xref: /openbmc/linux/arch/arm/crypto/sha512-armv4.pl (revision fb574682)
1#!/usr/bin/env perl
2# SPDX-License-Identifier: GPL-2.0
3
4# This code is taken from the OpenSSL project but the author (Andy Polyakov)
5# has relicensed it under the GPLv2. Therefore this program is free software;
6# you can redistribute it and/or modify it under the terms of the GNU General
7# Public License version 2 as published by the Free Software Foundation.
8#
9# The original headers, including the original license headers, are
10# included below for completeness.
11
12# ====================================================================
13# Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
14# project. The module is, however, dual licensed under OpenSSL and
15# CRYPTOGAMS licenses depending on where you obtain it. For further
16# details see http://www.openssl.org/~appro/cryptogams/.
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
60while (($output=shift) && ($output!~/^\w[\w\-]*\.\w+$/)) {}
61open STDOUT,">$output";
62
63$ctx="r0";	# parameter block
64$inp="r1";
65$len="r2";
66
67$Tlo="r3";
68$Thi="r4";
69$Alo="r5";
70$Ahi="r6";
71$Elo="r7";
72$Ehi="r8";
73$t0="r9";
74$t1="r10";
75$t2="r11";
76$t3="r12";
77############	r13 is stack pointer
78$Ktbl="r14";
79############	r15 is program counter
80
81$Aoff=8*0;
82$Boff=8*1;
83$Coff=8*2;
84$Doff=8*3;
85$Eoff=8*4;
86$Foff=8*5;
87$Goff=8*6;
88$Hoff=8*7;
89$Xoff=8*8;
90
91sub BODY_00_15() {
92my $magic = shift;
93$code.=<<___;
94	@ Sigma1(x)	(ROTR((x),14) ^ ROTR((x),18)  ^ ROTR((x),41))
95	@ LO		lo>>14^hi<<18 ^ lo>>18^hi<<14 ^ hi>>9^lo<<23
96	@ HI		hi>>14^lo<<18 ^ hi>>18^lo<<14 ^ lo>>9^hi<<23
97	mov	$t0,$Elo,lsr#14
98	str	$Tlo,[sp,#$Xoff+0]
99	mov	$t1,$Ehi,lsr#14
100	str	$Thi,[sp,#$Xoff+4]
101	eor	$t0,$t0,$Ehi,lsl#18
102	ldr	$t2,[sp,#$Hoff+0]	@ h.lo
103	eor	$t1,$t1,$Elo,lsl#18
104	ldr	$t3,[sp,#$Hoff+4]	@ h.hi
105	eor	$t0,$t0,$Elo,lsr#18
106	eor	$t1,$t1,$Ehi,lsr#18
107	eor	$t0,$t0,$Ehi,lsl#14
108	eor	$t1,$t1,$Elo,lsl#14
109	eor	$t0,$t0,$Ehi,lsr#9
110	eor	$t1,$t1,$Elo,lsr#9
111	eor	$t0,$t0,$Elo,lsl#23
112	eor	$t1,$t1,$Ehi,lsl#23	@ Sigma1(e)
113	adds	$Tlo,$Tlo,$t0
114	ldr	$t0,[sp,#$Foff+0]	@ f.lo
115	adc	$Thi,$Thi,$t1		@ T += Sigma1(e)
