• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1#! /usr/bin/env perl
2# Copyright 2014-2020 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# Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
11# project. The module is, however, dual licensed under OpenSSL and
12# CRYPTOGAMS licenses depending on where you obtain it. For further
13# details see http://www.openssl.org/~appro/cryptogams/.
14#
15# Permission to use under GPLv2 terms is granted.
16# ====================================================================
17#
18# SHA256/512 for ARMv8.
19#
20# Performance in cycles per processed byte and improvement coefficient
21# over code generated with "default" compiler:
22#
23#		SHA256-hw	SHA256(*)	SHA512
24# Apple A7	1.97		10.5 (+33%)	6.73 (-1%(**))
25# Cortex-A53	2.38		15.5 (+115%)	10.0 (+150%(***))
26# Cortex-A57	2.31		11.6 (+86%)	7.51 (+260%(***))
27# Denver	2.01		10.5 (+26%)	6.70 (+8%)
28# X-Gene			20.0 (+100%)	12.8 (+300%(***))
29# Mongoose	2.36		13.0 (+50%)	8.36 (+33%)
30# Kryo		1.92		17.4 (+30%)	11.2 (+8%)
31#
32# (*)	Software SHA256 results are of lesser relevance, presented
33#	mostly for informational purposes.
34# (**)	The result is a trade-off: it's possible to improve it by
35#	10% (or by 1 cycle per round), but at the cost of 20% loss
36#	on Cortex-A53 (or by 4 cycles per round).
37# (***)	Super-impressive coefficients over gcc-generated code are
38#	indication of some compiler "pathology", most notably code
39#	generated with -mgeneral-regs-only is significantly faster
40#	and the gap is only 40-90%.
41
42my ($flavour, $output) = @ARGV;
43
44if ($output =~ /sha512-armv8/) {
45	$BITS=512;
46	$SZ=8;
47	@Sigma0=(28,34,39);
48	@Sigma1=(14,18,41);
49	@sigma0=(1,  8, 7);
50	@sigma1=(19,61, 6);
51	$rounds=80;
52	$reg_t="x";
53} else {
54	$BITS=256;
55	$SZ=4;
56	@Sigma0=( 2,13,22);
57	@Sigma1=( 6,11,25);
58	@sigma0=( 7,18, 3);
59	@sigma1=(17,19,10);
60	$rounds=64;
61	$reg_t="w";
62}
63
64if ($flavour && $flavour ne "void") {
65    $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
66    ( $xlate="${dir}arm-xlate.pl" and -f $xlate ) or
67    ( $xlate="${dir}../../../perlasm/arm-xlate.pl" and -f $xlate) or
68    die "can't locate arm-xlate.pl";
69
70    open OUT,"| \"$^X\" \"$xlate\" $flavour \"$output\"";
71    *STDOUT=*OUT;
72} else {
73    open OUT,">$output";
74    *STDOUT=*OUT;
75}
76
77$func="sha${BITS}_block_data_order";
78
79($ctx,$inp,$num,$Ktbl)=map("x$_",(0..2,30));
80
81@X=map("$reg_t$_",(3..15,0..2));
82@V=($A,$B,$C,$D,$E,$F,$G,$H)=map("$reg_t$_",(20..27));
83($t0,$t1,$t2,$t3)=map("$reg_t$_",(16,17,19,28));
84
85sub BODY_00_xx {
86my ($i,$a,$b,$c,$d,$e,$f,$g,$h)=@_;
87my $j=($i+1)&15;
88my ($T0,$T1,$T2)=(@X[($i-8)&15],@X[($i-9)&15],@X[($i-10)&15]);
89   $T0=@X[$i+3] if ($i<11);
90
91$code.=<<___	if ($i<16);
92#ifndef	__AARCH64EB__
93	rev	@X[$i],@X[$i]			// $i
94#endif
95___
96$code.=<<___	if ($i<13 && ($i&1));
97	ldp	@X[$i+1],@X[$i+2],[$inp],#2*$SZ
98___
99$code.=<<___	if ($i==13);
100	ldp	@X[14],@X[15],[$inp]
101___
102$code.=<<___	if ($i>=14);
103	ldr	@X[($i-11)&15],[sp,#`$SZ*(($i-11)%4)`]
104___
105$code.=<<___	if ($i>0 && $i<16);
106	add	$a,$a,$t1			// h+=Sigma0(a)
107___
108$code.=<<___	if ($i>=11);
109	str	@X[($i-8)&15],[sp,#`$SZ*(($i-8)%4)`]
110___
111# While ARMv8 specifies merged rotate-n-logical operation such as
112# 'eor x,y,z,ror#n', it was found to negatively affect performance
113# on Apple A7. The reason seems to be that it requires even 'y' to
114# be available earlier. This means that such merged instruction is
115# not necessarily best choice on critical path... On the other hand
116# Cortex-A5x handles merged instructions much better than disjoint
117# rotate and logical... See (**) footnote above.
