xref: /freebsd/crypto/openssl/crypto/sha/asm/sha512-armv4.pl (revision b077aed33b7b6aefca7b17ddb250cf521f938613)
1#! /usr/bin/env perl
2# Copyright 2007-2020 The OpenSSL Project Authors. All Rights Reserved.
3#
4# Licensed under the Apache License 2.0 (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# Permission to use under GPL terms is granted.
17# ====================================================================
18
19# SHA512 block procedure for ARMv4. September 2007.
20
21# This code is ~4.5 (four and a half) times faster than code generated
22# by gcc 3.4 and it spends ~72 clock cycles per byte [on single-issue
23# Xscale PXA250 core].
24#
25# July 2010.
26#
27# Rescheduling for dual-issue pipeline resulted in 6% improvement on
28# Cortex A8 core and ~40 cycles per processed byte.
29
30# February 2011.
31#
32# Profiler-assisted and platform-specific optimization resulted in 7%
33# improvement on Coxtex A8 core and ~38 cycles per byte.
34
35# March 2011.
36#
37# Add NEON implementation. On Cortex A8 it was measured to process
38# one byte in 23.3 cycles or ~60% faster than integer-only code.
39
40# August 2012.
41#
42# Improve NEON performance by 12% on Snapdragon S4. In absolute
43# terms it's 22.6 cycles per byte, which is disappointing result.
44# Technical writers asserted that 3-way S4 pipeline can sustain
45# multiple NEON instructions per cycle, but dual NEON issue could
46# not be observed, see http://www.openssl.org/~appro/Snapdragon-S4.html
47# for further details. On side note Cortex-A15 processes one byte in
48# 16 cycles.
49
50# Byte order [in]dependence. =========================================
51#
52# Originally caller was expected to maintain specific *dword* order in
53# h[0-7], namely with most significant dword at *lower* address, which
54# was reflected in below two parameters as 0 and 4. Now caller is
55# expected to maintain native byte order for whole 64-bit values.
56$hi="HI";
57$lo="LO";
58# ====================================================================
59
60# $output is the last argument if it looks like a file (it has an extension)
61# $flavour is the first argument if it doesn't look like a file
62$output = $#ARGV >= 0 && $ARGV[$#ARGV] =~ m|\.\w+$| ? pop : undef;
63$flavour = $#ARGV >= 0 && $ARGV[0] !~ m|\.| ? shift : undef;
64
65if ($flavour && $flavour ne "void") {
66    $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
67    ( $xlate="${dir}arm-xlate.pl" and -f $xlate ) or
