• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1#!/usr/bin/env perl
2
3# ====================================================================
4# Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
5# project. The module is, however, dual licensed under OpenSSL and
6# CRYPTOGAMS licenses depending on where you obtain it. For further
7# details see http://www.openssl.org/~appro/cryptogams/.
8#
9# Permission to use under GPL terms is granted.
10# ====================================================================
11
12# SHA512 block procedure for ARMv4. September 2007.
13
14# This code is ~4.5 (four and a half) times faster than code generated
15# by gcc 3.4 and it spends ~72 clock cycles per byte [on single-issue
16# Xscale PXA250 core].
17#
18# July 2010.
19#
20# Rescheduling for dual-issue pipeline resulted in 6% improvement on
21# Cortex A8 core and ~40 cycles per processed byte.
22
23# February 2011.
24#
25# Profiler-assisted and platform-specific optimization resulted in 7%
26# improvement on Coxtex A8 core and ~38 cycles per byte.
27
28# March 2011.
29#
30# Add NEON implementation. On Cortex A8 it was measured to process
31# one byte in 23.3 cycles or ~60% faster than integer-only code.
32
33# August 2012.
34#
35# Improve NEON performance by 12% on Snapdragon S4. In absolute
36# terms it's 22.6 cycles per byte, which is disappointing result.
37# Technical writers asserted that 3-way S4 pipeline can sustain
38# multiple NEON instructions per cycle, but dual NEON issue could
39# not be observed, see http://www.openssl.org/~appro/Snapdragon-S4.html
40# for further details. On side note Cortex-A15 processes one byte in
41# 16 cycles.
42
43# Byte order [in]dependence. =========================================
44#
45# Originally caller was expected to maintain specific *dword* order in
46# h[0-7], namely with most significant dword at *lower* address, which
47# was reflected in below two parameters as 0 and 4. Now caller is
48# expected to maintain native byte order for whole 64-bit values.
49$hi="HI";
50$lo="LO";
51# ====================================================================
52
53$flavour = shift;
54if ($flavour=~/^\w[\w\-]*\.\w+$/) { $output=$flavour; undef $flavour; }
55else { while (($output=shift) && ($output!~/^\w[\w\-]*\.\w+$/)) {} }
56
57if ($flavour && $flavour ne "void") {
58    $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
59    ( $xlate="${dir}arm-xlate.pl" and -f $xlate ) or
60    ( $xlate="${dir}../../perlasm/arm-xlate.pl" and -f $xlate) or