118$code.=<<___	if ($i<15);
119	ror	$t0,$e,#$Sigma1[0]
120	add	$h,$h,$t2			// h+=K[i]
121	eor	$T0,$e,$e,ror#`$Sigma1[2]-$Sigma1[1]`
122	and	$t1,$f,$e
123	bic	$t2,$g,$e
124	add	$h,$h,@X[$i&15]			// h+=X[i]
125	orr	$t1,$t1,$t2			// Ch(e,f,g)
126	eor	$t2,$a,$b			// a^b, b^c in next round
127	eor	$t0,$t0,$T0,ror#$Sigma1[1]	// Sigma1(e)
128	ror	$T0,$a,#$Sigma0[0]
129	add	$h,$h,$t1			// h+=Ch(e,f,g)
130	eor	$t1,$a,$a,ror#`$Sigma0[2]-$Sigma0[1]`
131	add	$h,$h,$t0			// h+=Sigma1(e)
132	and	$t3,$t3,$t2			// (b^c)&=(a^b)
133	add	$d,$d,$h			// d+=h
134	eor	$t3,$t3,$b			// Maj(a,b,c)
135	eor	$t1,$T0,$t1,ror#$Sigma0[1]	// Sigma0(a)
136	add	$h,$h,$t3			// h+=Maj(a,b,c)
137	ldr	$t3,[$Ktbl],#$SZ		// *K++, $t2 in next round
138	//add	$h,$h,$t1			// h+=Sigma0(a)
139___
140$code.=<<___	if ($i>=15);
141	ror	$t0,$e,#$Sigma1[0]
142	add	$h,$h,$t2			// h+=K[i]
143	ror	$T1,@X[($j+1)&15],#$sigma0[0]
144	and	$t1,$f,$e
145	ror	$T2,@X[($j+14)&15],#$sigma1[0]
146	bic	$t2,$g,$e
147	ror	$T0,$a,#$Sigma0[0]
148	add	$h,$h,@X[$i&15]			// h+=X[i]
149	eor	$t0,$t0,$e,ror#$Sigma1[1]
150	eor	$T1,$T1,@X[($j+1)&15],ror#$sigma0[1]
151	orr	$t1,$t1,$t2			// Ch(e,f,g)
152	eor	$t2,$a,$b			// a^b, b^c in next round
153	eor	$t0,$t0,$e,ror#$Sigma1[2]	// Sigma1(e)
154	eor	$T0,$T0,$a,ror#$Sigma0[1]
155	add	$h,$h,$t1			// h+=Ch(e,f,g)
156	and	$t3,$t3,$t2			// (b^c)&=(a^b)
157	eor	$T2,$T2,@X[($j+14)&15],ror#$sigma1[1]
158	eor	$T1,$T1,@X[($j+1)&15],lsr#$sigma0[2]	// sigma0(X[i+1])
159	add	$h,$h,$t0			// h+=Sigma1(e)
160	eor	$t3,$t3,$b			// Maj(a,b,c)
161	eor	$t1,$T0,$a,ror#$Sigma0[2]	// Sigma0(a)
162	eor	$T2,$T2,@X[($j+14)&15],lsr#$sigma1[2]	// sigma1(X[i+14])
163	add	@X[$j],@X[$j],@X[($j+9)&15]
164	add	$d,$d,$h			// d+=h
165	add	$h,$h,$t3			// h+=Maj(a,b,c)
166	ldr	$t3,[$Ktbl],#$SZ		// *K++, $t2 in next round
167	add	@X[$j],@X[$j],$T1
168	add	$h,$h,$t1			// h+=Sigma0(a)
169	add	@X[$j],@X[$j],$T2
170___
171	($t2,$t3)=($t3,$t2);
172}
173
174$code.=<<___;
175#ifndef	__KERNEL__
176# include <ring-core/arm_arch.h>
177#endif
178
179.text
180
181.extern	OPENSSL_armcap_P
182.hidden	OPENSSL_armcap_P
183.globl	$func
184.type	$func,%function
185.align	6
186$func:
187	AARCH64_VALID_CALL_TARGET
188#ifndef	__KERNEL__
189#if defined(OPENSSL_HWASAN) && __clang_major__ >= 10
190	adrp	x16,:pg_hi21_nc:OPENSSL_armcap_P