68    ( $xlate="${dir}../../perlasm/arm-xlate.pl" and -f $xlate) or
69    die "can't locate arm-xlate.pl";
70
71    open STDOUT,"| \"$^X\" $xlate $flavour \"$output\""
72        or die "can't call $xlate: $!";
73} else {
74    $output and open STDOUT,">$output";
75}
76
77$ctx="r0";	# parameter block
78$inp="r1";
79$len="r2";
80
81$Tlo="r3";
82$Thi="r4";
83$Alo="r5";
84$Ahi="r6";
85$Elo="r7";
86$Ehi="r8";
87$t0="r9";
88$t1="r10";
89$t2="r11";
90$t3="r12";
91############	r13 is stack pointer
92$Ktbl="r14";
93############	r15 is program counter
94
95$Aoff=8*0;
96$Boff=8*1;
97$Coff=8*2;
98$Doff=8*3;
99$Eoff=8*4;
100$Foff=8*5;
101$Goff=8*6;
102$Hoff=8*7;
103$Xoff=8*8;
104
105sub BODY_00_15() {
106my $magic = shift;
107$code.=<<___;
108	@ Sigma1(x)	(ROTR((x),14) ^ ROTR((x),18)  ^ ROTR((x),41))
109	@ LO		lo>>14^hi<<18 ^ lo>>18^hi<<14 ^ hi>>9^lo<<23
110	@ HI		hi>>14^lo<<18 ^ hi>>18^lo<<14 ^ lo>>9^hi<<23
111	mov	$t0,$Elo,lsr#14
112	str	$Tlo,[sp,#$Xoff+0]
113	mov	$t1,$Ehi,lsr#14
114	str	$Thi,[sp,#$Xoff+4]
115	eor	$t0,$t0,$Ehi,lsl#18
116	ldr	$t2,[sp,#$Hoff+0]	@ h.lo
117	eor	$t1,$t1,$Elo,lsl#18
118	ldr	$t3,[sp,#$Hoff+4]	@ h.hi
119	eor	$t0,$t0,$Elo,lsr#18
120	eor	$t1,$t1,$Ehi,lsr#18
121	eor	$t0,$t0,$Ehi,lsl#14
122	eor	$t1,$t1,$Elo,lsl#14
123	eor	$t0,$t0,$Ehi,lsr#9
124	eor	$t1,$t1,$Elo,lsr#9
125	eor	$t0,$t0,$Elo,lsl#23
126	eor	$t1,$t1,$Ehi,lsl#23	@ Sigma1(e)
127	adds	$Tlo,$Tlo,$t0
128	ldr	$t0,[sp,#$Foff+0]	@ f.lo
129	adc	$Thi,$Thi,$t1		@ T += Sigma1(e)
130	ldr	$t1,[sp,#$Foff+4]	@ f.hi
131	adds	$Tlo,$Tlo,$t2
132	ldr	$t2,[sp,#$Goff+0]	@ g.lo
133	adc	$Thi,$Thi,$t3		@ T += h
134	ldr	$t3,[sp,#$Goff+4]	@ g.hi
135
136	eor	$t0,$t0,$t2
137	str	$Elo,[sp,#$Eoff+0]
138	eor	$t1,$t1,$t3
139	str	$Ehi,[sp,#$Eoff+4]
140	and	$t0,$t0,$Elo
141	str	$Alo,[sp,#$Aoff+0]
142	and	$t1,$t1,$Ehi
143	str	$Ahi,[sp,#$Aoff+4]
144	eor	$t0,$t0,$t2
145	ldr	$t2,[$Ktbl,#$lo]	@ K[i].lo
146	eor	$t1,$t1,$t3		@ Ch(e,f,g)
147	ldr	$t3,[$Ktbl,#$hi]	@ K[i].hi
148
149	adds	$Tlo,$Tlo,$t0
150	ldr	$Elo,[sp,#$Doff+0]	@ d.lo
151	adc	$Thi,$Thi,$t1		@ T += Ch(e,f,g)
152	ldr	$Ehi,[sp,#$Doff+4]	@ d.hi
153	adds	$Tlo,$Tlo,$t2
154	and	$t0,$t2,#0xff
155	adc	$Thi,$Thi,$t3		@ T += K[i]
156	adds	$Elo,$Elo,$Tlo
157	ldr	$t2,[sp,#$Boff+0]	@ b.lo
158	adc	$Ehi,$Ehi,$Thi		@ d += T
159	teq	$t0,#$magic
160
161	ldr	$t3,[sp,#$Coff+0]	@ c.lo
162#ifdef	__thumb2__
163	it	eq			@ Thumb2 thing, sanity check in ARM
164#endif
165	orreq	$Ktbl,$Ktbl,#1