116	ldr	$t1,[sp,#$Foff+4]	@ f.hi
117	adds	$Tlo,$Tlo,$t2
118	ldr	$t2,[sp,#$Goff+0]	@ g.lo
119	adc	$Thi,$Thi,$t3		@ T += h
120	ldr	$t3,[sp,#$Goff+4]	@ g.hi
121
122	eor	$t0,$t0,$t2
123	str	$Elo,[sp,#$Eoff+0]
124	eor	$t1,$t1,$t3
125	str	$Ehi,[sp,#$Eoff+4]
126	and	$t0,$t0,$Elo
127	str	$Alo,[sp,#$Aoff+0]
128	and	$t1,$t1,$Ehi
129	str	$Ahi,[sp,#$Aoff+4]
130	eor	$t0,$t0,$t2
131	ldr	$t2,[$Ktbl,#$lo]	@ K[i].lo
132	eor	$t1,$t1,$t3		@ Ch(e,f,g)
133	ldr	$t3,[$Ktbl,#$hi]	@ K[i].hi
134
135	adds	$Tlo,$Tlo,$t0
136	ldr	$Elo,[sp,#$Doff+0]	@ d.lo
137	adc	$Thi,$Thi,$t1		@ T += Ch(e,f,g)
138	ldr	$Ehi,[sp,#$Doff+4]	@ d.hi
139	adds	$Tlo,$Tlo,$t2
140	and	$t0,$t2,#0xff
141	adc	$Thi,$Thi,$t3		@ T += K[i]
142	adds	$Elo,$Elo,$Tlo
143	ldr	$t2,[sp,#$Boff+0]	@ b.lo
144	adc	$Ehi,$Ehi,$Thi		@ d += T
145	teq	$t0,#$magic
146
147	ldr	$t3,[sp,#$Coff+0]	@ c.lo
148#if __ARM_ARCH__>=7
149	it	eq			@ Thumb2 thing, sanity check in ARM
150#endif
151	orreq	$Ktbl,$Ktbl,#1
152	@ Sigma0(x)	(ROTR((x),28) ^ ROTR((x),34) ^ ROTR((x),39))
153	@ LO		lo>>28^hi<<4  ^ hi>>2^lo<<30 ^ hi>>7^lo<<25
154	@ HI		hi>>28^lo<<4  ^ lo>>2^hi<<30 ^ lo>>7^hi<<25
155	mov	$t0,$Alo,lsr#28
156	mov	$t1,$Ahi,lsr#28
157	eor	$t0,$t0,$Ahi,lsl#4
158	eor	$t1,$t1,$Alo,lsl#4
159	eor	$t0,$t0,$Ahi,lsr#2
160	eor	$t1,$t1,$Alo,lsr#2
161	eor	$t0,$t0,$Alo,lsl#30
162	eor	$t1,$t1,$Ahi,lsl#30
163	eor	$t0,$t0,$Ahi,lsr#7
164	eor	$t1,$t1,$Alo,lsr#7
165	eor	$t0,$t0,$Alo,lsl#25
166	eor	$t1,$t1,$Ahi,lsl#25	@ Sigma0(a)
167	adds	$Tlo,$Tlo,$t0
168	and	$t0,$Alo,$t2
169	adc	$Thi,$Thi,$t1		@ T += Sigma0(a)
170
171	ldr	$t1,[sp,#$Boff+4]	@ b.hi
172	orr	$Alo,$Alo,$t2
173	ldr	$t2,[sp,#$Coff+4]	@ c.hi
174	and	$Alo,$Alo,$t3
175	and	$t3,$Ahi,$t1
176	orr	$Ahi,$Ahi,$t1
177	orr	$Alo,$Alo,$t0		@ Maj(a,b,c).lo
178	and	$Ahi,$Ahi,$t2
179	adds	$Alo,$Alo,$Tlo
180	orr	$Ahi,$Ahi,$t3		@ Maj(a,b,c).hi
181	sub	sp,sp,#8
182	adc	$Ahi,$Ahi,$Thi		@ h += T
183	tst	$Ktbl,#1
184	add	$Ktbl,$Ktbl,#8
185___
186}
187$code=<<___;
188#ifndef __KERNEL__
189# include "arm_arch.h"
190# define VFP_ABI_PUSH	vstmdb	sp!,{d8-d15}
191# define VFP_ABI_POP	vldmia	sp!,{d8-d15}
192#else
193# define __ARM_ARCH__ __LINUX_ARM_ARCH__
194# define __ARM_MAX_ARCH__ 7
195# define VFP_ABI_PUSH
196# define VFP_ABI_POP
