa	(%rbx),@Xi[0]			# pull counters
446	mov	\$1,%ecx
447	cmp	4*0(%rbx),%ecx			# examine counters
448	pxor	$t2,$t2
449	cmovge	$Tbl,@ptr[0]			# cancel input
450	cmp	4*1(%rbx),%ecx
451	movdqa	@Xi[0],@Xi[1]
452	cmovge	$Tbl,@ptr[1]
453	cmp	4*2(%rbx),%ecx
454	pcmpgtd	$t2,@Xi[1]			# mask value
455	cmovge	$Tbl,@ptr[2]
456	cmp	4*3(%rbx),%ecx
457	paddd	@Xi[1],@Xi[0]			# counters--
458	cmovge	$Tbl,@ptr[3]
459
460	movdqu	0x00($ctx),$t0
461	pand	@Xi[1],$A
462	movdqu	0x20($ctx),$t1
463	pand	@Xi[1],$B
464	paddd	$t0,$A
465	movdqu	0x40($ctx),$t2
466	pand	@Xi[1],$C
467	paddd	$t1,$B
468	movdqu	0x60($ctx),$t3
469	pand	@Xi[1],$D
470	paddd	$t2,$C
471	movdqu	0x80($ctx),$tx
472	pand	@Xi[1],$E
473	movdqu	$A,0x00($ctx)
474	paddd	$t3,$D
475	movdqu	$B,0x20($ctx)
476	paddd	$tx,$E
477	movdqu	$C,0x40($ctx)
478	movdqu	$D,0x60($ctx)
479	movdqu	$E,0x80($ctx)
480
481	movdqa	@Xi[0],(%rbx)			# save counters
482	movdqa	0x60($Tbl),$tx			# pbswap_mask
483	movdqa	-0x20($Tbl),$K			# K_00_19
484	dec	$num
485	jnz	.Loop
486
487	mov	`$REG_SZ*17+8`(%rsp),$num
488	lea	$REG_SZ($ctx),$ctx
489	lea	`16*$REG_SZ/4`($inp),$inp
490	dec	$num
491	jnz	.Loop_grande
492
493.Ldone:
494	mov	`$REG_SZ*17`(%rsp),%rax		# original %rsp
495.cfi_def_cfa	%rax,8
496___
497$code.=<<___ if ($win64);
498	movaps	-0xb8(%rax),%xmm6
499	movaps	-0xa8(%rax),%xmm7
500	movaps	-0x98(%rax),%xmm8
501	movaps	-0x88(%rax),%xmm9
502	movaps	-0x78(%rax),%xmm10
503	movaps	-0x68(%rax),%xmm11
504	movaps	-0x58(%rax),%xmm12
505	movaps	-0x48(%rax),%xmm13
506	movaps	-0x38(%rax),%xmm14
507	movaps	-0x28(%rax),%xmm15
508___
509$code.=<<___;
510	mov	-16(%rax),%rbp
511.cfi_restore	%rbp
512	mov	-8(%rax),%rbx
513.cfi_restore	%rbx
514	lea	(%rax),%rsp
515.cfi_def_cfa_register	%rsp
516.Lepilogue:
517	ret
518.cfi_endproc
519.size	sha1_multi_block,.-sha1_multi_block
520___
521						{{{
522my ($ABCD0,$E0,$E0_,$BSWAP,$ABCD1,$E1,$E1_)=map("%xmm$_",(0..3,8..10));
523my @MSG0=map("%xmm$_",(4..7));
524my @MSG1=map("%xmm$_",(11..14));
525
526$code.=<<___;
527.type	sha1_multi_block_shaext,\@function,3
528.align	32
529sha1_multi_block_shaext:
530.cfi_startproc
531_shaext_shortcut:
532	mov	%rsp,%rax
533.cfi_def_cfa_register	%rax
534	push	%rbx
535.cfi_push	%rbx
536	push	%rbp
537.cfi_push	%rbp
538___
539$code.=<<___ if ($win64);
540	lea	-0xa8(%rsp),%rsp
541	movaps	%xmm6,(%rsp)
542	movaps	%xmm7,0x10(%rsp)
543	movaps	%xmm8,0x20(%rsp)
544	movaps	%xmm9,0x30(%rsp)
545	movaps	%xmm10,-0x78(%rax)
546	movaps	%xmm11,-0x68(%rax)
547	movaps	%xmm12,-0x58(%rax)
548	movaps	%xmm13,-0x48(%rax)
549	movaps	%xmm14,-0x38(%rax)
550	movaps	%xmm15,-0x28(%rax)
551___
552$code.=<<___;
553	sub	\$`$REG_SZ*18`,%rsp
554	shl	\$1,$num			# we process pair at a time
555	and	\$-256,%rsp
556	lea	0x40($ctx),$ctx			# size optimization
557	mov	%rax,`$REG_SZ*17`(%rsp)		# original %rsp
558.Lbody_shaext:
559	lea	`$REG_SZ*16`(%rsp),%rbx
560	movdqa	K_XX_XX+0x80(%rip),$BSWAP	# byte-n-word swap
561
562.Loop_grande_shaext:
563	mov	$num,`$REG_SZ*17+8`(%rsp)	# original $num
564	xor	$num,$num
565___
566for($i=0;$i<2;$i++) {
567    $code.=<<___;
568	mov	`16*$i+0`($inp),@ptr[$i]	# input pointer
569	mov	`16*$i+8`($inp),%ecx		# number of blocks
570	cmp	$num,%ecx
571	cmovg	%ecx,$num			# find maximum
572	test	%ecx,%ecx
573	mov	%ecx,`4*$i`(%rbx)		# initialize counters
574	cmovle	%rsp,@ptr[$i]			# cancel input
575___
576}
577$code.=<<___;
578	test	$num,$num
579	jz	.Ldone_shaext
580
581	movq		0x00-0x40($ctx),$ABCD0	# a1.a0
582	movq		0x20-0x40($ctx),@MSG0[0]# b1.b0
583	movq		0x40-0x40($ctx),@MSG0[1]# c1.c0
584	movq		0x60-0x40($ctx),@MSG0[2]# d1.d0
585	movq		0x80-0x40($ctx),@MSG0[3]# e1.e0
586
587	punpckldq	@MSG0[0],$ABCD0		# b1.a1.b0.a0
588	punpckldq	@MSG0[2],@MSG0[1]	# d1.c1.d0.c0
589
590	movdqa		$ABCD0,$ABCD1
591	punpcklqdq	@MSG0[1],$ABCD0		# d0.c0.b0.a0
592	punpckhqdq	@MSG0[1],$ABCD1		# d1.c1.b1.a1
593
594	pshufd		\$0b00111111,@MSG0[3],$E0
595	pshufd		\$0b01111111,@MSG0[3],$E1
596	pshufd		\$0b00011011,$ABCD0,$ABCD0
597	pshufd		\$0b00011011,$ABCD1,$ABCD1
598	jmp		.Loop_shaext
599
600.align	32
601.Loop_shaext:
602	movdqu		0x00(@ptr[0]),@MSG0[0]