191#else
192	adrp	x16,:pg_hi21:OPENSSL_armcap_P
193#endif
194	ldr	w16,[x16,:lo12:OPENSSL_armcap_P]
195___
196$code.=<<___	if ($SZ==4);
197	tst	w16,#ARMV8_SHA256
198	b.ne	.Lv8_entry
199___
200$code.=<<___	if ($SZ==8);
201	tst	w16,#ARMV8_SHA512
202	b.ne	.Lv8_entry
203___
204$code.=<<___;
205#endif
206	AARCH64_SIGN_LINK_REGISTER
207	stp	x29,x30,[sp,#-128]!
208	add	x29,sp,#0
209
210	stp	x19,x20,[sp,#16]
211	stp	x21,x22,[sp,#32]
212	stp	x23,x24,[sp,#48]
213	stp	x25,x26,[sp,#64]
214	stp	x27,x28,[sp,#80]
215	sub	sp,sp,#4*$SZ
216
217	ldp	$A,$B,[$ctx]				// load context
218	ldp	$C,$D,[$ctx,#2*$SZ]
219	ldp	$E,$F,[$ctx,#4*$SZ]
220	add	$num,$inp,$num,lsl#`log(16*$SZ)/log(2)`	// end of input
221	ldp	$G,$H,[$ctx,#6*$SZ]
222	adrp	$Ktbl,:pg_hi21:.LK$BITS
223	add	$Ktbl,$Ktbl,:lo12:.LK$BITS
224	stp	$ctx,$num,[x29,#96]
225
226.Loop:
227	ldp	@X[0],@X[1],[$inp],#2*$SZ
228	ldr	$t2,[$Ktbl],#$SZ			// *K++
229	eor	$t3,$B,$C				// magic seed
230	str	$inp,[x29,#112]
231___
232for ($i=0;$i<16;$i++)	{ &BODY_00_xx($i,@V); unshift(@V,pop(@V)); }
233$code.=".Loop_16_xx:\n";
234for (;$i<32;$i++)	{ &BODY_00_xx($i,@V); unshift(@V,pop(@V)); }
235$code.=<<___;
236	cbnz	$t2,.Loop_16_xx
237
238	ldp	$ctx,$num,[x29,#96]
239	ldr	$inp,[x29,#112]
240	sub	$Ktbl,$Ktbl,#`$SZ*($rounds+1)`		// rewind
241
242	ldp	@X[0],@X[1],[$ctx]
243	ldp	@X[2],@X[3],[$ctx,#2*$SZ]
244	add	$inp,$inp,#14*$SZ			// advance input pointer
245	ldp	@X[4],@X[5],[$ctx,#4*$SZ]
246	add	$A,$A,@X[0]
247	ldp	@X[6],@X[7],[$ctx,#6*$SZ]
248	add	$B,$B,@X[1]
249	add	$C,$C,@X[2]
250	add	$D,$D,@X[3]
251	stp	$A,$B,[$ctx]
252	add	$E,$E,@X[4]
253	add	$F,$F,@X[5]
254	stp	$C,$D,[$ctx,#2*$SZ]
255	add	$G,$G,@X[6]
256	add	$H,$H,@X[7]
257	cmp	$inp,$num
258	stp	$E,$F,[$ctx,#4*$SZ]
259	stp	$G,$H,[$ctx,#6*$SZ]
260	b.ne	.Loop
261
262	ldp	x19,x20,[x29,#16]
263	add	sp,sp,#4*$SZ
264	ldp	x21,x22,[x29,#32]
265	ldp	x23,x24,[x29,#48]
266	ldp	x25,x26,[x29,#64]
267	ldp	x27,x28,[x29,#80]
268	ldp	x29,x30,[sp],#128
269	AARCH64_VALIDATE_LINK_REGISTER
270	ret
271.size	$func,.-$func
272
273.section .rodata
274.align	6
275.type	.LK$BITS,%object
276.LK$BITS:
277___
278$code.=<<___ if ($SZ==8);
279	.quad	0x428a2f98d728ae22,0x7137449123ef65cd
280	.quad	0xb5c0fbcfec4d3b2f,0xe9b5dba58189dbbc
281	.quad	0x3956c25bf348b538,0x59f111f1b605d019