166	@ Sigma0(x)	(ROTR((x),28) ^ ROTR((x),34) ^ ROTR((x),39))
167	@ LO		lo>>28^hi<<4  ^ hi>>2^lo<<30 ^ hi>>7^lo<<25
168	@ HI		hi>>28^lo<<4  ^ lo>>2^hi<<30 ^ lo>>7^hi<<25
169	mov	$t0,$Alo,lsr#28
170	mov	$t1,$Ahi,lsr#28
171	eor	$t0,$t0,$Ahi,lsl#4
172	eor	$t1,$t1,$Alo,lsl#4
173	eor	$t0,$t0,$Ahi,lsr#2
174	eor	$t1,$t1,$Alo,lsr#2
175	eor	$t0,$t0,$Alo,lsl#30
176	eor	$t1,$t1,$Ahi,lsl#30
177	eor	$t0,$t0,$Ahi,lsr#7
178	eor	$t1,$t1,$Alo,lsr#7
179	eor	$t0,$t0,$Alo,lsl#25
180	eor	$t1,$t1,$Ahi,lsl#25	@ Sigma0(a)
181	adds	$Tlo,$Tlo,$t0
182	and	$t0,$Alo,$t2
183	adc	$Thi,$Thi,$t1		@ T += Sigma0(a)
184
185	ldr	$t1,[sp,#$Boff+4]	@ b.hi
186	orr	$Alo,$Alo,$t2
187	ldr	$t2,[sp,#$Coff+4]	@ c.hi
188	and	$Alo,$Alo,$t3
189	and	$t3,$Ahi,$t1
190	orr	$Ahi,$Ahi,$t1
191	orr	$Alo,$Alo,$t0		@ Maj(a,b,c).lo
192	and	$Ahi,$Ahi,$t2
193	adds	$Alo,$Alo,$Tlo
194	orr	$Ahi,$Ahi,$t3		@ Maj(a,b,c).hi
195	sub	sp,sp,#8
196	adc	$Ahi,$Ahi,$Thi		@ h += T
197	tst	$Ktbl,#1
198	add	$Ktbl,$Ktbl,#8
199___
200}
201
202my $_word = ($flavour =~ /win/ ? "DCDU" : ".word");
203
204$code=<<___;
205#ifndef __KERNEL__
206# include "arm_arch.h"
207# define VFP_ABI_PUSH	vstmdb	sp!,{d8-d15}
208# define VFP_ABI_POP	vldmia	sp!,{d8-d15}
209#else
210# define __ARM_ARCH__ __LINUX_ARM_ARCH__
211# define __ARM_MAX_ARCH__ 7
212# define VFP_ABI_PUSH
213# define VFP_ABI_POP
214#endif
215
216#ifdef __ARMEL__
217# define LO 0
218# define HI 4
219# define WORD64(hi0,lo0,hi1,lo1)	$_word	lo0,hi0, lo1,hi1
220#else
221# define HI 0
222# define LO 4
223# define WORD64(hi0,lo0,hi1,lo1)	$_word	hi0,lo0, hi1,lo1
224#endif
225
226#if defined(__thumb2__)
227.syntax unified
228.thumb
229# define adrl adr
230#else
231.code	32
232#endif
233
234.text
235
236.type	K512,%object
237.align	5
238K512:
239WORD64(0x428a2f98,0xd728ae22, 0x71374491,0x23ef65cd)
240WORD64(0xb5c0fbcf,0xec4d3b2f, 0xe9b5dba5,0x8189dbbc)
241WORD64(0x3956c25b,0xf348b538, 0x59f111f1,0xb605d019)
242WORD64(0x923f82a4,0xaf194f9b, 0xab1c5ed5,0xda6d8118)
243WORD64(0xd807aa98,0xa3030242, 0x12835b01,0x45706fbe)
244WORD64(0x243185be,0x4ee4b28c, 0x550c7dc3,0xd5ffb4e2)
245WORD64(0x72be5d74,0xf27b896f, 0x80deb1fe,0x3b1696b1)
246WORD64(0x9bdc06a7,0x25c71235, 0xc19bf174,0xcf692694)
247WORD64(0xe49b69c1,0x9ef14ad2, 0xefbe4786,0x384f25e3)
248WORD64(0x0fc19dc6,0x8b8cd5b5, 0x240ca1cc,0x77ac9c65)
249WORD64(0x2de92c6f,0x592b0275, 0x4a7484aa,0x6ea6e483)
250WORD64(0x5cb0a9dc,0xbd41fbd4, 0x76f988da,0x831153b5)
251WORD64(0x983e5152,0xee66dfab, 0xa831c66d,0x2db43210)