603	 movdqu		0x00(@ptr[1]),@MSG1[0]
604	movdqu		0x10(@ptr[0]),@MSG0[1]
605	 movdqu		0x10(@ptr[1]),@MSG1[1]
606	movdqu		0x20(@ptr[0]),@MSG0[2]
607	pshufb		$BSWAP,@MSG0[0]
608	 movdqu		0x20(@ptr[1]),@MSG1[2]
609	 pshufb		$BSWAP,@MSG1[0]
610	movdqu		0x30(@ptr[0]),@MSG0[3]
611	lea		0x40(@ptr[0]),@ptr[0]
612	pshufb		$BSWAP,@MSG0[1]
613	 movdqu		0x30(@ptr[1]),@MSG1[3]
614	 lea		0x40(@ptr[1]),@ptr[1]
615	 pshufb		$BSWAP,@MSG1[1]
616
617	movdqa		$E0,0x50(%rsp)		# offload
618	paddd		@MSG0[0],$E0
619	 movdqa		$E1,0x70(%rsp)
620	 paddd		@MSG1[0],$E1
621	movdqa		$ABCD0,0x40(%rsp)	# offload
622	movdqa		$ABCD0,$E0_
623	 movdqa		$ABCD1,0x60(%rsp)
624	 movdqa		$ABCD1,$E1_
625	sha1rnds4	\$0,$E0,$ABCD0		# 0-3
626	sha1nexte	@MSG0[1],$E0_
627	 sha1rnds4	\$0,$E1,$ABCD1		# 0-3
628	 sha1nexte	@MSG1[1],$E1_
629	pshufb		$BSWAP,@MSG0[2]
630	prefetcht0	127(@ptr[0])
631	sha1msg1	@MSG0[1],@MSG0[0]
632	 pshufb		$BSWAP,@MSG1[2]
633	 prefetcht0	127(@ptr[1])
634	 sha1msg1	@MSG1[1],@MSG1[0]
635
636	pshufb		$BSWAP,@MSG0[3]
637	movdqa		$ABCD0,$E0
638	 pshufb		$BSWAP,@MSG1[3]
639	 movdqa		$ABCD1,$E1
640	sha1rnds4	\$0,$E0_,$ABCD0		# 4-7
641	sha1nexte	@MSG0[2],$E0
642	 sha1rnds4	\$0,$E1_,$ABCD1		# 4-7
643	 sha1nexte	@MSG1[2],$E1
644	pxor		@MSG0[2],@MSG0[0]
645	sha1msg1	@MSG0[2],@MSG0[1]
646	 pxor		@MSG1[2],@MSG1[0]
647	 sha1msg1	@MSG1[2],@MSG1[1]
648___
649for($i=2;$i<20-4;$i++) {
650$code.=<<___;
651	movdqa		$ABCD0,$E0_
652	 movdqa		$ABCD1,$E1_
653	sha1rnds4	\$`int($i/5)`,$E0,$ABCD0	# 8-11
654	sha1nexte	@MSG0[3],$E0_
655	 sha1rnds4	\$`int($i/5)`,$E1,$ABCD1	# 8-11
656	 sha1nexte	@MSG1[3],$E1_
657	sha1msg2	@MSG0[3],@MSG0[0]
658	 sha1msg2	@MSG1[3],@MSG1[0]
659	pxor		@MSG0[3],@MSG0[1]
660	sha1msg1	@MSG0[3],@MSG0[2]
661	 pxor		@MSG1[3],@MSG1[1]
662	 sha1msg1	@MSG1[3],@MSG1[2]
663___
664	($E0,$E0_)=($E0_,$E0);		($E1,$E1_)=($E1_,$E1);
665	push(@MSG0,shift(@MSG0));	push(@MSG1,shift(@MSG1));
666}
667$code.=<<___;
668	movdqa		$ABCD0,$E0_
669	 movdqa		$ABCD1,$E1_
670	sha1rnds4	\$3,$E0,$ABCD0		# 64-67
671	sha1nexte	@MSG0[3],$E0_
672	 sha1rnds4	\$3,$E1,$ABCD1		# 64-67
673	 sha1nexte	@MSG1[3],$E1_
674	sha1msg2	@MSG0[3],@MSG0[0]
675	 sha1msg2	@MSG1[3],@MSG1[0]
676	pxor		@MSG0[3],@MSG0[1]
677	 pxor		@MSG1[3],@MSG1[1]
678
679	mov		\$1,%ecx
680	pxor		@MSG0[2],@MSG0[2]	# zero
681	cmp		4*0(%rbx),%ecx		# examine counters
682	cmovge		%rsp,@ptr[0]		# cancel input
683
684	movdqa		$ABCD0,$E0
685	 movdqa		$ABCD1,$E1
686	sha1rnds4	\$3,$E0_,$ABCD0		# 68-71
687	sha1nexte	@MSG0[0],$E0
688	 sha1rnds4	\$3,$E1_,$ABCD1		# 68-71
689	 sha1nexte	@MSG1[0],$E1
690	sha1msg2	@MSG0[0],@MSG0[1]
691	 sha1msg2	@MSG1[0],@MSG1[1]
692
693	cmp		4*1(%rbx),%ecx
694	cmovge		%rsp,@ptr[1]
695	movq		(%rbx),@MSG0[0]		# pull counters
696
697	movdqa		$ABCD0,$E0_
698	 movdqa		$ABCD1,$E1_
699	sha1rnds4	\$3,$E0,$ABCD0		# 72-75
700	sha1nexte	@MSG0[1],$E0_
701	 sha1rnds4	\$3,$E1,$ABCD1		# 72-75
702	 sha1nexte	@MSG1[1],$E1_
703
704	pshufd		\$0x00,@MSG0[0],@MSG1[2]
705	pshufd		\$0x55,@MSG0[0],@MSG1[3]
706	movdqa		@MSG0[0],@MSG0[1]
707	pcmpgtd		@MSG0[2],@MSG1[2]
708	pcmpgtd		@MSG0[2],@MSG1[3]
709
710	movdqa		$ABCD0,$E0
711	 movdqa		$ABCD1,$E1
712	sha1rnds4	\$3,$E0_,$ABCD0		# 76-79
713	sha1nexte	$MSG0[2],$E0
714	 sha1rnds4	\$3,$E1_,$ABCD1		# 76-79
715	 sha1nexte	$MSG0[2],$E1
716
717	pcmpgtd		@MSG0[2],@MSG0[1]	# counter mask
718	pand		@MSG1[2],$ABCD0
719	pand		@MSG1[2],$E0
720	 pand		@MSG1[3],$ABCD1
721	 pand		@MSG1[3],$E1
722	paddd		@MSG0[1],@MSG0[0]	# counters--
723
724	paddd		0x40(%rsp),$ABCD0
725	paddd		0x50(%rsp),$E0
726	 paddd		0x60(%rsp),$ABCD1
727	 paddd		0x70(%rsp),$E1
728
729	movq		@MSG0[0],(%rbx)		# save counters
730	dec		$num
731	jnz		.Loop_shaext
732
733	mov		`$REG_SZ*17+8`(%rsp),$num
734
735	pshufd		\$0b00011011,$ABCD0,$ABCD0
736	pshufd		\$0b00011011,$ABCD1,$ABCD1
737
738	movdqa		$ABCD0,@MSG0[0]
739	punpckldq	$ABCD1,$ABCD0		# b1.b0.a1.a0
740	punpckhdq	$ABCD1,@MSG0[0]		# d1.d0.c1.c0
741	punpckhdq	$E1,$E0			# e1.e0.xx.xx
742	movq		$ABCD0,0x00-0x40($ctx)	# a1.a0
743	psrldq		\$8,$ABCD0
744	movq		@MSG0[0],0x40-0x40($ctx)# c1.c0
745	psrldq		\$8,@MSG0[0]
746	movq		$ABCD0,0x20-0x40($ctx)	# b1.b0