282	.quad	0x923f82a4af194f9b,0xab1c5ed5da6d8118
283	.quad	0xd807aa98a3030242,0x12835b0145706fbe
284	.quad	0x243185be4ee4b28c,0x550c7dc3d5ffb4e2
285	.quad	0x72be5d74f27b896f,0x80deb1fe3b1696b1
286	.quad	0x9bdc06a725c71235,0xc19bf174cf692694
287	.quad	0xe49b69c19ef14ad2,0xefbe4786384f25e3
288	.quad	0x0fc19dc68b8cd5b5,0x240ca1cc77ac9c65
289	.quad	0x2de92c6f592b0275,0x4a7484aa6ea6e483
290	.quad	0x5cb0a9dcbd41fbd4,0x76f988da831153b5
291	.quad	0x983e5152ee66dfab,0xa831c66d2db43210
292	.quad	0xb00327c898fb213f,0xbf597fc7beef0ee4
293	.quad	0xc6e00bf33da88fc2,0xd5a79147930aa725
294	.quad	0x06ca6351e003826f,0x142929670a0e6e70
295	.quad	0x27b70a8546d22ffc,0x2e1b21385c26c926
296	.quad	0x4d2c6dfc5ac42aed,0x53380d139d95b3df
297	.quad	0x650a73548baf63de,0x766a0abb3c77b2a8
298	.quad	0x81c2c92e47edaee6,0x92722c851482353b
299	.quad	0xa2bfe8a14cf10364,0xa81a664bbc423001
300	.quad	0xc24b8b70d0f89791,0xc76c51a30654be30
301	.quad	0xd192e819d6ef5218,0xd69906245565a910
302	.quad	0xf40e35855771202a,0x106aa07032bbd1b8
303	.quad	0x19a4c116b8d2d0c8,0x1e376c085141ab53
304	.quad	0x2748774cdf8eeb99,0x34b0bcb5e19b48a8
305	.quad	0x391c0cb3c5c95a63,0x4ed8aa4ae3418acb
306	.quad	0x5b9cca4f7763e373,0x682e6ff3d6b2b8a3
307	.quad	0x748f82ee5defb2fc,0x78a5636f43172f60
308	.quad	0x84c87814a1f0ab72,0x8cc702081a6439ec
309	.quad	0x90befffa23631e28,0xa4506cebde82bde9
310	.quad	0xbef9a3f7b2c67915,0xc67178f2e372532b
311	.quad	0xca273eceea26619c,0xd186b8c721c0c207
312	.quad	0xeada7dd6cde0eb1e,0xf57d4f7fee6ed178
313	.quad	0x06f067aa72176fba,0x0a637dc5a2c898a6
314	.quad	0x113f9804bef90dae,0x1b710b35131c471b
315	.quad	0x28db77f523047d84,0x32caab7b40c72493
316	.quad	0x3c9ebe0a15c9bebc,0x431d67c49c100d4c
317	.quad	0x4cc5d4becb3e42b6,0x597f299cfc657e2a
318	.quad	0x5fcb6fab3ad6faec,0x6c44198c4a475817
319	.quad	0	// terminator
320___
321$code.=<<___ if ($SZ==4);
322	.long	0x428a2f98,0x71374491,0xb5c0fbcf,0xe9b5dba5
323	.long	0x3956c25b,0x59f111f1,0x923f82a4,0xab1c5ed5
324	.long	0xd807aa98,0x12835b01,0x243185be,0x550c7dc3
325	.long	0x72be5d74,0x80deb1fe,0x9bdc06a7,0xc19bf174
326	.long	0xe49b69c1,0xefbe4786,0x0fc19dc6,0x240ca1cc
327	.long	0x2de92c6f,0x4a7484aa,0x5cb0a9dc,0x76f988da
328	.long	0x983e5152,0xa831c66d,0xb00327c8,0xbf597fc7
329	.long	0xc6e00bf3,0xd5a79147,0x06ca6351,0x14292967