197#endif
198
199#ifdef __ARMEL__
200# define LO 0
201# define HI 4
202# define WORD64(hi0,lo0,hi1,lo1)	.word	lo0,hi0, lo1,hi1
203#else
204# define HI 0
205# define LO 4
206# define WORD64(hi0,lo0,hi1,lo1)	.word	hi0,lo0, hi1,lo1
207#endif
208
209.text
210#if __ARM_ARCH__<7
211.code	32
212#else
213.syntax unified
214# ifdef __thumb2__
215#  define adrl adr
216.thumb
217# else
218.code   32
219# endif
220#endif
221
222.type	K512,%object
223.align	5
224K512:
225WORD64(0x428a2f98,0xd728ae22, 0x71374491,0x23ef65cd)
226WORD64(0xb5c0fbcf,0xec4d3b2f, 0xe9b5dba5,0x8189dbbc)
227WORD64(0x3956c25b,0xf348b538, 0x59f111f1,0xb605d019)
228WORD64(0x923f82a4,0xaf194f9b, 0xab1c5ed5,0xda6d8118)
229WORD64(0xd807aa98,0xa3030242, 0x12835b01,0x45706fbe)
230WORD64(0x243185be,0x4ee4b28c, 0x550c7dc3,0xd5ffb4e2)
231WORD64(0x72be5d74,0xf27b896f, 0x80deb1fe,0x3b1696b1)
232WORD64(0x9bdc06a7,0x25c71235, 0xc19bf174,0xcf692694)
233WORD64(0xe49b69c1,0x9ef14ad2, 0xefbe4786,0x384f25e3)
234WORD64(0x0fc19dc6,0x8b8cd5b5, 0x240ca1cc,0x77ac9c65)
235WORD64(0x2de92c6f,0x592b0275, 0x4a7484aa,0x6ea6e483)
236WORD64(0x5cb0a9dc,0xbd41fbd4, 0x76f988da,0x831153b5)
237WORD64(0x983e5152,0xee66dfab, 0xa831c66d,0x2db43210)
238WORD64(0xb00327c8,0x98fb213f, 0xbf597fc7,0xbeef0ee4)
239WORD64(0xc6e00bf3,0x3da88fc2, 0xd5a79147,0x930aa725)
240WORD64(0x06ca6351,0xe003826f, 0x14292967,0x0a0e6e70)
241WORD64(0x27b70a85,0x46d22ffc, 0x2e1b2138,0x5c26c926)
242WORD64(0x4d2c6dfc,0x5ac42aed, 0x53380d13,0x9d95b3df)
243WORD64(0x650a7354,0x8baf63de, 0x766a0abb,0x3c77b2a8)
244WORD64(0x81c2c92e,0x47edaee6, 0x92722c85,0x1482353b)
245WORD64(0xa2bfe8a1,0x4cf10364, 0xa81a664b,0xbc423001)
246WORD64(0xc24b8b70,0xd0f89791, 0xc76c51a3,0x0654be30)
247WORD64(0xd192e819,0xd6ef5218, 0xd6990624,0x5565a910)
248WORD64(0xf40e3585,0x5771202a, 0x106aa070,0x32bbd1b8)
249WORD64(0x19a4c116,0xb8d2d0c8, 0x1e376c08,0x5141ab53)
250WORD64(0x2748774c,0xdf8eeb99, 0x34b0bcb5,0xe19b48a8)
251WORD64(0x391c0cb3,0xc5c95a63, 0x4ed8aa4a,0xe3418acb)
252WORD64(0x5b9cca4f,0x7763e373, 0x682e6ff3,0xd6b2b8a3)
253WORD64(0x748f82ee,0x5defb2fc, 0x78a5636f,0x43172f60)
254WORD64(0x84c87814,0xa1f0ab72, 0x8cc70208,0x1a6439ec)
255WORD64(0x90befffa,0x23631e28, 0xa4506ceb,0xde82bde9)