747	psrldq		\$8,$E0
748	movq		@MSG0[0],0x60-0x40($ctx)# d1.d0
749	movq		$E0,0x80-0x40($ctx)	# e1.e0
750
751	lea	`$REG_SZ/2`($ctx),$ctx
752	lea	`16*2`($inp),$inp
753	dec	$num
754	jnz	.Loop_grande_shaext
755
756.Ldone_shaext:
757	#mov	`$REG_SZ*17`(%rsp),%rax		# original %rsp
758___
759$code.=<<___ if ($win64);
760	movaps	-0xb8(%rax),%xmm6
761	movaps	-0xa8(%rax),%xmm7
762	movaps	-0x98(%rax),%xmm8
763	movaps	-0x88(%rax),%xmm9
764	movaps	-0x78(%rax),%xmm10
765	movaps	-0x68(%rax),%xmm11
766	movaps	-0x58(%rax),%xmm12
767	movaps	-0x48(%rax),%xmm13
768	movaps	-0x38(%rax),%xmm14
769	movaps	-0x28(%rax),%xmm15
770___
771$code.=<<___;
772	mov	-16(%rax),%rbp
773.cfi_restore	%rbp
774	mov	-8(%rax),%rbx
775.cfi_restore	%rbx
776	lea	(%rax),%rsp
777.cfi_def_cfa_register	%rsp
778.Lepilogue_shaext:
779	ret
780.cfi_endproc
781.size	sha1_multi_block_shaext,.-sha1_multi_block_shaext
782___
783						}}}
784
785						if ($avx) {{{
786sub BODY_00_19_avx {
787my ($i,$a,$b,$c,$d,$e)=@_;
788my $j=$i+1;
789my $k=$i+2;
790my $vpack = $REG_SZ==16 ? "vpunpckldq" : "vinserti128";
791my $ptr_n = $REG_SZ==16 ? @ptr[1] : @ptr[4];
792
793$code.=<<___ if ($i==0 && $REG_SZ==16);
794	vmovd		(@ptr[0]),@Xi[0]
795	 lea		`16*4`(@ptr[0]),@ptr[0]
796	vmovd		(@ptr[1]),@Xi[2]	# borrow Xi[2]
797	 lea		`16*4`(@ptr[1]),@ptr[1]
798	vpinsrd		\$1,(@ptr[2]),@Xi[0],@Xi[0]
799	 lea		`16*4`(@ptr[2]),@ptr[2]
800	vpinsrd		\$1,(@ptr[3]),@Xi[2],@Xi[2]
801	 lea		`16*4`(@ptr[3]),@ptr[3]
802	 vmovd		`4*$j-16*4`(@ptr[0]),@Xi[1]
803	vpunpckldq	@Xi[2],@Xi[0],@Xi[0]
804	 vmovd		`4*$j-16*4`($ptr_n),$t3
805	vpshufb		$tx,@Xi[0],@Xi[0]
806___
807$code.=<<___ if ($i<15 && $REG_SZ==16);		# just load input
808	 vpinsrd	\$1,`4*$j-16*4`(@ptr[2]),@Xi[1],@Xi[1]
809	 vpinsrd	\$1,`4*$j-16*4`(@ptr[3]),$t3,$t3
810___
811$code.=<<___ if ($i==0 && $REG_SZ==32);
812	vmovd		(@ptr[0]),@Xi[0]
813	 lea		`16*4`(@ptr[0]),@ptr[0]
814	vmovd		(@ptr[4]),@Xi[2]	# borrow Xi[2]
815	 lea		`16*4`(@ptr[4]),@ptr[4]
816	vmovd		(@ptr[1]),$t2
817	 lea		`16*4`(@ptr[1]),@ptr[1]
818	vmovd		(@ptr[5]),$t1
819	 lea		`16*4`(@ptr[5]),@ptr[5]
820	vpinsrd		\$1,(@ptr[2]),@Xi[0],@Xi[0]
821	 lea		`16*4`(@ptr[2]),@ptr[2]
822	vpinsrd		\$1,(@ptr[6]),@Xi[2],@Xi[2]
823	 lea		`16*4`(@ptr[6]),@ptr[6]
824	vpinsrd		\$1,(@ptr[3]),$t2,$t2
825	 lea		`16*4`(@ptr[3]),@ptr[3]
826	vpunpckldq	$t2,@Xi[0],@Xi[0]
827	vpinsrd		\$1,(@ptr[7]),$t1,$t1
828	 lea		`16*4`(@ptr[7]),@ptr[7]
829	vpunpckldq	$t1,@Xi[2],@Xi[2]
830	 vmovd		`4*$j-16*4`(@ptr[0]),@Xi[1]
831	vinserti128	@Xi[2],@Xi[0],@Xi[0]
832	 vmovd		`4*$j-16*4`($ptr_n),$t3
833	vpshufb		$tx,@Xi[0],@Xi[0]
834___
835$code.=<<___ if ($i<15 && $REG_SZ==32);		# just load input
836	 vmovd		`4*$j-16*4`(@ptr[1]),$t2
837	 vmovd		`4*$j-16*4`(@ptr[5]),$t1
838	 vpinsrd	\$1,`4*$j-16*4`(@ptr[2]),@Xi[1],@Xi[1]
839	 vpinsrd	\$1,`4*$j-16*4`(@ptr[6]),$t3,$t3
840	 vpinsrd	\$1,`4*$j-16*4`(@ptr[3]),$t2,$t2
841	 vpunpckldq	$t2,@Xi[1],@Xi[1]
842	 vpinsrd	\$1,`4*$j-16*4`(@ptr[7]),$t1,$t1
843	 vpunpckldq	$t1,$t3,$t3
844___
845$code.=<<___ if ($i<14);
846	vpaddd	$K,$e,$e			# e+=K_00_19
847	vpslld	\$5,$a,$t2
848	vpandn	$d,$b,$t1
849	vpand	$c,$b,$t0
850
851	vmovdqa	@Xi[0],`&Xi_off($i)`
852	vpaddd	@Xi[0],$e,$e			# e+=X[i]
853	 $vpack		$t3,@Xi[1],@Xi[1]
854	vpsrld	\$27,$a,$t3
855	vpxor	$t1,$t0,$t0			# Ch(b,c,d)
856	 vmovd		`4*$k-16*4`(@ptr[0]),@Xi[2]
857
858	vpslld	\$30,$b,$t1
859	vpor	$t3,$t2,$t2			# rol(a,5)
860	 vmovd		`4*$k-16*4`($ptr_n),$t3
861	vpaddd	$t0,$e,$e			# e+=Ch(b,c,d)
862
863	vpsrld	\$2,$b,$b
864	vpaddd	$t2,$e,$e			# e+=rol(a,5)
865	 vpshufb	$tx,@Xi[1],@Xi[1]
866	vpor	$t1,$b,$b			# b=rol(b,30)
867___
868$code.=<<___ if ($i==14);
869	vpaddd	$K,$e,$e			# e+=K_00_19
870	 prefetcht0	63(@ptr[0])
871	vpslld	\$5,$a,$t2
872	vpandn	$d,$b,$t1
873	vpand	$c,$b,$t0
874
875	vmovdqa	@Xi[0],`&Xi_off($i)`
876	vpaddd	@Xi[0],$e,$e			# e+=X[i]
877	 $vpack		$t3,@Xi[1],@Xi[1]
878	vpsrld	\$27,$a,$t3
879	 prefetcht0	63(@ptr[1])
880	vpxor	$t1,$t0,$t0			# Ch(b,c,d)
881
882	vpslld	\$30,$b,$t1
883	vpor	$t3,$t2,$t2			# rol(a,5)
884	 prefetcht0	63(@ptr[2])
885	vpaddd	$t0,$e,$e			# e+=Ch(b,c,d)
886
887	vpsrld	\$2,$b,$b