61    die "can't locate arm-xlate.pl";
62
63    open STDOUT,"| \"$^X\" $xlate $flavour $output";
64} else {
65    open STDOUT,">$output";
66}
67
68$ctx="r0";	# parameter block
69$inp="r1";
70$len="r2";
71
72$Tlo="r3";
73$Thi="r4";
74$Alo="r5";
75$Ahi="r6";
76$Elo="r7";
77$Ehi="r8";
78$t0="r9";
79$t1="r10";
80$t2="r11";
81$t3="r12";
82############	r13 is stack pointer
83$Ktbl="r14";
84############	r15 is program counter
85
86$Aoff=8*0;
87$Boff=8*1;
88$Coff=8*2;
89$Doff=8*3;
90$Eoff=8*4;
91$Foff=8*5;
92$Goff=8*6;
93$Hoff=8*7;
94$Xoff=8*8;
95
96sub BODY_00_15() {
97my $magic = shift;
98$code.=<<___;
99	@ Sigma1(x)	(ROTR((x),14) ^ ROTR((x),18)  ^ ROTR((x),41))
100	@ LO		lo>>14^hi<<18 ^ lo>>18^hi<<14 ^ hi>>9^lo<<23
101	@ HI		hi>>14^lo<<18 ^ hi>>18^lo<<14 ^ lo>>9^hi<<23
102	mov	$t0,$Elo,lsr#14
103	str	$Tlo,[sp,#$Xoff+0]
104	mov	$t1,$Ehi,lsr#14
105	str	$Thi,[sp,#$Xoff+4]
106	eor	$t0,$t0,$Ehi,lsl#18
107	ldr	$t2,[sp,#$Hoff+0]	@ h.lo
108	eor	$t1,$t1,$Elo,lsl#18
109	ldr	$t3,[sp,#$Hoff+4]	@ h.hi
110	eor	$t0,$t0,$Elo,lsr#18
111	eor	$t1,$t1,$Ehi,lsr#18
112	eor	$t0,$t0,$Ehi,lsl#14
113	eor	$t1,$t1,$Elo,lsl#14
114	eor	$t0,$t0,$Ehi,lsr#9
115	eor	$t1,$t1,$Elo,lsr#9
116	eor	$t0,$t0,$Elo,lsl#23
117	eor	$t1,$t1,$Ehi,lsl#23	@ Sigma1(e)
118	adds	$Tlo,$Tlo,$t0
119	ldr	$t0,[sp,#$Foff+0]	@ f.lo
120	adc	$Thi,$Thi,$t1		@ T += Sigma1(e)
121	ldr	$t1,[sp,#$Foff+4]	@ f.hi
122	adds	$Tlo,$Tlo,$t2
123	ldr	$t2,[sp,#$Goff+0]	@ g.lo
124	adc	$Thi,$Thi,$t3		@ T += h
125	ldr	$t3,[sp,#$Goff+4]	@ g.hi
126
127	eor	$t0,$t0,$t2
128	str	$Elo,[sp,#$Eoff+0]
129	eor	$t1,$t1,$t3
130	str	$Ehi,[sp,#$Eoff+4]
131	and	$t0,$t0,$Elo
132	str	$Alo,[sp,#$Aoff+0]
133	and	$t1,$t1,$Ehi
134	str	$Ahi,[sp,#$Aoff+4]
135	eor	$t0,$t0,$t2
136	ldr	$t2,[$Ktbl,#$lo]	@ K[i].lo
137	eor	$t1,$t1,$t3		@ Ch(e,f,g)
138	ldr	$t3,[$Ktbl,#$hi]	@ K[i].hi
139
140	adds	$Tlo,$Tlo,$t0
141	ldr	$Elo,[sp,#$Doff+0]	@ d.lo
142	adc	$Thi,$Thi,$t1		@ T += Ch(e,f,g)
143	ldr	$Ehi,[sp,#$Doff+4]	@ d.hi
144	adds	$Tlo,$Tlo,$t2
145	and	$t0,$t2,#0xff
146	adc	$Thi,$Thi,$t3		@ T += K[i]
147	adds	$Elo,$Elo,$Tlo
148	ldr	$t2,[sp,#$Boff+0]	@ b.lo
149	adc	$Ehi,$Ehi,$Thi		@ d += T
150	teq	$t0,#$magic
151
152	ldr	$t3,[sp,#$Coff+0]	@ c.lo
153#if __ARM_ARCH__>=7
154	it	eq			@ Thumb2 thing, sanity check in ARM
155#endif
156	orreq	$Ktbl,$Ktbl,#1
157	@ Sigma0(x)	(ROTR((x),28) ^ ROTR((x),34) ^ ROTR((x),39))
158	@ LO		lo>>28^hi<<4  ^ hi>>2^lo<<30 ^ hi>>7^lo<<25
159	@ HI		hi>>28^lo<<4  ^ lo>>2^hi<<30 ^ lo>>7^hi<<25
160	mov	$t0,$Alo,lsr#28
161	mov	$t1,$Ahi,lsr#28
162	eor	$t0,$t0,$Ahi,lsl#4
163	eor	$t1,$t1,$Alo,lsl#4