256WORD64(0xbef9a3f7,0xb2c67915, 0xc67178f2,0xe372532b)
257WORD64(0xca273ece,0xea26619c, 0xd186b8c7,0x21c0c207)
258WORD64(0xeada7dd6,0xcde0eb1e, 0xf57d4f7f,0xee6ed178)
259WORD64(0x06f067aa,0x72176fba, 0x0a637dc5,0xa2c898a6)
260WORD64(0x113f9804,0xbef90dae, 0x1b710b35,0x131c471b)
261WORD64(0x28db77f5,0x23047d84, 0x32caab7b,0x40c72493)
262WORD64(0x3c9ebe0a,0x15c9bebc, 0x431d67c4,0x9c100d4c)
263WORD64(0x4cc5d4be,0xcb3e42b6, 0x597f299c,0xfc657e2a)
264WORD64(0x5fcb6fab,0x3ad6faec, 0x6c44198c,0x4a475817)
265.size	K512,.-K512
266#if __ARM_MAX_ARCH__>=7 && !defined(__KERNEL__)
267.LOPENSSL_armcap:
268.word	OPENSSL_armcap_P-sha512_block_data_order
269.skip	32-4
270#else
271.skip	32
272#endif
273
274.global	sha512_block_data_order
275.type	sha512_block_data_order,%function
276sha512_block_data_order:
277.Lsha512_block_data_order:
278#if __ARM_ARCH__<7
279	sub	r3,pc,#8		@ sha512_block_data_order
280#else
281	adr	r3,.Lsha512_block_data_order
282#endif
283#if __ARM_MAX_ARCH__>=7 && !defined(__KERNEL__)
284	ldr	r12,.LOPENSSL_armcap
285	ldr	r12,[r3,r12]		@ OPENSSL_armcap_P
286	tst	r12,#1
287	bne	.LNEON
288#endif
289	add	$len,$inp,$len,lsl#7	@ len to point at the end of inp
290	stmdb	sp!,{r4-r12,lr}
291	sub	$Ktbl,r3,#672		@ K512
292	sub	sp,sp,#9*8
293
294	ldr	$Elo,[$ctx,#$Eoff+$lo]
295	ldr	$Ehi,[$ctx,#$Eoff+$hi]
296	ldr	$t0, [$ctx,#$Goff+$lo]
297	ldr	$t1, [$ctx,#$Goff+$hi]
298	ldr	$t2, [$ctx,#$Hoff+$lo]
299	ldr	$t3, [$ctx,#$Hoff+$hi]
300.Loop:
301	str	$t0, [sp,#$Goff+0]
302	str	$t1, [sp,#$Goff+4]
303	str	$t2, [sp,#$Hoff+0]
304	str	$t3, [sp,#$Hoff+4]
305	ldr	$Alo,[$ctx,#$Aoff+$lo]
306	ldr	$Ahi,[$ctx,#$Aoff+$hi]
307	ldr	$Tlo,[$ctx,#$Boff+$lo]
308	ldr	$Thi,[$ctx,#$Boff+$hi]
309	ldr	$t0, [$ctx,#$Coff+$lo]
310	ldr	$t1, [$ctx,#$Coff+$hi]
311	ldr	$t2, [$ctx,#$Doff+$lo]
312	ldr	$t3, [$ctx,#$Doff+$hi]
313	str	$Tlo,[sp,#$Boff+0]
314	str	$Thi,[sp,#$Boff+4]
315	str	$t0, [sp,#$Coff+0]
316	str	$t1, [sp,#$Coff+4]
317	str	$t2, [sp,#$Doff+0]
318	str	$t3, [sp,#$Doff+4]
319	ldr	$Tlo,[$ctx,#$Foff+$lo]
320	ldr	$Thi,[$ctx,#$Foff+$hi]
321	str	$Tlo,[sp,#$Foff+0]
322	str	$Thi,[sp,#$Foff+4]
323
324.L00_15:
325#if __ARM_ARCH__<7
326	ldrb	$Tlo,[$inp,#7]
327	ldrb	$t0, [$inp,#6]
328	ldrb	$t1, [$inp,#5]