888	vpaddd	$t2,$e,$e			# e+=rol(a,5)
889	 prefetcht0	63(@ptr[3])
890	 vpshufb	$tx,@Xi[1],@Xi[1]
891	vpor	$t1,$b,$b			# b=rol(b,30)
892___
893$code.=<<___ if ($i>=13 && $i<15);
894	vmovdqa	`&Xi_off($j+2)`,@Xi[3]		# preload "X[2]"
895___
896$code.=<<___ if ($i>=15);			# apply Xupdate
897	vpxor	@Xi[-2],@Xi[1],@Xi[1]		# "X[13]"
898	vmovdqa	`&Xi_off($j+2)`,@Xi[3]		# "X[2]"
899
900	vpaddd	$K,$e,$e			# e+=K_00_19
901	vpslld	\$5,$a,$t2
902	vpandn	$d,$b,$t1
903	 `"prefetcht0	63(@ptr[4])"		if ($i==15 && $REG_SZ==32)`
904	vpand	$c,$b,$t0
905
906	vmovdqa	@Xi[0],`&Xi_off($i)`
907	vpaddd	@Xi[0],$e,$e			# e+=X[i]
908	 vpxor	`&Xi_off($j+8)`,@Xi[1],@Xi[1]
909	vpsrld	\$27,$a,$t3
910	vpxor	$t1,$t0,$t0			# Ch(b,c,d)
911	 vpxor	@Xi[3],@Xi[1],@Xi[1]
912	 `"prefetcht0	63(@ptr[5])"		if ($i==15 && $REG_SZ==32)`
913
914	vpslld	\$30,$b,$t1
915	vpor	$t3,$t2,$t2			# rol(a,5)
916	vpaddd	$t0,$e,$e			# e+=Ch(b,c,d)
917	 `"prefetcht0	63(@ptr[6])"		if ($i==15 && $REG_SZ==32)`
918	 vpsrld	\$31,@Xi[1],$tx
919	 vpaddd	@Xi[1],@Xi[1],@Xi[1]
920
921	vpsrld	\$2,$b,$b
922	 `"prefetcht0	63(@ptr[7])"		if ($i==15 && $REG_SZ==32)`
923	vpaddd	$t2,$e,$e			# e+=rol(a,5)
924	 vpor	$tx,@Xi[1],@Xi[1]		# rol	\$1,@Xi[1]
925	vpor	$t1,$b,$b			# b=rol(b,30)
926___
927push(@Xi,shift(@Xi));
928}
929
930sub BODY_20_39_avx {
931my ($i,$a,$b,$c,$d,$e)=@_;
932my $j=$i+1;
933
934$code.=<<___ if ($i<79);
935	vpxor	@Xi[-2],@Xi[1],@Xi[1]		# "X[13]"
936	vmovdqa	`&Xi_off($j+2)`,@Xi[3]		# "X[2]"
937
938	vpslld	\$5,$a,$t2
939	vpaddd	$K,$e,$e			# e+=K_20_39
940	vpxor	$b,$d,$t0
941___
942$code.=<<___ if ($i<72);
943	vmovdqa	@Xi[0],`&Xi_off($i)`
944___
945$code.=<<___ if ($i<79);
946	vpaddd	@Xi[0],$e,$e			# e+=X[i]
947	 vpxor	`&Xi_off($j+8)`,@Xi[1],@Xi[1]
948	vpsrld	\$27,$a,$t3
949	vpxor	$c,$t0,$t0			# Parity(b,c,d)
950	 vpxor	@Xi[3],@Xi[1],@Xi[1]
951
952	vpslld	\$30,$b,$t1
953	vpor	$t3,$t2,$t2			# rol(a,5)
954	vpaddd	$t0,$e,$e			# e+=Parity(b,c,d)
955	 vpsrld	\$31,@Xi[1],$tx
956	 vpaddd	@Xi[1],@Xi[1],@Xi[1]
957
958	vpsrld	\$2,$b,$b
959	vpaddd	$t2,$e,$e			# e+=rol(a,5)
960	 vpor	$tx,@Xi[1],@Xi[1]		# rol(@Xi[1],1)
961	vpor	$t1,$b,$b			# b=rol(b,30)
962___
963$code.=<<___ if ($i==79);
964	vpslld	\$5,$a,$t2
965	vpaddd	$K,$e,$e			# e+=K_20_39
966	vpxor	$b,$d,$t0
967
968	vpsrld	\$27,$a,$t3
969	vpaddd	@Xi[0],$e,$e			# e+=X[i]
970	vpxor	$c,$t0,$t0			# Parity(b,c,d)
971
972	vpslld	\$30,$b,$t1
973	vpor	$t3,$t2,$t2			# rol(a,5)
974	vpaddd	$t0,$e,$e			# e+=Parity(b,c,d)
975
976	vpsrld	\$2,$b,$b
977	vpaddd	$t2,$e,$e			# e+=rol(a,5)
978	vpor	$t1,$b,$b			# b=rol(b,30)
979___
980push(@Xi,shift(@Xi));
981}
982
983sub BODY_40_59_avx {
984my ($i,$a,$b,$c,$d,$e)=@_;
985my $j=$i+1;
986
987$code.=<<___;
988	vpxor	@Xi[-2],@Xi[1],@Xi[1]		# "X[13]"
989	vmovdqa	`&Xi_off($j+2)`,@Xi[3]		# "X[2]"
990
991	vpaddd	$K,$e,$e			# e+=K_40_59
992	vpslld	\$5,$a,$t2
993	vpand	$c,$d,$t1
994	 vpxor	`&Xi_off($j+8)`,@Xi[1],@Xi[1]
995
996	vpaddd	$t1,$e,$e
997	vpsrld	\$27,$a,$t3
998	vpxor	$c,$d,$t0
999	 vpxor	@Xi[3],@Xi[1],@Xi[1]
1000
1001	vmovdqu	@Xi[0],`&Xi_off($i)`
1002	vpaddd	@Xi[0],$e,$e			# e+=X[i]
1003	vpor	$t3,$t2,$t2			# rol(a,5)
1004	 vpsrld	\$31,@Xi[1],$tx
1005	vpand	$b,$t0,$t0
1006	 vpaddd	@Xi[1],@Xi[1],@Xi[1]
1007
1008	vpslld	\$30,$b,$t1
1009	vpaddd	$t0,$e,$e			# e+=Maj(b,d,c)
1010
1011	vpsrld	\$2,$b,$b
1012	vpaddd	$t2,$e,$e			# e+=rol(a,5)
1013	 vpor	$tx,@Xi[1],@Xi[1]		# rol(@X[1],1)
1014	vpor	$t1,$b,$b			# b=rol(b,30)
1015___
1016push(@Xi,shift(@Xi));
1017}
1018
1019$code.=<<___;
1020.type	sha1_multi_block_avx,\@function,3
1021.align	32
1022sha1_multi_block_avx:
1023.cfi_startproc
1024_avx_shortcut:
1025___
1026$code.=<<___ if ($avx>1);
1027	shr	\$32,%rcx
1028	cmp	\$2,$num
1029	jb	.Lavx
1030	test	\$`1<<5`,%ecx
1031	jnz	_avx2_shortcut
1032	jmp	.Lavx
1033.align	32
1034.Lavx:
1035___
1036$code.=<<___;
1037	mov	%rsp,%rax
1038.cfi_def_cfa_register	%rax
1039	push	%rbx
1040.cfi_push	%rbx
1041	push	%rbp
1042.cfi_push	%rbp
1043___
1044$code.=<<___ if ($win64);
1045	lea	-0xa8(%rsp),%rsp
1046	movaps	%xmm6,(%rsp)
1047	movaps	%xmm7,0x10(%rsp)
1048	movaps	%xmm8,0x20(%rsp)
1049	movaps	%xmm9,0x30(%rsp)