330	.long	0x27b70a85,0x2e1b2138,0x4d2c6dfc,0x53380d13
331	.long	0x650a7354,0x766a0abb,0x81c2c92e,0x92722c85
332	.long	0xa2bfe8a1,0xa81a664b,0xc24b8b70,0xc76c51a3
333	.long	0xd192e819,0xd6990624,0xf40e3585,0x106aa070
334	.long	0x19a4c116,0x1e376c08,0x2748774c,0x34b0bcb5
335	.long	0x391c0cb3,0x4ed8aa4a,0x5b9cca4f,0x682e6ff3
336	.long	0x748f82ee,0x78a5636f,0x84c87814,0x8cc70208
337	.long	0x90befffa,0xa4506ceb,0xbef9a3f7,0xc67178f2
338	.long	0	//terminator
339___
340$code.=<<___;
341.size	.LK$BITS,.-.LK$BITS
342.asciz	"SHA$BITS block transform for ARMv8, CRYPTOGAMS by <appro\@openssl.org>"
343.align	2
344___
345
346if ($SZ==4) {
347my $Ktbl="x3";
348
349my ($ABCD,$EFGH,$abcd)=map("v$_.16b",(0..2));
350my @MSG=map("v$_.16b",(4..7));
351my ($W0,$W1)=("v16.4s","v17.4s");
352my ($ABCD_SAVE,$EFGH_SAVE)=("v18.16b","v19.16b");
353
354$code.=<<___;
355.text
356#ifndef	__KERNEL__
357.type	sha256_block_armv8,%function
358.align	6
359sha256_block_armv8:
360.Lv8_entry:
361	// Armv8.3-A PAuth: even though x30 is pushed to stack it is not popped later.
362	stp		x29,x30,[sp,#-16]!
363	add		x29,sp,#0
364
365	ld1.32		{$ABCD,$EFGH},[$ctx]
366	adrp		$Ktbl,:pg_hi21:.LK256
367	add		$Ktbl,$Ktbl,:lo12:.LK256
368
369.Loop_hw:
370	ld1		{@MSG[0]-@MSG[3]},[$inp],#64
371	sub		$num,$num,#1
372	ld1.32		{$W0},[$Ktbl],#16
373	rev32		@MSG[0],@MSG[0]
374	rev32		@MSG[1],@MSG[1]
375	rev32		@MSG[2],@MSG[2]
376	rev32		@MSG[3],@MSG[3]
377	orr		$ABCD_SAVE,$ABCD,$ABCD		// offload
378	orr		$EFGH_SAVE,$EFGH,$EFGH
379___
380for($i=0;$i<12;$i++) {
381$code.=<<___;
382	ld1.32		{$W1},[$Ktbl],#16
383	add.i32		$W0,$W0,@MSG[0]
384	sha256su0	@MSG[0],@MSG[1]
385	orr		$abcd,$ABCD,$ABCD
386	sha256h		$ABCD,$EFGH,$W0
387	sha256h2	$EFGH,$abcd,$W0
388	sha256su1	@MSG[0],@MSG[2],@MSG[3]
389___
390	($W0,$W1)=($W1,$W0);	push(@MSG,shift(@MSG));
391}
392$code.=<<___;
393	ld1.32		{$W1},[$Ktbl],#16
394	add.i32		$W0,$W0,@MSG[0]
395	orr		$abcd,$ABCD,$ABCD
396	sha256h		$ABCD,$EFGH,$W0
397	sha256h2	$EFGH,$abcd,$W0
398
399	ld1.32		{$W0},[$Ktbl],#16
400	add.i32		$W1,$W1,@MSG[1]
401	orr		$abcd,$ABCD,$ABCD
402	sha256h		$ABCD,$EFGH,$W1
403	sha256h2	$EFGH,$abcd,$W1
404
405	ld1.32		{$W1},[$Ktbl]
406	add.i32		$W0,$W0,@MSG[2]
407	sub		$Ktbl,$Ktbl,#$rounds*$SZ-16	// rewind
408	orr		$abcd,$ABCD,$ABCD
409	sha256h		$ABCD,$EFGH,$W0