252WORD64(0xb00327c8,0x98fb213f, 0xbf597fc7,0xbeef0ee4)
253WORD64(0xc6e00bf3,0x3da88fc2, 0xd5a79147,0x930aa725)
254WORD64(0x06ca6351,0xe003826f, 0x14292967,0x0a0e6e70)
255WORD64(0x27b70a85,0x46d22ffc, 0x2e1b2138,0x5c26c926)
256WORD64(0x4d2c6dfc,0x5ac42aed, 0x53380d13,0x9d95b3df)
257WORD64(0x650a7354,0x8baf63de, 0x766a0abb,0x3c77b2a8)
258WORD64(0x81c2c92e,0x47edaee6, 0x92722c85,0x1482353b)
259WORD64(0xa2bfe8a1,0x4cf10364, 0xa81a664b,0xbc423001)
260WORD64(0xc24b8b70,0xd0f89791, 0xc76c51a3,0x0654be30)
261WORD64(0xd192e819,0xd6ef5218, 0xd6990624,0x5565a910)
262WORD64(0xf40e3585,0x5771202a, 0x106aa070,0x32bbd1b8)
263WORD64(0x19a4c116,0xb8d2d0c8, 0x1e376c08,0x5141ab53)
264WORD64(0x2748774c,0xdf8eeb99, 0x34b0bcb5,0xe19b48a8)
265WORD64(0x391c0cb3,0xc5c95a63, 0x4ed8aa4a,0xe3418acb)
266WORD64(0x5b9cca4f,0x7763e373, 0x682e6ff3,0xd6b2b8a3)
267WORD64(0x748f82ee,0x5defb2fc, 0x78a5636f,0x43172f60)
268WORD64(0x84c87814,0xa1f0ab72, 0x8cc70208,0x1a6439ec)
269WORD64(0x90befffa,0x23631e28, 0xa4506ceb,0xde82bde9)
270WORD64(0xbef9a3f7,0xb2c67915, 0xc67178f2,0xe372532b)
271WORD64(0xca273ece,0xea26619c, 0xd186b8c7,0x21c0c207)
272WORD64(0xeada7dd6,0xcde0eb1e, 0xf57d4f7f,0xee6ed178)
273WORD64(0x06f067aa,0x72176fba, 0x0a637dc5,0xa2c898a6)
274WORD64(0x113f9804,0xbef90dae, 0x1b710b35,0x131c471b)
275WORD64(0x28db77f5,0x23047d84, 0x32caab7b,0x40c72493)
276WORD64(0x3c9ebe0a,0x15c9bebc, 0x431d67c4,0x9c100d4c)
277WORD64(0x4cc5d4be,0xcb3e42b6, 0x597f299c,0xfc657e2a)
278WORD64(0x5fcb6fab,0x3ad6faec, 0x6c44198c,0x4a475817)
279.size	K512,.-K512
280#if __ARM_MAX_ARCH__>=7 && !defined(__KERNEL__)
281.LOPENSSL_armcap:
282# ifdef	_WIN32
283.word	OPENSSL_armcap_P
284# else
285.word	OPENSSL_armcap_P-.Lsha512_block_data_order
286# endif
287.skip	32-4
288#else
289.skip	32
290#endif
291
292.global	sha512_block_data_order
293.type	sha512_block_data_order,%function
294sha512_block_data_order:
295.Lsha512_block_data_order:
296#if __ARM_ARCH__<7 && !defined(__thumb2__)
297	sub	r3,pc,#8		@ sha512_block_data_order
298#else
299	adr	r3,.Lsha512_block_data_order
300#endif
301#if __ARM_MAX_ARCH__>=7 && !defined(__KERNEL__)
302	ldr	r12,.LOPENSSL_armcap
303# if !defined(_WIN32)
304	ldr	r12,[r3,r12]		@ OPENSSL_armcap_P
305# endif
306# if defined(__APPLE__) || defined(_WIN32)
307	ldr	r12,[r12]
308# endif
309	tst	r12,#ARMV7_NEON
310	bne	.LNEON
311#endif
312	add	$len,$inp,$len,lsl#7	@ len to point at the end of inp
313	stmdb	sp!,{r4-r12,lr}
314	sub	$Ktbl,r3,#672		@ K512