1050	movaps	%xmm10,-0x78(%rax)
1051	movaps	%xmm11,-0x68(%rax)
1052	movaps	%xmm12,-0x58(%rax)
1053	movaps	%xmm13,-0x48(%rax)
1054	movaps	%xmm14,-0x38(%rax)
1055	movaps	%xmm15,-0x28(%rax)
1056___
1057$code.=<<___;
1058	sub	\$`$REG_SZ*18`, %rsp
1059	and	\$-256,%rsp
1060	mov	%rax,`$REG_SZ*17`(%rsp)		# original %rsp
1061.cfi_cfa_expression	%rsp+`$REG_SZ*17`,deref,+8
1062.Lbody_avx:
1063	lea	K_XX_XX(%rip),$Tbl
1064	lea	`$REG_SZ*16`(%rsp),%rbx
1065
1066	vzeroupper
1067.Loop_grande_avx:
1068	mov	$num,`$REG_SZ*17+8`(%rsp)	# original $num
1069	xor	$num,$num
1070___
1071for($i=0;$i<4;$i++) {
1072    $code.=<<___;
1073	mov	`16*$i+0`($inp),@ptr[$i]	# input pointer
1074	mov	`16*$i+8`($inp),%ecx		# number of blocks
1075	cmp	$num,%ecx
1076	cmovg	%ecx,$num			# find maximum
1077	test	%ecx,%ecx
1078	mov	%ecx,`4*$i`(%rbx)		# initialize counters
1079	cmovle	$Tbl,@ptr[$i]			# cancel input
1080___
1081}
1082$code.=<<___;
1083	test	$num,$num
1084	jz	.Ldone_avx
1085
1086	vmovdqu	0x00($ctx),$A			# load context
1087	 lea	128(%rsp),%rax
1088	vmovdqu	0x20($ctx),$B
1089	vmovdqu	0x40($ctx),$C
1090	vmovdqu	0x60($ctx),$D
1091	vmovdqu	0x80($ctx),$E
1092	vmovdqu	0x60($Tbl),$tx			# pbswap_mask
1093	jmp	.Loop_avx
1094
1095.align	32
1096.Loop_avx:
1097___
1098$code.="	vmovdqa	-0x20($Tbl),$K\n";	# K_00_19
1099for($i=0;$i<20;$i++)	{ &BODY_00_19_avx($i,@V); unshift(@V,pop(@V)); }
1100$code.="	vmovdqa	0x00($Tbl),$K\n";	# K_20_39
1101for(;$i<40;$i++)	{ &BODY_20_39_avx($i,@V); unshift(@V,pop(@V)); }
1102$code.="	vmovdqa	0x20($Tbl),$K\n";	# K_40_59
1103for(;$i<60;$i++)	{ &BODY_40_59_avx($i,@V); unshift(@V,pop(@V)); }
1104$code.="	vmovdqa	0x40($Tbl),$K\n";	# K_60_79
1105for(;$i<80;$i++)	{ &BODY_20_39_avx($i,@V); unshift(@V,pop(@V)); }
1106$code.=<<___;
1107	mov	\$1,%ecx
1108___
1109for($i=0;$i<4;$i++) {
1110    $code.=<<___;
1111	cmp	`4*$i`(%rbx),%ecx		# examine counters
1112	cmovge	$Tbl,@ptr[$i]			# cancel input
1113___
1114}
1115$code.=<<___;
1116	vmovdqu	(%rbx),$t0			# pull counters
1117	vpxor	$t2,$t2,$t2
1118	vmovdqa	$t0,$t1
1119	vpcmpgtd $t2,$t1,$t1			# mask value
1120	vpaddd	$t1,$t0,$t0			# counters--
1121
1122	vpand	$t1,$A,$A
1123	vpand	$t1,$B,$B
1124	vpaddd	0x00($ctx),$A,$A
1125	vpand	$t1,$C,$C
1126	vpaddd	0x20($ctx),$B,$B
1127	vpand	$t1,$D,$D
1128	vpaddd	0x40($ctx),$C,$C
1129	vpand	$t1,$E,$E
1130	vpaddd	0x60($ctx),$D,$D
1131	vpaddd	0x80($ctx),$E,$E
1132	vmovdqu	$A,0x00($ctx)
1133	vmovdqu	$B,0x20($ctx)
1134	vmovdqu	$C,0x40($ctx)
1135	vmovdqu	$D,0x60($ctx)
1136	vmovdqu	$E,0x80($ctx)
1137
1138	vmovdqu	$t0,(%rbx)			# save counters
1139	vmovdqu	0x60($Tbl),$tx			# pbswap_mask
1140	dec	$num
1141	jnz	.Loop_avx
1142
1143	mov	`$REG_SZ*17+8`(%rsp),$num
1144	lea	$REG_SZ($ctx),$ctx
1145	lea	`16*$REG_SZ/4`($inp),$inp
1146	dec	$num
1147	jnz	.Loop_grande_avx
1148
1149.Ldone_avx:
1150	mov	`$REG_SZ*17`(%rsp),%rax		# original %rsp
1151.cfi_def_cfa	%rax,8
1152	vzeroupper
1153___
1154$code.=<<___ if ($win64);
1155	movaps	-0xb8(%rax),%xmm6
1156	movaps	-0xa8(%rax),%xmm7
1157	movaps	-0x98(%rax),%xmm8
1158	movaps	-0x88(%rax),%xmm9
1159	movaps	-0x78(%rax),%xmm10
1160	movaps	-0x68(%rax),%xmm11
1161	movaps	-0x58(%rax),%xmm12
1162	movaps	-0x48(%rax),%xmm13
1163	movaps	-0x38(%rax),%xmm14
1164	movaps	-0x28(%rax),%xmm15
1165___
1166$code.=<<___;
1167	mov	-16(%rax),%rbp
1168.cfi_restore	%rbp
1169	mov	-8(%rax),%rbx
1170.cfi_restore	%rbx
1171	lea	(%rax),%rsp
1172.cfi_def_cfa_register	%rsp
1173.Lepilogue_avx:
1174	ret
1175.cfi_endproc
1176.size	sha1_multi_block_avx,.-sha1_multi_block_avx
1177___
1178
1179						if ($avx>1) {
1180$code =~ s/\`([^\`]*)\`/eval $1/gem;
1181
1182$REG_SZ=32;
1183
1184@ptr=map("%r$_",(12..15,8..11));
1185
1186@V=($A,$B,$C,$D,$E)=map("%ymm$_",(0..4));
1187($t0,$t1,$t2,$t3,$tx)=map("%ymm$_",(5..9));
1188@Xi=map("%ymm$_",(10..14));
1189$K="%ymm15";
1190
1191$code.=<<___;
1192.type	sha1_multi_block_avx2,\@function,3
1193.align	32
1194sha1_multi_block_avx2:
1195.cfi_startproc
1196_avx2_shortcut:
1197	mov	%rsp,%rax
1198.cfi_def_cfa_register	%rax
1199	push	%rbx
1200.cfi_push	%rbx
1201	push	%rbp
1202.cfi_push	%rbp
1203	push	%r12