410	sha256h2	$EFGH,$abcd,$W0
411
412	add.i32		$W1,$W1,@MSG[3]
413	orr		$abcd,$ABCD,$ABCD
414	sha256h		$ABCD,$EFGH,$W1
415	sha256h2	$EFGH,$abcd,$W1
416
417	add.i32		$ABCD,$ABCD,$ABCD_SAVE
418	add.i32		$EFGH,$EFGH,$EFGH_SAVE
419
420	cbnz		$num,.Loop_hw
421
422	st1.32		{$ABCD,$EFGH},[$ctx]
423
424	ldr		x29,[sp],#16
425	ret
426.size	sha256_block_armv8,.-sha256_block_armv8
427#endif
428___
429}
430
431if ($SZ==8) {
432my $Ktbl="x3";
433
434my @H = map("v$_.16b",(0..4));
435my ($fg,$de,$m9_10)=map("v$_.16b",(5..7));
436my @MSG=map("v$_.16b",(16..23));
437my ($W0,$W1)=("v24.2d","v25.2d");
438my ($AB,$CD,$EF,$GH)=map("v$_.16b",(26..29));
439
440$code.=<<___;
441.text
442#ifndef	__KERNEL__
443.type	sha512_block_armv8,%function
444.align	6
445sha512_block_armv8:
446.Lv8_entry:
447	stp		x29,x30,[sp,#-16]!
448	add		x29,sp,#0
449
450	ld1		{@MSG[0]-@MSG[3]},[$inp],#64	// load input
451	ld1		{@MSG[4]-@MSG[7]},[$inp],#64
452
453	ld1.64		{@H[0]-@H[3]},[$ctx]		// load context
454	adrp		$Ktbl,:pg_hi21:.LK512
455	add		$Ktbl,$Ktbl,:lo12:.LK512
456
457	rev64		@MSG[0],@MSG[0]
458	rev64		@MSG[1],@MSG[1]
459	rev64		@MSG[2],@MSG[2]
460	rev64		@MSG[3],@MSG[3]
461	rev64		@MSG[4],@MSG[4]
462	rev64		@MSG[5],@MSG[5]
463	rev64		@MSG[6],@MSG[6]
464	rev64		@MSG[7],@MSG[7]
465	b		.Loop_hw
466
467.align	4
468.Loop_hw:
469	ld1.64		{$W0},[$Ktbl],#16
470	subs		$num,$num,#1
471	sub		x4,$inp,#128
472	orr		$AB,@H[0],@H[0]			// offload
473	orr		$CD,@H[1],@H[1]
474	orr		$EF,@H[2],@H[2]
475	orr		$GH,@H[3],@H[3]
476	csel		$inp,$inp,x4,ne			// conditional rewind
477___
478for($i=0;$i<32;$i++) {
479$code.=<<___;
480	add.i64		$W0,$W0,@MSG[0]
481	ld1.64		{$W1},[$Ktbl],#16
482	ext		$W0,$W0,$W0,#8
483	ext		$fg,@H[2],@H[3],#8
484	ext		$de,@H[1],@H[2],#8
485	add.i64		@H[3],@H[3],$W0			// "T1 + H + K512[i]"
486	 sha512su0	@MSG[0],@MSG[1]
487	 ext		$m9_10,@MSG[4],@MSG[5],#8
488	sha512h		@H[3],$fg,$de
489	 sha512su1	@MSG[0],@MSG[7],$m9_10
490	add.i64		@H[4],@H[1],@H[3]		// "D + T1"
491	sha512h2	@H[3],$H[1],@H[0]
492___
493	($W0,$W1)=($W1,$W0);	push(@MSG,shift(@MSG));
494	@H = (@H[3],@H[0],@H[4],@H[2],@H[1]);
495}
496for(;$i<40;$i++) {
497$code.=<<___	if ($i<39);
498	ld1.64		{$W1},[$Ktbl],#16
499___
500$code.=<<___	if ($i==39);
501	sub		$Ktbl,$Ktbl,#$rounds*$SZ	// rewind
502___