315	sub	sp,sp,#9*8
316
317	ldr	$Elo,[$ctx,#$Eoff+$lo]
318	ldr	$Ehi,[$ctx,#$Eoff+$hi]
319	ldr	$t0, [$ctx,#$Goff+$lo]
320	ldr	$t1, [$ctx,#$Goff+$hi]
321	ldr	$t2, [$ctx,#$Hoff+$lo]
322	ldr	$t3, [$ctx,#$Hoff+$hi]
323.Loop:
324	str	$t0, [sp,#$Goff+0]
325	str	$t1, [sp,#$Goff+4]
326	str	$t2, [sp,#$Hoff+0]
327	str	$t3, [sp,#$Hoff+4]
328	ldr	$Alo,[$ctx,#$Aoff+$lo]
329	ldr	$Ahi,[$ctx,#$Aoff+$hi]
330	ldr	$Tlo,[$ctx,#$Boff+$lo]
331	ldr	$Thi,[$ctx,#$Boff+$hi]
332	ldr	$t0, [$ctx,#$Coff+$lo]
333	ldr	$t1, [$ctx,#$Coff+$hi]
334	ldr	$t2, [$ctx,#$Doff+$lo]
335	ldr	$t3, [$ctx,#$Doff+$hi]
336	str	$Tlo,[sp,#$Boff+0]
337	str	$Thi,[sp,#$Boff+4]
338	str	$t0, [sp,#$Coff+0]
339	str	$t1, [sp,#$Coff+4]
340	str	$t2, [sp,#$Doff+0]
341	str	$t3, [sp,#$Doff+4]
342	ldr	$Tlo,[$ctx,#$Foff+$lo]
343	ldr	$Thi,[$ctx,#$Foff+$hi]
344	str	$Tlo,[sp,#$Foff+0]
345	str	$Thi,[sp,#$Foff+4]
346
347.L00_15:
348#if __ARM_ARCH__<7
349	ldrb	$Tlo,[$inp,#7]
350	ldrb	$t0, [$inp,#6]
351	ldrb	$t1, [$inp,#5]
352	ldrb	$t2, [$inp,#4]
353	ldrb	$Thi,[$inp,#3]
354	ldrb	$t3, [$inp,#2]
355	orr	$Tlo,$Tlo,$t0,lsl#8
356	ldrb	$t0, [$inp,#1]
357	orr	$Tlo,$Tlo,$t1,lsl#16
358	ldrb	$t1, [$inp],#8
359	orr	$Tlo,$Tlo,$t2,lsl#24
360	orr	$Thi,$Thi,$t3,lsl#8
361	orr	$Thi,$Thi,$t0,lsl#16
362	orr	$Thi,$Thi,$t1,lsl#24
363#else
364	ldr	$Tlo,[$inp,#4]
365	ldr	$Thi,[$inp],#8
366#ifdef __ARMEL__
367	rev	$Tlo,$Tlo
368	rev	$Thi,$Thi
369#endif
370#endif
371___
372	&BODY_00_15(0x94);
373$code.=<<___;
374	tst	$Ktbl,#1
375	beq	.L00_15
376	ldr	$t0,[sp,#`$Xoff+8*(16-1)`+0]
377	ldr	$t1,[sp,#`$Xoff+8*(16-1)`+4]
378	bic	$Ktbl,$Ktbl,#1
379.L16_79:
380	@ sigma0(x)	(ROTR((x),1)  ^ ROTR((x),8)  ^ ((x)>>7))
381	@ LO		lo>>1^hi<<31  ^ lo>>8^hi<<24 ^ lo>>7^hi<<25
382	@ HI		hi>>1^lo<<31  ^ hi>>8^lo<<24 ^ hi>>7
383	mov	$Tlo,$t0,lsr#1
384	ldr	$t2,[sp,#`$Xoff+8*(16-14)`+0]
385	mov	$Thi,$t1,lsr#1
386	ldr	$t3,[sp,#`$Xoff+8*(16-14)`+4]
387	eor	$Tlo,$Tlo,$t1,lsl#31
388	eor	$Thi,$Thi,$t0,lsl#31
389	eor	$Tlo,$Tlo,$t0,lsr#8
390	eor	$Thi,$Thi,$t1,lsr#8
391	eor	$Tlo,$Tlo,$t1,lsl#24
392	eor	$Thi,$Thi,$t0,lsl#24
393	eor	$Tlo,$Tlo,$t0,lsr#7
394	eor	$Thi,$Thi,$t1,lsr#7
395	eor	$Tlo,$Tlo,$t1,lsl#25
396
397	@ sigma1(x)	(ROTR((x),19) ^ ROTR((x),61) ^ ((x)>>6))
398	@ LO		lo>>19^hi<<13 ^ hi>>29^lo<<3 ^ lo>>6^hi<<26
399	@ HI		hi>>19^lo<<13 ^ lo>>29^hi<<3 ^ hi>>6
400	mov	$t0,$t2,lsr#19
401	mov	$t1,$t3,lsr#19
402	eor	$t0,$t0,$t3,lsl#13
403	eor	$t1,$t1,$t2,lsl#13
404	eor	$t0,$t0,$t3,lsr#29
405	eor	$t1,$t1,$t2,lsr#29
406	eor	$t0,$t0,$t2,lsl#3