1204.cfi_push	%r12
1205	push	%r13
1206.cfi_push	%r13
1207	push	%r14
1208.cfi_push	%r14
1209	push	%r15
1210.cfi_push	%r15
1211___
1212$code.=<<___ if ($win64);
1213	lea	-0xa8(%rsp),%rsp
1214	movaps	%xmm6,(%rsp)
1215	movaps	%xmm7,0x10(%rsp)
1216	movaps	%xmm8,0x20(%rsp)
1217	movaps	%xmm9,0x30(%rsp)
1218	movaps	%xmm10,0x40(%rsp)
1219	movaps	%xmm11,0x50(%rsp)
1220	movaps	%xmm12,-0x78(%rax)
1221	movaps	%xmm13,-0x68(%rax)
1222	movaps	%xmm14,-0x58(%rax)
1223	movaps	%xmm15,-0x48(%rax)
1224___
1225$code.=<<___;
1226	sub	\$`$REG_SZ*18`, %rsp
1227	and	\$-256,%rsp
1228	mov	%rax,`$REG_SZ*17`(%rsp)		# original %rsp
1229.cfi_cfa_expression	%rsp+`$REG_SZ*17`,deref,+8
1230.Lbody_avx2:
1231	lea	K_XX_XX(%rip),$Tbl
1232	shr	\$1,$num
1233
1234	vzeroupper
1235.Loop_grande_avx2:
1236	mov	$num,`$REG_SZ*17+8`(%rsp)	# original $num
1237	xor	$num,$num
1238	lea	`$REG_SZ*16`(%rsp),%rbx
1239___
1240for($i=0;$i<8;$i++) {
1241    $code.=<<___;
1242	mov	`16*$i+0`($inp),@ptr[$i]	# input pointer
1243	mov	`16*$i+8`($inp),%ecx		# number of blocks
1244	cmp	$num,%ecx
1245	cmovg	%ecx,$num			# find maximum
1246	test	%ecx,%ecx
1247	mov	%ecx,`4*$i`(%rbx)		# initialize counters
1248	cmovle	$Tbl,@ptr[$i]			# cancel input
1249___
1250}
1251$code.=<<___;
1252	vmovdqu	0x00($ctx),$A			# load context
1253	 lea	128(%rsp),%rax
1254	vmovdqu	0x20($ctx),$B
1255	 lea	256+128(%rsp),%rbx
1256	vmovdqu	0x40($ctx),$C
1257	vmovdqu	0x60($ctx),$D
1258	vmovdqu	0x80($ctx),$E
1259	vmovdqu	0x60($Tbl),$tx			# pbswap_mask
1260	jmp	.Loop_avx2
1261
1262.align	32
1263.Loop_avx2:
1264___
1265$code.="	vmovdqa	-0x20($Tbl),$K\n";	# K_00_19
1266for($i=0;$i<20;$i++)	{ &BODY_00_19_avx($i,@V); unshift(@V,pop(@V)); }
1267$code.="	vmovdqa	0x00($Tbl),$K\n";	# K_20_39
1268for(;$i<40;$i++)	{ &BODY_20_39_avx($i,@V); unshift(@V,pop(@V)); }
1269$code.="	vmovdqa	0x20($Tbl),$K\n";	# K_40_59
1270for(;$i<60;$i++)	{ &BODY_40_59_avx($i,@V); unshift(@V,pop(@V)); }
1271$code.="	vmovdqa	0x40($Tbl),$K\n";	# K_60_79
1272for(;$i<80;$i++)	{ &BODY_20_39_avx($i,@V); unshift(@V,pop(@V)); }
1273$code.=<<___;
1274	mov	\$1,%ecx
1275	lea	`$REG_SZ*16`(%rsp),%rbx
1276___
1277for($i=0;$i<8;$i++) {
1278    $code.=<<___;
1279	cmp	`4*$i`(%rbx),%ecx		# examine counters
1280	cmovge	$Tbl,@ptr[$i]			# cancel input
1281___
1282}
1283$code.=<<___;
1284	vmovdqu	(%rbx),$t0		# pull counters
1285	vpxor	$t2,$t2,$t2
1286	vmovdqa	$t0,$t1
1287	vpcmpgtd $t2,$t1,$t1			# mask value
1288	vpaddd	$t1,$t0,$t0			# counters--
1289
1290	vpand	$t1,$A,$A
1291	vpand	$t1,$B,$B
1292	vpaddd	0x00($ctx),$A,$A
1293	vpand	$t1,$C,$C
1294	vpaddd	0x20($ctx),$B,$B
1295	vpand	$t1,$D,$D
1296	vpaddd	0x40($ctx),$C,$C
1297	vpand	$t1,$E,$E
1298	vpaddd	0x60($ctx),$D,$D
1299	vpaddd	0x80($ctx),$E,$E
1300	vmovdqu	$A,0x00($ctx)
1301	vmovdqu	$B,0x20($ctx)
1302	vmovdqu	$C,0x40($ctx)
1303	vmovdqu	$D,0x60($ctx)
1304	vmovdqu	$E,0x80($ctx)
1305
1306	vmovdqu	$t0,(%rbx)			# save counters
1307	lea	256+128(%rsp),%rbx
1308	vmovdqu	0x60($Tbl),$tx			# pbswap_mask
1309	dec	$num
1310	jnz	.Loop_avx2
1311
1312	#mov	`$REG_SZ*17+8`(%rsp),$num
1313	#lea	$REG_SZ($ctx),$ctx
1314	#lea	`16*$REG_SZ/4`($inp),$inp
1315	#dec	$num
1316	#jnz	.Loop_grande_avx2
1317
1318.Ldone_avx2:
1319	mov	`$REG_SZ*17`(%rsp),%rax		# original %rsp
1320.cfi_def_cfa	%rax,8
1321	vzeroupper
1322___
1323$code.=<<___ if ($win64);
1324	movaps	-0xd8(%rax),%xmm6
1325	movaps	-0xc8(%rax),%xmm7
1326	movaps	-0xb8(%rax),%xmm8
1327	movaps	-0xa8(%rax),%xmm9
1328	movaps	-0x98(%rax),%xmm10
1329	movaps	-0x88(%rax),%xmm11
1330	movaps	-0x78(%rax),%xmm12
1331	movaps	-0x68(%rax),%xmm13
1332	movaps	-0x58(%rax),%xmm14
1333	movaps	-0x48(%rax),%xmm15
1334___
1335$code.=<<___;
1336	mov	-48(%rax),%r15
1337.cfi_restore	%r15
1338	mov	-40(%rax),%r14
1339.cfi_restore	%r14
1340	mov	-32(%rax),%r13
1341.cfi_restore	%r13
1342	mov	-24(%rax),%r12
1343.cfi_restore	%r12
1344	mov	-16(%rax),%rbp
1345.cfi_restore	%rbp
1346	mov	-8(%rax),%rbx
1347.cfi_restore	%rbx
1348	lea	(%rax),%rsp
1349.cfi_def_cfa_register	%rsp
1350.Lepilogue_avx2:
1351	ret
1352.cfi_endproc
1353.size	sha1_multi_block_avx2,.-sha1_multi_block_avx2
1354___
1355						}	}}}
