xref: /openbmc/linux/arch/arm/crypto/sha512-armv4.pl (revision 63705da3)
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 https://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 https://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.thumb
216# else
217.code   32
218# endif
219#endif
220
221.type	K512,%object
222.align	5
223K512:
224WORD64(0x428a2f98,0xd728ae22, 0x71374491,0x23ef65cd)
225WORD64(0xb5c0fbcf,0xec4d3b2f, 0xe9b5dba5,0x8189dbbc)
226WORD64(0x3956c25b,0xf348b538, 0x59f111f1,0xb605d019)
227WORD64(0x923f82a4,0xaf194f9b, 0xab1c5ed5,0xda6d8118)
228WORD64(0xd807aa98,0xa3030242, 0x12835b01,0x45706fbe)
229WORD64(0x243185be,0x4ee4b28c, 0x550c7dc3,0xd5ffb4e2)
230WORD64(0x72be5d74,0xf27b896f, 0x80deb1fe,0x3b1696b1)
231WORD64(0x9bdc06a7,0x25c71235, 0xc19bf174,0xcf692694)
232WORD64(0xe49b69c1,0x9ef14ad2, 0xefbe4786,0x384f25e3)
233WORD64(0x0fc19dc6,0x8b8cd5b5, 0x240ca1cc,0x77ac9c65)
234WORD64(0x2de92c6f,0x592b0275, 0x4a7484aa,0x6ea6e483)
235WORD64(0x5cb0a9dc,0xbd41fbd4, 0x76f988da,0x831153b5)
236WORD64(0x983e5152,0xee66dfab, 0xa831c66d,0x2db43210)
237WORD64(0xb00327c8,0x98fb213f, 0xbf597fc7,0xbeef0ee4)
238WORD64(0xc6e00bf3,0x3da88fc2, 0xd5a79147,0x930aa725)
239WORD64(0x06ca6351,0xe003826f, 0x14292967,0x0a0e6e70)
240WORD64(0x27b70a85,0x46d22ffc, 0x2e1b2138,0x5c26c926)
241WORD64(0x4d2c6dfc,0x5ac42aed, 0x53380d13,0x9d95b3df)
242WORD64(0x650a7354,0x8baf63de, 0x766a0abb,0x3c77b2a8)
243WORD64(0x81c2c92e,0x47edaee6, 0x92722c85,0x1482353b)
244WORD64(0xa2bfe8a1,0x4cf10364, 0xa81a664b,0xbc423001)
245WORD64(0xc24b8b70,0xd0f89791, 0xc76c51a3,0x0654be30)
246WORD64(0xd192e819,0xd6ef5218, 0xd6990624,0x5565a910)
247WORD64(0xf40e3585,0x5771202a, 0x106aa070,0x32bbd1b8)
248WORD64(0x19a4c116,0xb8d2d0c8, 0x1e376c08,0x5141ab53)
249WORD64(0x2748774c,0xdf8eeb99, 0x34b0bcb5,0xe19b48a8)
250WORD64(0x391c0cb3,0xc5c95a63, 0x4ed8aa4a,0xe3418acb)
251WORD64(0x5b9cca4f,0x7763e373, 0x682e6ff3,0xd6b2b8a3)
252WORD64(0x748f82ee,0x5defb2fc, 0x78a5636f,0x43172f60)
253WORD64(0x84c87814,0xa1f0ab72, 0x8cc70208,0x1a6439ec)
254WORD64(0x90befffa,0x23631e28, 0xa4506ceb,0xde82bde9)
255WORD64(0xbef9a3f7,0xb2c67915, 0xc67178f2,0xe372532b)
256WORD64(0xca273ece,0xea26619c, 0xd186b8c7,0x21c0c207)
257WORD64(0xeada7dd6,0xcde0eb1e, 0xf57d4f7f,0xee6ed178)
258WORD64(0x06f067aa,0x72176fba, 0x0a637dc5,0xa2c898a6)
259WORD64(0x113f9804,0xbef90dae, 0x1b710b35,0x131c471b)