329	ldrb	$t2, [$inp,#4]
330	ldrb	$Thi,[$inp,#3]
331	ldrb	$t3, [$inp,#2]
332	orr	$Tlo,$Tlo,$t0,lsl#8
333	ldrb	$t0, [$inp,#1]
334	orr	$Tlo,$Tlo,$t1,lsl#16
335	ldrb	$t1, [$inp],#8
336	orr	$Tlo,$Tlo,$t2,lsl#24
337	orr	$Thi,$Thi,$t3,lsl#8
338	orr	$Thi,$Thi,$t0,lsl#16
339	orr	$Thi,$Thi,$t1,lsl#24
340#else
341	ldr	$Tlo,[$inp,#4]
342	ldr	$Thi,[$inp],#8
343#ifdef __ARMEL__
344	rev	$Tlo,$Tlo
345	rev	$Thi,$Thi
346#endif
347#endif
348___
349	&BODY_00_15(0x94);
350$code.=<<___;
351	tst	$Ktbl,#1
352	beq	.L00_15
353	ldr	$t0,[sp,#`$Xoff+8*(16-1)`+0]
354	ldr	$t1,[sp,#`$Xoff+8*(16-1)`+4]
355	bic	$Ktbl,$Ktbl,#1
356.L16_79:
357	@ sigma0(x)	(ROTR((x),1)  ^ ROTR((x),8)  ^ ((x)>>7))
358	@ LO		lo>>1^hi<<31  ^ lo>>8^hi<<24 ^ lo>>7^hi<<25
359	@ HI		hi>>1^lo<<31  ^ hi>>8^lo<<24 ^ hi>>7
360	mov	$Tlo,$t0,lsr#1
361	ldr	$t2,[sp,#`$Xoff+8*(16-14)`+0]
362	mov	$Thi,$t1,lsr#1
363	ldr	$t3,[sp,#`$Xoff+8*(16-14)`+4]
364	eor	$Tlo,$Tlo,$t1,lsl#31
365	eor	$Thi,$Thi,$t0,lsl#31
366	eor	$Tlo,$Tlo,$t0,lsr#8
367	eor	$Thi,$Thi,$t1,lsr#8
368	eor	$Tlo,$Tlo,$t1,lsl#24
369	eor	$Thi,$Thi,$t0,lsl#24
370	eor	$Tlo,$Tlo,$t0,lsr#7
371	eor	$Thi,$Thi,$t1,lsr#7
372	eor	$Tlo,$Tlo,$t1,lsl#25
373
374	@ sigma1(x)	(ROTR((x),19) ^ ROTR((x),61) ^ ((x)>>6))
375	@ LO		lo>>19^hi<<13 ^ hi>>29^lo<<3 ^ lo>>6^hi<<26
376	@ HI		hi>>19^lo<<13 ^ lo>>29^hi<<3 ^ hi>>6
377	mov	$t0,$t2,lsr#19
378	mov	$t1,$t3,lsr#19
379	eor	$t0,$t0,$t3,lsl#13
380	eor	$t1,$t1,$t2,lsl#13
381	eor	$t0,$t0,$t3,lsr#29
382	eor	$t1,$t1,$t2,lsr#29
383	eor	$t0,$t0,$t2,lsl#3
384	eor	$t1,$t1,$t3,lsl#3
385	eor	$t0,$t0,$t2,lsr#6
386	eor	$t1,$t1,$t3,lsr#6
387	ldr	$t2,[sp,#`$Xoff+8*(16-9)`+0]
388	eor	$t0,$t0,$t3,lsl#26
389
390	ldr	$t3,[sp,#`$Xoff+8*(16-9)`+4]
391	adds	$Tlo,$Tlo,$t0
392	ldr	$t0,[sp,#`$Xoff+8*16`+0]
393	adc	$Thi,$Thi,$t1
394
395	ldr	$t1,[sp,#`$Xoff+8*16`+4]
396	adds	$Tlo,$Tlo,$t2
397	adc	$Thi,$Thi,$t3
398	adds	$Tlo,$Tlo,$t0
399	adc	$Thi,$Thi,$t1
400___
401	&BODY_00_15(0x17);
402$code.=<<___;
403#if __ARM_ARCH__>=7
404	ittt	eq			@ Thumb2 thing, sanity check in ARM
405#endif
406	ldreq	$t0,[sp,#`$Xoff+8*(16-1)`+0]
407	ldreq	$t1,[sp,#`$Xoff+8*(16-1)`+4]
408	beq	.L16_79
409	bic	$Ktbl,$Ktbl,#1
410
411	ldr	$Tlo,[sp,#$Boff+0]
412	ldr	$Thi,[sp,#$Boff+4]