1356$code.=<<___;
1357
1358.align	256
1359	.long	0x5a827999,0x5a827999,0x5a827999,0x5a827999	# K_00_19
1360	.long	0x5a827999,0x5a827999,0x5a827999,0x5a827999	# K_00_19
1361K_XX_XX:
1362	.long	0x6ed9eba1,0x6ed9eba1,0x6ed9eba1,0x6ed9eba1	# K_20_39
1363	.long	0x6ed9eba1,0x6ed9eba1,0x6ed9eba1,0x6ed9eba1	# K_20_39
1364	.long	0x8f1bbcdc,0x8f1bbcdc,0x8f1bbcdc,0x8f1bbcdc	# K_40_59
1365	.long	0x8f1bbcdc,0x8f1bbcdc,0x8f1bbcdc,0x8f1bbcdc	# K_40_59
1366	.long	0xca62c1d6,0xca62c1d6,0xca62c1d6,0xca62c1d6	# K_60_79
1367	.long	0xca62c1d6,0xca62c1d6,0xca62c1d6,0xca62c1d6	# K_60_79
1368	.long	0x00010203,0x04050607,0x08090a0b,0x0c0d0e0f	# pbswap
1369	.long	0x00010203,0x04050607,0x08090a0b,0x0c0d0e0f	# pbswap
1370	.byte	0xf,0xe,0xd,0xc,0xb,0xa,0x9,0x8,0x7,0x6,0x5,0x4,0x3,0x2,0x1,0x0
1371	.asciz	"SHA1 multi-block transform for x86_64, CRYPTOGAMS by <appro\@openssl.org>"
1372___
1373
1374if ($win64) {
1375# EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
1376#		CONTEXT *context,DISPATCHER_CONTEXT *disp)
1377$rec="%rcx";
1378$frame="%rdx";
1379$context="%r8";
1380$disp="%r9";
1381
1382$code.=<<___;
1383.extern	__imp_RtlVirtualUnwind
1384.type	se_handler,\@abi-omnipotent
1385.align	16
1386se_handler:
1387	push	%rsi
1388	push	%rdi
1389	push	%rbx
1390	push	%rbp
1391	push	%r12
1392	push	%r13
1393	push	%r14
1394	push	%r15
1395	pushfq
1396	sub	\$64,%rsp
1397
1398	mov	120($context),%rax	# pull context->Rax
1399	mov	248($context),%rbx	# pull context->Rip
1400
1401	mov	8($disp),%rsi		# disp->ImageBase
1402	mov	56($disp),%r11		# disp->HandlerData
1403
1404	mov	0(%r11),%r10d		# HandlerData[0]
1405	lea	(%rsi,%r10),%r10	# end of prologue label
1406	cmp	%r10,%rbx		# context->Rip<.Lbody
1407	jb	.Lin_prologue
1408
1409	mov	152($context),%rax	# pull context->Rsp
1410
1411	mov	4(%r11),%r10d		# HandlerData[1]
1412	lea	(%rsi,%r10),%r10	# epilogue label
1413	cmp	%r10,%rbx		# context->Rip>=.Lepilogue
1414	jae	.Lin_prologue
1415
1416	mov	`16*17`(%rax),%rax	# pull saved stack pointer
1417
1418	mov	-8(%rax),%rbx
1419	mov	-16(%rax),%rbp
1420	mov	%rbx,144($context)	# restore context->Rbx
1421	mov	%rbp,160($context)	# restore context->Rbp
1422
1423	lea	-24-10*16(%rax),%rsi
1424	lea	512($context),%rdi	# &context.Xmm6
1425	mov	\$20,%ecx
1426	.long	0xa548f3fc		# cld; rep movsq
1427
1428.Lin_prologue:
1429	mov	8(%rax),%rdi
1430	mov	16(%rax),%rsi
1431	mov	%rax,152($context)	# restore context->Rsp
1432	mov	%rsi,168($context)	# restore context->Rsi
1433	mov	%rdi,176($context)	# restore context->Rdi
1434
1435	mov	40($disp),%rdi		# disp->ContextRecord
1436	mov	$context,%rsi		# context
1437	mov	\$154,%ecx		# sizeof(CONTEXT)
1438	.long	0xa548f3fc		# cld; rep movsq
1439
1440	mov	$disp,%rsi
1441	xor	%rcx,%rcx		# arg1, UNW_FLAG_NHANDLER
1442	mov	8(%rsi),%rdx		# arg2, disp->ImageBase
1443	mov	0(%rsi),%r8		# arg3, disp->ControlPc
1444	mov	16(%rsi),%r9		# arg4, disp->FunctionEntry
1445	mov	40(%rsi),%r10		# disp->ContextRecord
1446	lea	56(%rsi),%r11		# &disp->HandlerData
1447	lea	24(%rsi),%r12		# &disp->EstablisherFrame
1448	mov	%r10,32(%rsp)		# arg5
1449	mov	%r11,40(%rsp)		# arg6
1450	mov	%r12,48(%rsp)		# arg7
1451	mov	%rcx,56(%rsp)		# arg8, (NULL)
1452	call	*__imp_RtlVirtualUnwind(%rip)
1453
1454	mov	\$1,%eax		# ExceptionContinueSearch
1455	add	\$64,%rsp
1456	popfq
1457	pop	%r15
1458	pop	%r14
1459	pop	%r13
1460	pop	%r12
1461	pop	%rbp
1462	pop	%rbx
1463	pop	%rdi
1464	pop	%rsi
1465	ret
1466.size	se_handler,.-se_handler
1467___
1468$code.=<<___ if ($avx>1);
1469.type	avx2_handler,\@abi-omnipotent
1470.align	16
1471avx2_handler:
1472	push	%rsi
1473	push	%rdi
1474	push	%rbx
1475	push	%rbp
1476	push	%r12
1477	push	%r13
1478	push	%r14
1479	push	%r15
1480	pushfq
1481	sub	\$64,%rsp
1482
1483	mov	120($context),%rax	# pull context->Rax
1484	mov	248($context),%rbx	# pull context->Rip
1485
1486	mov	8($disp),%rsi		# disp->ImageBase
1487	mov	56($disp),%r11		# disp->HandlerData
1488
1489	mov	0(%r11),%r10d		# HandlerData[0]
1490	lea	(%rsi,%r10),%r10	# end of prologue label
1491	cmp	%r10,%rbx		# context->Rip<body label
1492	jb	.Lin_prologue
1493