260WORD64(0x28db77f5,0x23047d84, 0x32caab7b,0x40c72493)
261WORD64(0x3c9ebe0a,0x15c9bebc, 0x431d67c4,0x9c100d4c)
262WORD64(0x4cc5d4be,0xcb3e42b6, 0x597f299c,0xfc657e2a)
263WORD64(0x5fcb6fab,0x3ad6faec, 0x6c44198c,0x4a475817)
264.size	K512,.-K512
265#if __ARM_MAX_ARCH__>=7 && !defined(__KERNEL__)
266.LOPENSSL_armcap:
267.word	OPENSSL_armcap_P-sha512_block_data_order
268.skip	32-4
269#else
270.skip	32
271#endif
272
273.global	sha512_block_data_order
274.type	sha512_block_data_order,%function
275sha512_block_data_order:
276.Lsha512_block_data_order:
277#if __ARM_ARCH__<7
278	sub	r3,pc,#8		@ sha512_block_data_order
279#else
280	adr	r3,.Lsha512_block_data_order
281#endif
282#if __ARM_MAX_ARCH__>=7 && !defined(__KERNEL__)
283	ldr	r12,.LOPENSSL_armcap
284	ldr	r12,[r3,r12]		@ OPENSSL_armcap_P
285	tst	r12,#1
286	bne	.LNEON
287#endif
288	add	$len,$inp,$len,lsl#7	@ len to point at the end of inp
289	stmdb	sp!,{r4-r12,lr}
290	sub	$Ktbl,r3,#672		@ K512
291	sub	sp,sp,#9*8
292
293	ldr	$Elo,[$ctx,#$Eoff+$lo]
294	ldr	$Ehi,[$ctx,#$Eoff+$hi]
295	ldr	$t0, [$ctx,#$Goff+$lo]
296	ldr	$t1, [$ctx,#$Goff+$hi]
297	ldr	$t2, [$ctx,#$Hoff+$lo]
298	ldr	$t3, [$ctx,#$Hoff+$hi]
299.Loop:
300	str	$t0, [sp,#$Goff+0]
301	str	$t1, [sp,#$Goff+4]
302	str	$t2, [sp,#$Hoff+0]
303	str	$t3, [sp,#$Hoff+4]
304	ldr	$Alo,[$ctx,#$Aoff+$lo]
305	ldr	$Ahi,[$ctx,#$Aoff+$hi]
306	ldr	$Tlo,[$ctx,#$Boff+$lo]
307	ldr	$Thi,[$ctx,#$Boff+$hi]
308	ldr	$t0, [$ctx,#$Coff+$lo]
309	ldr	$t1, [$ctx,#$Coff+$hi]
310	ldr	$t2, [$ctx,#$Doff+$lo]
311	ldr	$t3, [$ctx,#$Doff+$hi]
312	str	$Tlo,[sp,#$Boff+0]
313	str	$Thi,[sp,#$Boff+4]
314	str	$t0, [sp,#$Coff+0]
315	str	$t1, [sp,#$Coff+4]
316	str	$t2, [sp,#$Doff+0]
317	str	$t3, [sp,#$Doff+4]
318	ldr	$Tlo,[$ctx,#$Foff+$lo]
319	ldr	$Thi,[$ctx,#$Foff+$hi]
320	str	$Tlo,[sp,#$Foff+0]
321	str	$Thi,[sp,#$Foff+4]
322
323.L00_15:
324#if __ARM_ARCH__<7
325	ldrb	$Tlo,[$inp,#7]
326	ldrb	$t0, [$inp,#6]
327	ldrb	$t1, [$inp,#5]
328	ldrb	$t2, [$inp,#4]
329	ldrb	$Thi,[$inp,#3]
330	ldrb	$t3, [$inp,#2]
331	orr	$Tlo,$Tlo,$t0,lsl#8
332	ldrb	$t0, [$inp,#1]
333	orr	$Tlo,$Tlo,$t1,lsl#16
334	ldrb	$t1, [$inp],#8
335	orr	$Tlo,$Tlo,$t2,lsl#24
336	orr	$Thi,$Thi,$t3,lsl#8
337	orr	$Thi,$Thi,$t0,lsl#16
338	orr	$Thi,$Thi,$t1,lsl#24
339#else
340	ldr	$Tlo,[$inp,#4]
341	ldr	$Thi,[$inp],#8
342#ifdef __ARMEL__
343	rev	$Tlo,$Tlo
344	rev	$Thi,$Thi
345#endif
346#endif
347___
348	&BODY_00_15(0x94);
349$code.=<<___;