164	eor	$t0,$t0,$Ahi,lsr#2
165	eor	$t1,$t1,$Alo,lsr#2
166	eor	$t0,$t0,$Alo,lsl#30
167	eor	$t1,$t1,$Ahi,lsl#30
168	eor	$t0,$t0,$Ahi,lsr#7
169	eor	$t1,$t1,$Alo,lsr#7
170	eor	$t0,$t0,$Alo,lsl#25
171	eor	$t1,$t1,$Ahi,lsl#25	@ Sigma0(a)
172	adds	$Tlo,$Tlo,$t0
173	and	$t0,$Alo,$t2
174	adc	$Thi,$Thi,$t1		@ T += Sigma0(a)
175
176	ldr	$t1,[sp,#$Boff+4]	@ b.hi
177	orr	$Alo,$Alo,$t2
178	ldr	$t2,[sp,#$Coff+4]	@ c.hi
179	and	$Alo,$Alo,$t3
180	and	$t3,$Ahi,$t1
181	orr	$Ahi,$Ahi,$t1
182	orr	$Alo,$Alo,$t0		@ Maj(a,b,c).lo
183	and	$Ahi,$Ahi,$t2
184	adds	$Alo,$Alo,$Tlo
185	orr	$Ahi,$Ahi,$t3		@ Maj(a,b,c).hi
186	sub	sp,sp,#8
187	adc	$Ahi,$Ahi,$Thi		@ h += T
188	tst	$Ktbl,#1
189	add	$Ktbl,$Ktbl,#8
190___
191}
192$code=<<___;
193#ifndef __KERNEL__
194# include <openssl/arm_arch.h>
195# define VFP_ABI_PUSH	vstmdb	sp!,{d8-d15}
196# define VFP_ABI_POP	vldmia	sp!,{d8-d15}
197#else
198# define __ARM_ARCH__ __LINUX_ARM_ARCH__
199# define __ARM_MAX_ARCH__ 7
200# define VFP_ABI_PUSH
201# define VFP_ABI_POP
202#endif
203
204#ifdef __ARMEL__
205# define LO 0
206# define HI 4
207# define WORD64(hi0,lo0,hi1,lo1)	.word	lo0,hi0, lo1,hi1
208#else
209# define HI 0
210# define LO 4
211# define WORD64(hi0,lo0,hi1,lo1)	.word	hi0,lo0, hi1,lo1
212#endif
213
214.text
215#if __ARM_ARCH__<7 || defined(__APPLE__)
216.code	32
217#else
218.syntax unified
219# ifdef __thumb2__
220#  define adrl adr
221.thumb
222# else
223.code   32
224# endif
225#endif
226
227.type	K512,%object
228.align	5
229K512:
230WORD64(0x428a2f98,0xd728ae22, 0x71374491,0x23ef65cd)
231WORD64(0xb5c0fbcf,0xec4d3b2f, 0xe9b5dba5,0x8189dbbc)
232WORD64(0x3956c25b,0xf348b538, 0x59f111f1,0xb605d019)
233WORD64(0x923f82a4,0xaf194f9b, 0xab1c5ed5,0xda6d8118)
234WORD64(0xd807aa98,0xa3030242, 0x12835b01,0x45706fbe)
235WORD64(0x243185be,0x4ee4b28c, 0x550c7dc3,0xd5ffb4e2)
236WORD64(0x72be5d74,0xf27b896f, 0x80deb1fe,0x3b1696b1)
237WORD64(0x9bdc06a7,0x25c71235, 0xc19bf174,0xcf692694)
238WORD64(0xe49b69c1,0x9ef14ad2, 0xefbe4786,0x384f25e3)
239WORD64(0x0fc19dc6,0x8b8cd5b5, 0x240ca1cc,0x77ac9c65)
240WORD64(0x2de92c6f,0x592b0275, 0x4a7484aa,0x6ea6e483)
241WORD64(0x5cb0a9dc,0xbd41fbd4, 0x76f988da,0x831153b5)
242WORD64(0x983e5152,0xee66dfab, 0xa831c66d,0x2db43210)
243WORD64(0xb00327c8,0x98fb213f, 0xbf597fc7,0xbeef0ee4)
244WORD64(0xc6e00bf3,0x3da88fc2, 0xd5a79147,0x930aa725)
245WORD64(0x06ca6351,0xe003826f, 0x14292967,0x0a0e6e70)
246WORD64(0x27b70a85,0x46d22ffc, 0x2e1b2138,0x5c26c926)
247WORD64(0x4d2c6dfc,0x5ac42aed, 0x53380d13,0x9d95b3df)
248WORD64(0x650a7354,0x8baf63de, 0x766a0abb,0x3c77b2a8)