503$code.=<<___;
504	add.i64		$W0,$W0,@MSG[0]
505	 ld1		{@MSG[0]},[$inp],#16		// load next input
506	ext		$W0,$W0,$W0,#8
507	ext		$fg,@H[2],@H[3],#8
508	ext		$de,@H[1],@H[2],#8
509	add.i64		@H[3],@H[3],$W0			// "T1 + H + K512[i]"
510	sha512h		@H[3],$fg,$de
511	 rev64		@MSG[0],@MSG[0]
512	add.i64		@H[4],@H[1],@H[3]		// "D + T1"
513	sha512h2	@H[3],$H[1],@H[0]
514___
515	($W0,$W1)=($W1,$W0);	push(@MSG,shift(@MSG));
516	@H = (@H[3],@H[0],@H[4],@H[2],@H[1]);
517}
518$code.=<<___;
519	add.i64		@H[0],@H[0],$AB			// accumulate
520	add.i64		@H[1],@H[1],$CD
521	add.i64		@H[2],@H[2],$EF
522	add.i64		@H[3],@H[3],$GH
523
524	cbnz		$num,.Loop_hw
525
526	st1.64		{@H[0]-@H[3]},[$ctx]		// store context
527
528	ldr		x29,[sp],#16
529	ret
530.size	sha512_block_armv8,.-sha512_block_armv8
531#endif
532___
533}
534
535{   my  %opcode = (
536	"sha256h"	=> 0x5e004000,	"sha256h2"	=> 0x5e005000,
537	"sha256su0"	=> 0x5e282800,	"sha256su1"	=> 0x5e006000	);
538
539    sub unsha256 {
540	my ($mnemonic,$arg)=@_;
541
542	$arg =~ m/[qv]([0-9]+)[^,]*,\s*[qv]([0-9]+)[^,]*(?:,\s*[qv]([0-9]+))?/o
543	&&
544	sprintf ".inst\t0x%08x\t//%s %s",
545			$opcode{$mnemonic}|$1|($2<<5)|($3<<16),
546			$mnemonic,$arg;
547    }
548}
549
550{   my  %opcode = (
551	"sha512h"	=> 0xce608000,	"sha512h2"	=> 0xce608400,
552	"sha512su0"	=> 0xcec08000,	"sha512su1"	=> 0xce608800	);
553
554    sub unsha512 {
555	my ($mnemonic,$arg)=@_;
556
557	$arg =~ m/[qv]([0-9]+)[^,]*,\s*[qv]([0-9]+)[^,]*(?:,\s*[qv]([0-9]+))?/o
558	&&
559	sprintf ".inst\t0x%08x\t//%s %s",
560			$opcode{$mnemonic}|$1|($2<<5)|($3<<16),
561			$mnemonic,$arg;
562    }
563}
564
565open SELF,$0;
566while(<SELF>) {
567        next if (/^#!/);
568        last if (!s/^#/\/\// and !/^$/);
569        print;
570}
571close SELF;
572
573foreach(split("\n",$code)) {
574
575	s/\`([^\`]*)\`/eval($1)/ge;
576
577	s/\b(sha512\w+)\s+([qv].*)/unsha512($1,$2)/ge	or
578	s/\b(sha256\w+)\s+([qv].*)/unsha256($1,$2)/ge;
579
580	s/\bq([0-9]+)\b/v$1.16b/g;		# old->new registers
581
582	s/\.[ui]?8(\s)/$1/;
583	s/\.\w?64\b//		and s/\.16b/\.2d/g	or
584	s/\.\w?32\b//		and s/\.16b/\.4s/g;
585	m/\bext\b/		and s/\.2d/\.16b/g	or
586	m/(ld|st)1[^\[]+\[0\]/	and s/\.4s/\.s/g;
587
588	print $_,"\n";
589}
590
591close STDOUT or die "error closing STDOUT: $!";
592