350	tst	$Ktbl,#1
351	beq	.L00_15
352	ldr	$t0,[sp,#`$Xoff+8*(16-1)`+0]
353	ldr	$t1,[sp,#`$Xoff+8*(16-1)`+4]
354	bic	$Ktbl,$Ktbl,#1
355.L16_79:
356	@ sigma0(x)	(ROTR((x),1)  ^ ROTR((x),8)  ^ ((x)>>7))
357	@ LO		lo>>1^hi<<31  ^ lo>>8^hi<<24 ^ lo>>7^hi<<25
358	@ HI		hi>>1^lo<<31  ^ hi>>8^lo<<24 ^ hi>>7
359	mov	$Tlo,$t0,lsr#1
360	ldr	$t2,[sp,#`$Xoff+8*(16-14)`+0]
361	mov	$Thi,$t1,lsr#1
362	ldr	$t3,[sp,#`$Xoff+8*(16-14)`+4]
363	eor	$Tlo,$Tlo,$t1,lsl#31
364	eor	$Thi,$Thi,$t0,lsl#31
365	eor	$Tlo,$Tlo,$t0,lsr#8
366	eor	$Thi,$Thi,$t1,lsr#8
367	eor	$Tlo,$Tlo,$t1,lsl#24
368	eor	$Thi,$Thi,$t0,lsl#24
369	eor	$Tlo,$Tlo,$t0,lsr#7
370	eor	$Thi,$Thi,$t1,lsr#7
371	eor	$Tlo,$Tlo,$t1,lsl#25
372
373	@ sigma1(x)	(ROTR((x),19) ^ ROTR((x),61) ^ ((x)>>6))
374	@ LO		lo>>19^hi<<13 ^ hi>>29^lo<<3 ^ lo>>6^hi<<26
375	@ HI		hi>>19^lo<<13 ^ lo>>29^hi<<3 ^ hi>>6
376	mov	$t0,$t2,lsr#19
377	mov	$t1,$t3,lsr#19
378	eor	$t0,$t0,$t3,lsl#13
379	eor	$t1,$t1,$t2,lsl#13
380	eor	$t0,$t0,$t3,lsr#29
381	eor	$t1,$t1,$t2,lsr#29
382	eor	$t0,$t0,$t2,lsl#3
383	eor	$t1,$t1,$t3,lsl#3
384	eor	$t0,$t0,$t2,lsr#6
385	eor	$t1,$t1,$t3,lsr#6
386	ldr	$t2,[sp,#`$Xoff+8*(16-9)`+0]
387	eor	$t0,$t0,$t3,lsl#26
388
389	ldr	$t3,[sp,#`$Xoff+8*(16-9)`+4]
390	adds	$Tlo,$Tlo,$t0
391	ldr	$t0,[sp,#`$Xoff+8*16`+0]
392	adc	$Thi,$Thi,$t1
393
394	ldr	$t1,[sp,#`$Xoff+8*16`+4]
395	adds	$Tlo,$Tlo,$t2
396	adc	$Thi,$Thi,$t3
397	adds	$Tlo,$Tlo,$t0
398	adc	$Thi,$Thi,$t1
399___
400	&BODY_00_15(0x17);
401$code.=<<___;
402#if __ARM_ARCH__>=7
403	ittt	eq			@ Thumb2 thing, sanity check in ARM
404#endif
405	ldreq	$t0,[sp,#`$Xoff+8*(16-1)`+0]
406	ldreq	$t1,[sp,#`$Xoff+8*(16-1)`+4]
407	beq	.L16_79
408	bic	$Ktbl,$Ktbl,#1
409
410	ldr	$Tlo,[sp,#$Boff+0]
411	ldr	$Thi,[sp,#$Boff+4]
412	ldr	$t0, [$ctx,#$Aoff+$lo]
413	ldr	$t1, [$ctx,#$Aoff+$hi]
414	ldr	$t2, [$ctx,#$Boff+$lo]
415	ldr	$t3, [$ctx,#$Boff+$hi]
416	adds	$t0,$Alo,$t0
417	str	$t0, [$ctx,#$Aoff+$lo]
418	adc	$t1,$Ahi,$t1
419	str	$t1, [$ctx,#$Aoff+$hi]
420	adds	$t2,$Tlo,$t2
421	str	$t2, [$ctx,#$Boff+$lo]
422	adc	$t3,$Thi,$t3
423	str	$t3, [$ctx,#$Boff+$hi]
424
425	ldr	$Alo,[sp,#$Coff+0]
426	ldr	$Ahi,[sp,#$Coff+4]
427	ldr	$Tlo,[sp,#$Doff+0]
428	ldr	$Thi,[sp,#$Doff+4]
429	ldr	$t0, [$ctx,#$Coff+$lo]
430	ldr	$t1, [$ctx,#$Coff+$hi]
431	ldr	$t2, [$ctx,#$Doff+$lo]
432	ldr	$t3, [$ctx,#$Doff+$hi]
433	adds	$t0,$Alo,$t0
434	str	$t0, [$ctx,#$Coff+$lo]
435	adc	$t1,$Ahi,$t1