249WORD64(0x81c2c92e,0x47edaee6, 0x92722c85,0x1482353b)
250WORD64(0xa2bfe8a1,0x4cf10364, 0xa81a664b,0xbc423001)
251WORD64(0xc24b8b70,0xd0f89791, 0xc76c51a3,0x0654be30)
252WORD64(0xd192e819,0xd6ef5218, 0xd6990624,0x5565a910)
253WORD64(0xf40e3585,0x5771202a, 0x106aa070,0x32bbd1b8)
254WORD64(0x19a4c116,0xb8d2d0c8, 0x1e376c08,0x5141ab53)
255WORD64(0x2748774c,0xdf8eeb99, 0x34b0bcb5,0xe19b48a8)
256WORD64(0x391c0cb3,0xc5c95a63, 0x4ed8aa4a,0xe3418acb)
257WORD64(0x5b9cca4f,0x7763e373, 0x682e6ff3,0xd6b2b8a3)
258WORD64(0x748f82ee,0x5defb2fc, 0x78a5636f,0x43172f60)
259WORD64(0x84c87814,0xa1f0ab72, 0x8cc70208,0x1a6439ec)
260WORD64(0x90befffa,0x23631e28, 0xa4506ceb,0xde82bde9)
261WORD64(0xbef9a3f7,0xb2c67915, 0xc67178f2,0xe372532b)
262WORD64(0xca273ece,0xea26619c, 0xd186b8c7,0x21c0c207)
263WORD64(0xeada7dd6,0xcde0eb1e, 0xf57d4f7f,0xee6ed178)
264WORD64(0x06f067aa,0x72176fba, 0x0a637dc5,0xa2c898a6)
265WORD64(0x113f9804,0xbef90dae, 0x1b710b35,0x131c471b)
266WORD64(0x28db77f5,0x23047d84, 0x32caab7b,0x40c72493)
267WORD64(0x3c9ebe0a,0x15c9bebc, 0x431d67c4,0x9c100d4c)
268WORD64(0x4cc5d4be,0xcb3e42b6, 0x597f299c,0xfc657e2a)
269WORD64(0x5fcb6fab,0x3ad6faec, 0x6c44198c,0x4a475817)
270.size	K512,.-K512
271#if __ARM_MAX_ARCH__>=7 && !defined(__KERNEL__)
272.LOPENSSL_armcap:
273.word	OPENSSL_armcap_P-.Lsha512_block_data_order
274.skip	32-4
275#else
276.skip	32
277#endif
278
279.global	sha512_block_data_order
280.type	sha512_block_data_order,%function
281sha512_block_data_order:
282.Lsha512_block_data_order:
283#if __ARM_ARCH__<7
284	sub	r3,pc,#8		@ sha512_block_data_order
285#else
286	adr	r3,sha512_block_data_order
287#endif
288#if __ARM_MAX_ARCH__>=7 && !defined(__KERNEL__)
289	ldr	r12,.LOPENSSL_armcap
290	ldr	r12,[r3,r12]		@ OPENSSL_armcap_P
291#ifdef	__APPLE__
292	ldr	r12,[r12]
293#endif
294	tst	r12,#1
295	bne	.LNEON
296#endif
297	add	$len,$inp,$len,lsl#7	@ len to point at the end of inp
298	stmdb	sp!,{r4-r12,lr}
299	sub	$Ktbl,r3,#672		@ K512
300	sub	sp,sp,#9*8
301
302	ldr	$Elo,[$ctx,#$Eoff+$lo]
303	ldr	$Ehi,[$ctx,#$Eoff+$hi]
304	ldr	$t0, [$ctx,#$Goff+$lo]
305	ldr	$t1, [$ctx,#$Goff+$hi]
306	ldr	$t2, [$ctx,#$Hoff+$lo]
307	ldr	$t3, [$ctx,#$Hoff+$hi]
308.Loop:
309	str	$t0, [sp,#$Goff+0]
310	str	$t1, [sp,#$Goff+4]
311	str	$t2, [sp,#$Hoff+0]
312	str	$t3, [sp,#$Hoff+4]
313	ldr	$Alo,[$ctx,#$Aoff+$lo]
314	ldr	$Ahi,[$ctx,#$Aoff+$hi]
315	ldr	$Tlo,[$ctx,#$Boff+$lo]
316	ldr	$Thi,[$ctx,#$Boff+$hi]
317	ldr	$t0, [$ctx,#$Coff+$lo]
318	ldr	$t1, [$ctx,#$Coff+$hi]
319	ldr	$t2, [$ctx,#$Doff+$lo]
320	ldr	$t3, [$ctx,#$Doff+$hi]