413	ldr	$t0, [$ctx,#$Aoff+$lo]
414	ldr	$t1, [$ctx,#$Aoff+$hi]
415	ldr	$t2, [$ctx,#$Boff+$lo]
416	ldr	$t3, [$ctx,#$Boff+$hi]
417	adds	$t0,$Alo,$t0
418	str	$t0, [$ctx,#$Aoff+$lo]
419	adc	$t1,$Ahi,$t1
420	str	$t1, [$ctx,#$Aoff+$hi]
421	adds	$t2,$Tlo,$t2
422	str	$t2, [$ctx,#$Boff+$lo]
423	adc	$t3,$Thi,$t3
424	str	$t3, [$ctx,#$Boff+$hi]
425
426	ldr	$Alo,[sp,#$Coff+0]
427	ldr	$Ahi,[sp,#$Coff+4]
428	ldr	$Tlo,[sp,#$Doff+0]
429	ldr	$Thi,[sp,#$Doff+4]
430	ldr	$t0, [$ctx,#$Coff+$lo]
431	ldr	$t1, [$ctx,#$Coff+$hi]
432	ldr	$t2, [$ctx,#$Doff+$lo]
433	ldr	$t3, [$ctx,#$Doff+$hi]
434	adds	$t0,$Alo,$t0
435	str	$t0, [$ctx,#$Coff+$lo]
436	adc	$t1,$Ahi,$t1
437	str	$t1, [$ctx,#$Coff+$hi]
438	adds	$t2,$Tlo,$t2
439	str	$t2, [$ctx,#$Doff+$lo]
440	adc	$t3,$Thi,$t3
441	str	$t3, [$ctx,#$Doff+$hi]
442
443	ldr	$Tlo,[sp,#$Foff+0]
444	ldr	$Thi,[sp,#$Foff+4]
445	ldr	$t0, [$ctx,#$Eoff+$lo]
446	ldr	$t1, [$ctx,#$Eoff+$hi]
447	ldr	$t2, [$ctx,#$Foff+$lo]
448	ldr	$t3, [$ctx,#$Foff+$hi]
449	adds	$Elo,$Elo,$t0
450	str	$Elo,[$ctx,#$Eoff+$lo]
451	adc	$Ehi,$Ehi,$t1
452	str	$Ehi,[$ctx,#$Eoff+$hi]
453	adds	$t2,$Tlo,$t2
454	str	$t2, [$ctx,#$Foff+$lo]
455	adc	$t3,$Thi,$t3
456	str	$t3, [$ctx,#$Foff+$hi]
457
458	ldr	$Alo,[sp,#$Goff+0]
459	ldr	$Ahi,[sp,#$Goff+4]
460	ldr	$Tlo,[sp,#$Hoff+0]
461	ldr	$Thi,[sp,#$Hoff+4]
462	ldr	$t0, [$ctx,#$Goff+$lo]
463	ldr	$t1, [$ctx,#$Goff+$hi]
464	ldr	$t2, [$ctx,#$Hoff+$lo]
465	ldr	$t3, [$ctx,#$Hoff+$hi]
466	adds	$t0,$Alo,$t0
467	str	$t0, [$ctx,#$Goff+$lo]
468	adc	$t1,$Ahi,$t1
469	str	$t1, [$ctx,#$Goff+$hi]
470	adds	$t2,$Tlo,$t2
471	str	$t2, [$ctx,#$Hoff+$lo]
472	adc	$t3,$Thi,$t3
473	str	$t3, [$ctx,#$Hoff+$hi]
474
475	add	sp,sp,#640
476	sub	$Ktbl,$Ktbl,#640
477
478	teq	$inp,$len
479	bne	.Loop
480
481	add	sp,sp,#8*9		@ destroy frame
482#if __ARM_ARCH__>=5
483	ldmia	sp!,{r4-r12,pc}
484#else
485	ldmia	sp!,{r4-r12,lr}
486	tst	lr,#1
487	moveq	pc,lr			@ be binary compatible with V4, yet
488	bx	lr			@ interoperable with Thumb ISA:-)
489#endif
490.size	sha512_block_data_order,.-sha512_block_data_order
491___
492
493{
494my @Sigma0=(28,34,39);
495my @Sigma1=(14,18,41);
496my @sigma0=(1, 8, 7);
497my @sigma1=(19,61,6);
498
499my $Ktbl="r3";
500my $cnt="r12";	# volatile register known as ip, intra-procedure-call scratch