436	str	$t1, [$ctx,#$Coff+$hi]
437	adds	$t2,$Tlo,$t2
438	str	$t2, [$ctx,#$Doff+$lo]
439	adc	$t3,$Thi,$t3
440	str	$t3, [$ctx,#$Doff+$hi]
441
442	ldr	$Tlo,[sp,#$Foff+0]
443	ldr	$Thi,[sp,#$Foff+4]
444	ldr	$t0, [$ctx,#$Eoff+$lo]
445	ldr	$t1, [$ctx,#$Eoff+$hi]
446	ldr	$t2, [$ctx,#$Foff+$lo]
447	ldr	$t3, [$ctx,#$Foff+$hi]
448	adds	$Elo,$Elo,$t0
449	str	$Elo,[$ctx,#$Eoff+$lo]
450	adc	$Ehi,$Ehi,$t1
451	str	$Ehi,[$ctx,#$Eoff+$hi]
452	adds	$t2,$Tlo,$t2
453	str	$t2, [$ctx,#$Foff+$lo]
454	adc	$t3,$Thi,$t3
455	str	$t3, [$ctx,#$Foff+$hi]
456
457	ldr	$Alo,[sp,#$Goff+0]
458	ldr	$Ahi,[sp,#$Goff+4]
459	ldr	$Tlo,[sp,#$Hoff+0]
460	ldr	$Thi,[sp,#$Hoff+4]
461	ldr	$t0, [$ctx,#$Goff+$lo]
462	ldr	$t1, [$ctx,#$Goff+$hi]
463	ldr	$t2, [$ctx,#$Hoff+$lo]
464	ldr	$t3, [$ctx,#$Hoff+$hi]
465	adds	$t0,$Alo,$t0
466	str	$t0, [$ctx,#$Goff+$lo]
467	adc	$t1,$Ahi,$t1
468	str	$t1, [$ctx,#$Goff+$hi]
469	adds	$t2,$Tlo,$t2
470	str	$t2, [$ctx,#$Hoff+$lo]
471	adc	$t3,$Thi,$t3
472	str	$t3, [$ctx,#$Hoff+$hi]
473
474	add	sp,sp,#640
475	sub	$Ktbl,$Ktbl,#640
476
477	teq	$inp,$len
478	bne	.Loop
479
480	add	sp,sp,#8*9		@ destroy frame
481#if __ARM_ARCH__>=5
482	ldmia	sp!,{r4-r12,pc}
483#else
484	ldmia	sp!,{r4-r12,lr}
485	tst	lr,#1
486	moveq	pc,lr			@ be binary compatible with V4, yet
487	bx	lr			@ interoperable with Thumb ISA:-)
488#endif
489.size	sha512_block_data_order,.-sha512_block_data_order
490___
491
492{
493my @Sigma0=(28,34,39);
494my @Sigma1=(14,18,41);
495my @sigma0=(1, 8, 7);
496my @sigma1=(19,61,6);
497
498my $Ktbl="r3";
499my $cnt="r12";	# volatile register known as ip, intra-procedure-call scratch
500
501my @X=map("d$_",(0..15));
502my @V=($A,$B,$C,$D,$E,$F,$G,$H)=map("d$_",(16..23));
503
504sub NEON_00_15() {
505my $i=shift;
506my ($a,$b,$c,$d,$e,$f,$g,$h)=@_;
507my ($t0,$t1,$t2,$T1,$K,$Ch,$Maj)=map("d$_",(24..31));	# temps
508
509$code.=<<___ if ($i<16 || $i&1);
510	vshr.u64	$t0,$e,#@Sigma1[0]	@ $i
511#if $i<16
512	vld1.64		{@X[$i%16]},[$inp]!	@ handles unaligned
513#endif
514	vshr.u64	$t1,$e,#@Sigma1[1]
515#if $i>0
516	 vadd.i64	$a,$Maj			@ h+=Maj from the past
517#endif
518	vshr.u64	$t2,$e,#@Sigma1[2]
519___
520$code.=<<___;
521	vld1.64		{$K},[$Ktbl,:64]!	@ K[i++]
522	vsli.64		$t0,$e,#`64-@Sigma1[0]`
523	vsli.64		$t1,$e,#`64-@Sigma1[1]`
524	vmov		$Ch,$e
525	vsli.64		$t2,$e,#`64-@Sigma1[2]`
526#if $i<16 && defined(__ARMEL__)
527	vrev64.8	@X[$i],@X[$i]
528#endif
529	veor		$t1,$t0