407	eor	$t1,$t1,$t3,lsl#3
408	eor	$t0,$t0,$t2,lsr#6
409	eor	$t1,$t1,$t3,lsr#6
410	ldr	$t2,[sp,#`$Xoff+8*(16-9)`+0]
411	eor	$t0,$t0,$t3,lsl#26
412
413	ldr	$t3,[sp,#`$Xoff+8*(16-9)`+4]
414	adds	$Tlo,$Tlo,$t0
415	ldr	$t0,[sp,#`$Xoff+8*16`+0]
416	adc	$Thi,$Thi,$t1
417
418	ldr	$t1,[sp,#`$Xoff+8*16`+4]
419	adds	$Tlo,$Tlo,$t2
420	adc	$Thi,$Thi,$t3
421	adds	$Tlo,$Tlo,$t0
422	adc	$Thi,$Thi,$t1
423___
424	&BODY_00_15(0x17);
425$code.=<<___;
426#ifdef	__thumb2__
427	ittt	eq			@ Thumb2 thing, sanity check in ARM
428#endif
429	ldreq	$t0,[sp,#`$Xoff+8*(16-1)`+0]
430	ldreq	$t1,[sp,#`$Xoff+8*(16-1)`+4]
431	beq	.L16_79
432	bic	$Ktbl,$Ktbl,#1
433
434	ldr	$Tlo,[sp,#$Boff+0]
435	ldr	$Thi,[sp,#$Boff+4]
436	ldr	$t0, [$ctx,#$Aoff+$lo]
437	ldr	$t1, [$ctx,#$Aoff+$hi]
438	ldr	$t2, [$ctx,#$Boff+$lo]
439	ldr	$t3, [$ctx,#$Boff+$hi]
440	adds	$t0,$Alo,$t0
441	str	$t0, [$ctx,#$Aoff+$lo]
442	adc	$t1,$Ahi,$t1
443	str	$t1, [$ctx,#$Aoff+$hi]
444	adds	$t2,$Tlo,$t2
445	str	$t2, [$ctx,#$Boff+$lo]
446	adc	$t3,$Thi,$t3
447	str	$t3, [$ctx,#$Boff+$hi]
448
449	ldr	$Alo,[sp,#$Coff+0]
450	ldr	$Ahi,[sp,#$Coff+4]
451	ldr	$Tlo,[sp,#$Doff+0]
452	ldr	$Thi,[sp,#$Doff+4]
453	ldr	$t0, [$ctx,#$Coff+$lo]
454	ldr	$t1, [$ctx,#$Coff+$hi]
455	ldr	$t2, [$ctx,#$Doff+$lo]
456	ldr	$t3, [$ctx,#$Doff+$hi]
457	adds	$t0,$Alo,$t0
458	str	$t0, [$ctx,#$Coff+$lo]
459	adc	$t1,$Ahi,$t1
460	str	$t1, [$ctx,#$Coff+$hi]
461	adds	$t2,$Tlo,$t2
462	str	$t2, [$ctx,#$Doff+$lo]
463	adc	$t3,$Thi,$t3
464	str	$t3, [$ctx,#$Doff+$hi]
465
466	ldr	$Tlo,[sp,#$Foff+0]
467	ldr	$Thi,[sp,#$Foff+4]
468	ldr	$t0, [$ctx,#$Eoff+$lo]
469	ldr	$t1, [$ctx,#$Eoff+$hi]
470	ldr	$t2, [$ctx,#$Foff+$lo]
471	ldr	$t3, [$ctx,#$Foff+$hi]
472	adds	$Elo,$Elo,$t0
473	str	$Elo,[$ctx,#$Eoff+$lo]
474	adc	$Ehi,$Ehi,$t1
475	str	$Ehi,[$ctx,#$Eoff+$hi]
476	adds	$t2,$Tlo,$t2
477	str	$t2, [$ctx,#$Foff+$lo]
478	adc	$t3,$Thi,$t3
479	str	$t3, [$ctx,#$Foff+$hi]
480
481	ldr	$Alo,[sp,#$Goff+0]
482	ldr	$Ahi,[sp,#$Goff+4]
483	ldr	$Tlo,[sp,#$Hoff+0]
484	ldr	$Thi,[sp,#$Hoff+4]
485	ldr	$t0, [$ctx,#$Goff+$lo]
486	ldr	$t1, [$ctx,#$Goff+$hi]
487	ldr	$t2, [$ctx,#$Hoff+$lo]
488	ldr	$t3, [$ctx,#$Hoff+$hi]
489	adds	$t0,$Alo,$t0
490	str	$t0, [$ctx,#$Goff+$lo]
491	adc	$t1,$Ahi,$t1
492	str	$t1, [$ctx,#$Goff+$hi]
493	adds	$t2,$Tlo,$t2
494	str	$t2, [$ctx,#$Hoff+$lo]
495	adc	$t3,$Thi,$t3
496	str	$t3, [$ctx,#$Hoff+$hi]
497
498	add	sp,sp,#640
499	sub	$Ktbl,$Ktbl,#640
500
501	teq	$inp,$len
502	bne	.Loop
503
504	add	sp,sp,#8*9		@ destroy frame