501
502my @X=map("d$_",(0..15));
503my @V=($A,$B,$C,$D,$E,$F,$G,$H)=map("d$_",(16..23));
504
505sub NEON_00_15() {
506my $i=shift;
507my ($a,$b,$c,$d,$e,$f,$g,$h)=@_;
508my ($t0,$t1,$t2,$T1,$K,$Ch,$Maj)=map("d$_",(24..31));	# temps
509
510$code.=<<___ if ($i<16 || $i&1);
511	vshr.u64	$t0,$e,#@Sigma1[0]	@ $i
512#if $i<16
513	vld1.64		{@X[$i%16]},[$inp]!	@ handles unaligned
514#endif
515	vshr.u64	$t1,$e,#@Sigma1[1]
516#if $i>0
517	 vadd.i64	$a,$Maj			@ h+=Maj from the past
518#endif
519	vshr.u64	$t2,$e,#@Sigma1[2]
520___
521$code.=<<___;
522	vld1.64		{$K},[$Ktbl,:64]!	@ K[i++]
523	vsli.64		$t0,$e,#`64-@Sigma1[0]`
524	vsli.64		$t1,$e,#`64-@Sigma1[1]`
525	vmov		$Ch,$e
526	vsli.64		$t2,$e,#`64-@Sigma1[2]`
527#if $i<16 && defined(__ARMEL__)
528	vrev64.8	@X[$i],@X[$i]
529#endif
530	veor		$t1,$t0
531	vbsl		$Ch,$f,$g		@ Ch(e,f,g)
532	vshr.u64	$t0,$a,#@Sigma0[0]
533	veor		$t2,$t1			@ Sigma1(e)
534	vadd.i64	$T1,$Ch,$h
535	vshr.u64	$t1,$a,#@Sigma0[1]
536	vsli.64		$t0,$a,#`64-@Sigma0[0]`
537	vadd.i64	$T1,$t2
538	vshr.u64	$t2,$a,#@Sigma0[2]
539	vadd.i64	$K,@X[$i%16]
540	vsli.64		$t1,$a,#`64-@Sigma0[1]`
541	veor		$Maj,$a,$b
542	vsli.64		$t2,$a,#`64-@Sigma0[2]`
543	veor		$h,$t0,$t1
544	vadd.i64	$T1,$K
545	vbsl		$Maj,$c,$b		@ Maj(a,b,c)
546	veor		$h,$t2			@ Sigma0(a)
547	vadd.i64	$d,$T1
548	vadd.i64	$Maj,$T1
549	@ vadd.i64	$h,$Maj
550___
551}
552
553sub NEON_16_79() {
554my $i=shift;
555
556if ($i&1)	{ &NEON_00_15($i,@_); return; }
557
558# 2x-vectorized, therefore runs every 2nd round
559my @X=map("q$_",(0..7));			# view @X as 128-bit vector
560my ($t0,$t1,$s0,$s1) = map("q$_",(12..15));	# temps
561my ($d0,$d1,$d2) = map("d$_",(24..26));		# temps from NEON_00_15
562my $e=@_[4];					# $e from NEON_00_15
563$i /= 2;
564$code.=<<___;
565	vshr.u64	$t0,@X[($i+7)%8],#@sigma1[0]
566	vshr.u64	$t1,@X[($i+7)%8],#@sigma1[1]
567	 vadd.i64	@_[0],d30			@ h+=Maj from the past
568	vshr.u64	$s1,@X[($i+7)%8],#@sigma1[2]
569	vsli.64		$t0,@X[($i+7)%8],#`64-@sigma1[0]`
570	vext.8		$s0,@X[$i%8],@X[($i+1)%8],#8	@ X[i+1]
571	vsli.64		$t1,@X[($i+7)%8],#`64-@sigma1[1]`
572	veor		$s1,$t0
573	vshr.u64	$t0,$s0,#@sigma0[0]
574	veor		$s1,$t1				@ sigma1(X[i+14])
575	vshr.u64	$t1,$s0,#@sigma0[1]