1494	mov	152($context),%rax	# pull context->Rsp
1495
1496	mov	4(%r11),%r10d		# HandlerData[1]
1497	lea	(%rsi,%r10),%r10	# epilogue label
1498	cmp	%r10,%rbx		# context->Rip>=epilogue label
1499	jae	.Lin_prologue
1500
1501	mov	`32*17`($context),%rax	# pull saved stack pointer
1502
1503	mov	-8(%rax),%rbx
1504	mov	-16(%rax),%rbp
1505	mov	-24(%rax),%r12
1506	mov	-32(%rax),%r13
1507	mov	-40(%rax),%r14
1508	mov	-48(%rax),%r15
1509	mov	%rbx,144($context)	# restore context->Rbx
1510	mov	%rbp,160($context)	# restore context->Rbp
1511	mov	%r12,216($context)	# restore context->R12
1512	mov	%r13,224($context)	# restore context->R13
1513	mov	%r14,232($context)	# restore context->R14
1514	mov	%r15,240($context)	# restore context->R15
1515
1516	lea	-56-10*16(%rax),%rsi
1517	lea	512($context),%rdi	# &context.Xmm6
1518	mov	\$20,%ecx
1519	.long	0xa548f3fc		# cld; rep movsq
1520
1521	jmp	.Lin_prologue
1522.size	avx2_handler,.-avx2_handler
1523___
1524$code.=<<___;
1525.section	.pdata
1526.align	4
1527	.rva	.LSEH_begin_sha1_multi_block
1528	.rva	.LSEH_end_sha1_multi_block
1529	.rva	.LSEH_info_sha1_multi_block
1530	.rva	.LSEH_begin_sha1_multi_block_shaext
1531	.rva	.LSEH_end_sha1_multi_block_shaext
1532	.rva	.LSEH_info_sha1_multi_block_shaext
1533___
1534$code.=<<___ if ($avx);
1535	.rva	.LSEH_begin_sha1_multi_block_avx
1536	.rva	.LSEH_end_sha1_multi_block_avx
1537	.rva	.LSEH_info_sha1_multi_block_avx
1538___
1539$code.=<<___ if ($avx>1);
1540	.rva	.LSEH_begin_sha1_multi_block_avx2
1541	.rva	.LSEH_end_sha1_multi_block_avx2
1542	.rva	.LSEH_info_sha1_multi_block_avx2
1543___
1544$code.=<<___;
1545.section	.xdata
1546.align	8
1547.LSEH_info_sha1_multi_block:
1548	.byte	9,0,0,0
1549	.rva	se_handler
1550	.rva	.Lbody,.Lepilogue			# HandlerData[]
1551.LSEH_info_sha1_multi_block_shaext:
1552	.byte	9,0,0,0
1553	.rva	se_handler
1554	.rva	.Lbody_shaext,.Lepilogue_shaext	# HandlerData[]
1555___
1556$code.=<<___ if ($avx);
1557.LSEH_info_sha1_multi_block_avx:
1558	.byte	9,0,0,0
1559	.rva	se_handler
1560	.rva	.Lbody_avx,.Lepilogue_avx		# HandlerData[]
1561___
1562$code.=<<___ if ($avx>1);
1563.LSEH_info_sha1_multi_block_avx2:
1564	.byte	9,0,0,0
1565	.rva	avx2_handler
1566	.rva	.Lbody_avx2,.Lepilogue_avx2		# HandlerData[]
1567___
1568}
1569####################################################################
1570
1571sub rex {
1572  local *opcode=shift;
1573  my ($dst,$src)=@_;
1574  my $rex=0;
1575
1576    $rex|=0x04			if ($dst>=8);
1577    $rex|=0x01			if ($src>=8);
1578    unshift @opcode,$rex|0x40	if ($rex);
1579}
1580
1581sub sha1rnds4 {
1582    if (@_[0] =~ /\$([x0-9a-f]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) {
1583      my @opcode=(0x0f,0x3a,0xcc);
1584	rex(\@opcode,$3,$2);
1585	push @opcode,0xc0|($2&7)|(($3&7)<<3);		# ModR/M
1586	my $c=$1;
1587	push @opcode,$c=~/^0/?oct($c):$c;
1588	return ".byte\t".join(',',@opcode);
1589    } else {
1590	return "sha1rnds4\t".@_[0];
1591    }
1592}
1593
1594sub sha1op38 {
1595    my $instr = shift;
1596    my %opcodelet = (
1597		"sha1nexte" => 0xc8,
1598  		"sha1msg1"  => 0xc9,
1599		"sha1msg2"  => 0xca	);
1600
1601    if (defined($opcodelet{$instr}) && @_[0] =~ /%xmm([0-9]+),\s*%xmm([0-9]+)/) {
1602      my @opcode=(0x0f,0x38);
1603	rex(\@opcode,$2,$1);
1604	push @opcode,$opcodelet{$instr};
1605	push @opcode,0xc0|($1&7)|(($2&7)<<3);		# ModR/M
1606	return ".byte\t".join(',',@opcode);
1607    } else {
1608	return $instr."\t".@_[0];
1609    }
1610}
1611
1612foreach (split("\n",$code)) {
1613	s/\`([^\`]*)\`/eval($1)/ge;
1614
1615	s/\b(sha1rnds4)\s+(.*)/sha1rnds4($2)/geo		or
1616	s/\b(sha1[^\s]*)\s+(.*)/sha1op38($1,$2)/geo		or
1617
1618	s/\b(vmov[dq])\b(.+)%ymm([0-9]+)/$1$2%xmm$3/go		or
1619	s/\b(vmovdqu)\b(.+)%x%ymm([0-9]+)/$1$2%xmm$3/go		or
1620	s/\b(vpinsr[qd])\b(.+)%ymm([0-9]+),%ymm([0-9]+)/$1$2%xmm$3,%xmm$4/go	or
1621	s/\b(vpextr[qd])\b(.+)%ymm([0-9]+)/$1$2%xmm$3/go	or
1622	s/\b(vinserti128)\b(\s+)%ymm/$1$2\$1,%xmm/go		or
1623	s/\b(vpbroadcast[qd]\s+)%ymm([0-9]+)/$1%xmm$2/go;
1624
1625	print $_,"\n";
1626}
1627
1628close STDOUT or die "error closing STDOUT: $!";
1629