530	vbsl		$Ch,$f,$g		@ Ch(e,f,g)
531	vshr.u64	$t0,$a,#@Sigma0[0]
532	veor		$t2,$t1			@ Sigma1(e)
533	vadd.i64	$T1,$Ch,$h
534	vshr.u64	$t1,$a,#@Sigma0[1]
535	vsli.64		$t0,$a,#`64-@Sigma0[0]`
536	vadd.i64	$T1,$t2
537	vshr.u64	$t2,$a,#@Sigma0[2]
538	vadd.i64	$K,@X[$i%16]
539	vsli.64		$t1,$a,#`64-@Sigma0[1]`
540	veor		$Maj,$a,$b
541	vsli.64		$t2,$a,#`64-@Sigma0[2]`
542	veor		$h,$t0,$t1
543	vadd.i64	$T1,$K
544	vbsl		$Maj,$c,$b		@ Maj(a,b,c)
545	veor		$h,$t2			@ Sigma0(a)
546	vadd.i64	$d,$T1
547	vadd.i64	$Maj,$T1
548	@ vadd.i64	$h,$Maj
549___
550}
551
552sub NEON_16_79() {
553my $i=shift;
554
555if ($i&1)	{ &NEON_00_15($i,@_); return; }
556
557# 2x-vectorized, therefore runs every 2nd round
558my @X=map("q$_",(0..7));			# view @X as 128-bit vector
559my ($t0,$t1,$s0,$s1) = map("q$_",(12..15));	# temps
560my ($d0,$d1,$d2) = map("d$_",(24..26));		# temps from NEON_00_15
561my $e=@_[4];					# $e from NEON_00_15
562$i /= 2;
563$code.=<<___;
564	vshr.u64	$t0,@X[($i+7)%8],#@sigma1[0]
565	vshr.u64	$t1,@X[($i+7)%8],#@sigma1[1]
566	 vadd.i64	@_[0],d30			@ h+=Maj from the past
567	vshr.u64	$s1,@X[($i+7)%8],#@sigma1[2]
568	vsli.64		$t0,@X[($i+7)%8],#`64-@sigma1[0]`
569	vext.8		$s0,@X[$i%8],@X[($i+1)%8],#8	@ X[i+1]
570	vsli.64		$t1,@X[($i+7)%8],#`64-@sigma1[1]`
571	veor		$s1,$t0
572	vshr.u64	$t0,$s0,#@sigma0[0]
573	veor		$s1,$t1				@ sigma1(X[i+14])
574	vshr.u64	$t1,$s0,#@sigma0[1]
575	vadd.i64	@X[$i%8],$s1
576	vshr.u64	$s1,$s0,#@sigma0[2]
577	vsli.64		$t0,$s0,#`64-@sigma0[0]`
578	vsli.64		$t1,$s0,#`64-@sigma0[1]`
579	vext.8		$s0,@X[($i+4)%8],@X[($i+5)%8],#8	@ X[i+9]
580	veor		$s1,$t0
581	vshr.u64	$d0,$e,#@Sigma1[0]		@ from NEON_00_15
582	vadd.i64	@X[$i%8],$s0
583	vshr.u64	$d1,$e,#@Sigma1[1]		@ from NEON_00_15
584	veor		$s1,$t1				@ sigma0(X[i+1])
585	vshr.u64	$d2,$e,#@Sigma1[2]		@ from NEON_00_15
586	vadd.i64	@X[$i%8],$s1
587___
588	&NEON_00_15(2*$i,@_);
589}
590
591$code.=<<___;
592#if __ARM_MAX_ARCH__>=7
593.arch	armv7-a
594.fpu	neon
595
596.global	sha512_block_data_order_neon
597.type	sha512_block_data_order_neon,%function
598.align	4
599sha512_block_data_order_neon:
600.LNEON:
601	dmb				@ errata #451034 on early Cortex A8
602	add	$len,$inp,$len,lsl#7	@ len to point at the end of inp
603	VFP_ABI_PUSH
604	adr	$Ktbl,.Lsha512_block_data_order
605	sub	$Ktbl,$Ktbl,.Lsha512_block_data_order-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