576	vadd.i64	@X[$i%8],$s1
577	vshr.u64	$s1,$s0,#@sigma0[2]
578	vsli.64		$t0,$s0,#`64-@sigma0[0]`
579	vsli.64		$t1,$s0,#`64-@sigma0[1]`
580	vext.8		$s0,@X[($i+4)%8],@X[($i+5)%8],#8	@ X[i+9]
581	veor		$s1,$t0
582	vshr.u64	$d0,$e,#@Sigma1[0]		@ from NEON_00_15
583	vadd.i64	@X[$i%8],$s0
584	vshr.u64	$d1,$e,#@Sigma1[1]		@ from NEON_00_15
585	veor		$s1,$t1				@ sigma0(X[i+1])
586	vshr.u64	$d2,$e,#@Sigma1[2]		@ from NEON_00_15
587	vadd.i64	@X[$i%8],$s1
588___
589	&NEON_00_15(2*$i,@_);
590}
591
592$code.=<<___;
593#if __ARM_MAX_ARCH__>=7
594.arch	armv7-a
595.fpu	neon
596
597.global	sha512_block_data_order_neon
598.type	sha512_block_data_order_neon,%function
599.align	4
600sha512_block_data_order_neon:
601.LNEON:
602	dmb				@ errata #451034 on early Cortex A8
603	add	$len,$inp,$len,lsl#7	@ len to point at the end of inp
604	VFP_ABI_PUSH
605	adrl	$Ktbl,K512
606	vldmia	$ctx,{$A-$H}		@ load context
607.Loop_neon:
608___
609for($i=0;$i<16;$i++)	{ &NEON_00_15($i,@V); unshift(@V,pop(@V)); }
610$code.=<<___;
611	mov		$cnt,#4
612.L16_79_neon:
613	subs		$cnt,#1
614___
615for(;$i<32;$i++)	{ &NEON_16_79($i,@V); unshift(@V,pop(@V)); }
616$code.=<<___;
617	bne		.L16_79_neon
618
619	 vadd.i64	$A,d30		@ h+=Maj from the past
620	vldmia		$ctx,{d24-d31}	@ load context to temp
621	vadd.i64	q8,q12		@ vectorized accumulate
622	vadd.i64	q9,q13
623	vadd.i64	q10,q14
624	vadd.i64	q11,q15
625	vstmia		$ctx,{$A-$H}	@ save context
626	teq		$inp,$len
627	sub		$Ktbl,#640	@ rewind K512
628	bne		.Loop_neon
629
630	VFP_ABI_POP
631	ret				@ bx lr
632.size	sha512_block_data_order_neon,.-sha512_block_data_order_neon
633#endif
634___
635}
636$code.=<<___;
637.asciz	"SHA512 block transform for ARMv4/NEON, CRYPTOGAMS by <appro\@openssl.org>"
638.align	2
639#if __ARM_MAX_ARCH__>=7 && !defined(__KERNEL__)
640.comm	OPENSSL_armcap_P,4,4
641#endif
642___
643
644$code =~ s/\`([^\`]*)\`/eval $1/gem;
645$code =~ s/\bbx\s+lr\b/.word\t0xe12fff1e/gm;	# make it possible to compile with -march=armv4
646$code =~ s/\bret\b/bx	lr/gm;
647
648open SELF,$0;
649while(<SELF>) {
650	next if (/^#!/);
651	last if (!s/^#/@/ and !/^$/);
652	print;
653}
654close SELF;
655
656print $code;
657close STDOUT; # enforce flush
658