505#if __ARM_ARCH__>=5
506	ldmia	sp!,{r4-r12,pc}
507#else
508	ldmia	sp!,{r4-r12,lr}
509	tst	lr,#1
510	moveq	pc,lr			@ be binary compatible with V4, yet
511	bx	lr			@ interoperable with Thumb ISA:-)
512#endif
513.size	sha512_block_data_order,.-sha512_block_data_order
514___
515
516{
517my @Sigma0=(28,34,39);
518my @Sigma1=(14,18,41);
519my @sigma0=(1, 8, 7);
520my @sigma1=(19,61,6);
521
522my $Ktbl="r3";
523my $cnt="r12";	# volatile register known as ip, intra-procedure-call scratch
524
525my @X=map("d$_",(0..15));
526my @V=($A,$B,$C,$D,$E,$F,$G,$H)=map("d$_",(16..23));
527
528sub NEON_00_15() {
529my $i=shift;
530my ($a,$b,$c,$d,$e,$f,$g,$h)=@_;
531my ($t0,$t1,$t2,$T1,$K,$Ch,$Maj)=map("d$_",(24..31));	# temps
532
533$code.=<<___ if ($i<16 || $i&1);
534	vshr.u64	$t0,$e,#@Sigma1[0]	@ $i
535#if $i<16
536	vld1.64		{@X[$i%16]},[$inp]!	@ handles unaligned
537#endif
538	vshr.u64	$t1,$e,#@Sigma1[1]
539#if $i>0
540	 vadd.i64	$a,$Maj			@ h+=Maj from the past
541#endif
542	vshr.u64	$t2,$e,#@Sigma1[2]
543___
544$code.=<<___;
545	vld1.64		{$K},[$Ktbl,:64]!	@ K[i++]
546	vsli.64		$t0,$e,#`64-@Sigma1[0]`
547	vsli.64		$t1,$e,#`64-@Sigma1[1]`
548	vmov		$Ch,$e
549	vsli.64		$t2,$e,#`64-@Sigma1[2]`
550#if $i<16 && defined(__ARMEL__)
551	vrev64.8	@X[$i],@X[$i]
552#endif
553	veor		$t1,$t0
554	vbsl		$Ch,$f,$g		@ Ch(e,f,g)
555	vshr.u64	$t0,$a,#@Sigma0[0]
556	veor		$t2,$t1			@ Sigma1(e)
557	vadd.i64	$T1,$Ch,$h
558	vshr.u64	$t1,$a,#@Sigma0[1]
559	vsli.64		$t0,$a,#`64-@Sigma0[0]`
560	vadd.i64	$T1,$t2
561	vshr.u64	$t2,$a,#@Sigma0[2]
562	vadd.i64	$K,@X[$i%16]
563	vsli.64		$t1,$a,#`64-@Sigma0[1]`
564	veor		$Maj,$a,$b
565	vsli.64		$t2,$a,#`64-@Sigma0[2]`
566	veor		$h,$t0,$t1
567	vadd.i64	$T1,$K
568	vbsl		$Maj,$c,$b		@ Maj(a,b,c)
569	veor		$h,$t2			@ Sigma0(a)
570	vadd.i64	$d,$T1
571	vadd.i64	$Maj,$T1
572	@ vadd.i64	$h,$Maj
573___
574}
575
576sub NEON_16_79() {
577my $i=shift;
578
579if ($i&1)	{ &NEON_00_15($i,@_); return; }
580
581# 2x-vectorized, therefore runs every 2nd round
582my @X=map("q$_",(0..7));			# view @X as 128-bit vector
583my ($t0,$t1,$s0,$s1) = map("q$_",(12..15));	# temps
584my ($d0,$d1,$d2) = map("d$_",(24..26));		# temps from NEON_00_15
585my $e=@_[4];					# $e from NEON_00_15
586$i /= 2;
587$code.=<<___;
588	vshr.u64	$t0,@X[($i+7)%8],#@sigma1[0]
589	vshr.u64	$t1,@X[($i+7)%8],#@sigma1[1]
590	 vadd.i64	@_[0],d30			@ h+=Maj from the past
591	vshr.u64	$s1,@X[($i+7)%8],#@sigma1[2]
592	vsli.64		$t0,@X[($i+7)%8],#`64-@sigma1[0]`
593	vext.8		$s0,@X[$i%8],@X[($i+1)%8],#8	@ X[i+1]
594	vsli.64		$t1,@X[($i+7)%8],#`64-@sigma1[1]`
595	veor		$s1,$t0
596	vshr.u64	$t0,$s0,#@sigma0[0]
597	veor		$s1,$t1				@ sigma1(X[i+14])
598	vshr.u64	$t1,$s0,#@sigma0[1]
599	vadd.i64	@X[$i%8],$s1
600	vshr.u64	$s1,$s0,#@sigma0[2]
601	vsli.64		$t0,$s0,#`64-@sigma0[0]`
602	vsli.64		$t1,$s0,#`64-@sigma0[1]`
603	vext.8		$s0,@X[($i+4)%8],@X[($i+5)%8],#8	@ X[i+9]
604	veor		$s1,$t0
605	vshr.u64	$d0,$e,#@Sigma1[0]		@ from NEON_00_15
606	vadd.i64	@X[$i%8],$s0
607	vshr.u64	$d1,$e,#@Sigma1[1]		@ from NEON_00_15
608	veor		$s1,$t1				@ sigma0(X[i+1])
609	vshr.u64	$d2,$e,#@Sigma1[2]		@ from NEON_00_15
610	vadd.i64	@X[$i%8],$s1
611___
612	&NEON_00_15(2*$i,@_);
613}
614
615$code.=<<___;
616#if __ARM_MAX_ARCH__>=7
617.arch	armv7-a
618.fpu	neon
619
620.global	sha512_block_data_order_neon
621.type	sha512_block_data_order_neon,%function
622.align	4
623sha512_block_data_order_neon:
624.LNEON:
625	dmb				@ errata #451034 on early Cortex A8
626	add	$len,$inp,$len,lsl#7	@ len to point at the end of inp
627	adr	$Ktbl,K512
628	VFP_ABI_PUSH
629	vldmia	$ctx,{$A-$H}		@ load context
630.Loop_neon:
631___
632for($i=0;$i<16;$i++)	{ &NEON_00_15($i,@V); unshift(@V,pop(@V)); }
633$code.=<<___;
634	mov		$cnt,#4
635.L16_79_neon:
636	subs		$cnt,#1
637___
638for(;$i<32;$i++)	{ &NEON_16_79($i,@V); unshift(@V,pop(@V)); }
639$code.=<<___;
640	bne		.L16_79_neon
641
642	 vadd.i64	$A,d30		@ h+=Maj from the past
643	vldmia		$ctx,{d24-d31}	@ load context to temp
644	vadd.i64	q8,q12		@ vectorized accumulate
645	vadd.i64	q9,q13
646	vadd.i64	q10,q14
647	vadd.i64	q11,q15
648	vstmia		$ctx,{$A-$H}	@ save context
649	teq		$inp,$len
650	sub		$Ktbl,#640	@ rewind K512
651	bne		.Loop_neon
652
653	VFP_ABI_POP
654	ret				@ bx lr
655.size	sha512_block_data_order_neon,.-sha512_block_data_order_neon
656#endif
657___
658}
659$code.=<<___;
660.asciz	"SHA512 block transform for ARMv4/NEON, CRYPTOGAMS by <appro\@openssl.org>"
661.align	2
662#if __ARM_MAX_ARCH__>=7 && !defined(__KERNEL__)
663.comm	OPENSSL_armcap_P,4,4
664#endif
665___
666
667$code =~ s/\`([^\`]*)\`/eval $1/gem;
668$code =~ s/\bbx\s+lr\b/.word\t0xe12fff1e/gm;	# make it possible to compile with -march=armv4
669$code =~ s/\bret\b/bx	lr/gm;
670
671open SELF,$0;
672while(<SELF>) {
673	next if (/^#!/);
674	last if (!s/^#/@/ and !/^$/);
675	print;
676}
677close SELF;
678
679print $code;
680close STDOUT or die "error closing STDOUT: $!"; # enforce flush
681