321	str	$Tlo,[sp,#$Boff+0]
322	str	$Thi,[sp,#$Boff+4]
323	str	$t0, [sp,#$Coff+0]
324	str	$t1, [sp,#$Coff+4]
325	str	$t2, [sp,#$Doff+0]
326	str	$t3, [sp,#$Doff+4]
327	ldr	$Tlo,[$ctx,#$Foff+$lo]
328	ldr	$Thi,[$ctx,#$Foff+$hi]
329	str	$Tlo,[sp,#$Foff+0]
330	str	$Thi,[sp,#$Foff+4]
331
332.L00_15:
333#if __ARM_ARCH__<7
334	ldrb	$Tlo,[$inp,#7]
335	ldrb	$t0, [$inp,#6]
336	ldrb	$t1, [$inp,#5]
337	ldrb	$t2, [$inp,#4]
338	ldrb	$Thi,[$inp,#3]
339	ldrb	$t3, [$inp,#2]
340	orr	$Tlo,$Tlo,$t0,lsl#8
341	ldrb	$t0, [$inp,#1]
342	orr	$Tlo,$Tlo,$t1,lsl#16
343	ldrb	$t1, [$inp],#8
344	orr	$Tlo,$Tlo,$t2,lsl#24
345	orr	$Thi,$Thi,$t3,lsl#8
346	orr	$Thi,$Thi,$t0,lsl#16
347	orr	$Thi,$Thi,$t1,lsl#24
348#else
349	ldr	$Tlo,[$inp,#4]
350	ldr	$Thi,[$inp],#8
351#ifdef __ARMEL__
352	rev	$Tlo,$Tlo
353	rev	$Thi,$Thi
354#endif
355#endif
356___
357	&BODY_00_15(0x94);
358$code.=<<___;
359	tst	$Ktbl,#1
360	beq	.L00_15
361	ldr	$t0,[sp,#`$Xoff+8*(16-1)`+0]
362	ldr	$t1,[sp,#`$Xoff+8*(16-1)`+4]
363	bic	$Ktbl,$Ktbl,#1
364.L16_79:
365	@ sigma0(x)	(ROTR((x),1)  ^ ROTR((x),8)  ^ ((x)>>7))
366	@ LO		lo>>1^hi<<31  ^ lo>>8^hi<<24 ^ lo>>7^hi<<25
367	@ HI		hi>>1^lo<<31  ^ hi>>8^lo<<24 ^ hi>>7
368	mov	$Tlo,$t0,lsr#1
369	ldr	$t2,[sp,#`$Xoff+8*(16-14)`+0]
370	mov	$Thi,$t1,lsr#1
371	ldr	$t3,[sp,#`$Xoff+8*(16-14)`+4]
372	eor	$Tlo,$Tlo,$t1,lsl#31
373	eor	$Thi,$Thi,$t0,lsl#31
374	eor	$Tlo,$Tlo,$t0,lsr#8
375	eor	$Thi,$Thi,$t1,lsr#8
376	eor	$Tlo,$Tlo,$t1,lsl#24
377	eor	$Thi,$Thi,$t0,lsl#24
378	eor	$Tlo,$Tlo,$t0,lsr#7
379	eor	$Thi,$Thi,$t1,lsr#7
380	eor	$Tlo,$Tlo,$t1,lsl#25
381
382	@ sigma1(x)	(ROTR((x),19) ^ ROTR((x),61) ^ ((x)>>6))
383	@ LO		lo>>19^hi<<13 ^ hi>>29^lo<<3 ^ lo>>6^hi<<26
384	@ HI		hi>>19^lo<<13 ^ lo>>29^hi<<3 ^ hi>>6
385	mov	$t0,$t2,lsr#19
386	mov	$t1,$t3,lsr#19
387	eor	$t0,$t0,$t3,lsl#13
388	eor	$t1,$t1,$t2,lsl#13
389	eor	$t0,$t0,$t3,lsr#29
390	eor	$t1,$t1,$t2,lsr#29
391	eor	$t0,$t0,$t2,lsl#3
392	eor	$t1,$t1,$t3,lsl#3
393	eor	$t0,$t0,$t2,lsr#6
394	eor	$t1,$t1,$t3,lsr#6
395	ldr	$t2,[sp,#`$Xoff+8*(16-9)`+0]
396	eor	$t0,$t0,$t3,lsl#26
397
398	ldr	$t3,[sp,#`$Xoff+8*(16-9)`+4]
399	adds	$Tlo,$Tlo,$t0
400	ldr	$t0,[sp,#`$Xoff+8*16`+0]
401	adc	$Thi,$Thi,$t1
402
403	ldr	$t1,[sp,#`$Xoff+8*16`+4]
404	adds	$Tlo,$Tlo,$t2
405	adc	$Thi,$Thi,$t3
406	adds	$Tlo,$Tlo,$t0
407	adc	$Thi,$Thi,$t1
408___
409	&BODY_00_15(0x17);
410$code.=<<___;
411#if __ARM_ARCH__>=7
412	ittt	eq			@ Thumb2 thing, sanity check in ARM
413#endif
414	ldreq	$t0,[sp,#`$Xoff+8*(16-1)`+0]
415	ldreq	$t1,[sp,#`$Xoff+8*(16-1)`+4]
416	beq	.L16_79
417	bic	$Ktbl,$Ktbl,#1
418
419	ldr	$Tlo,[sp,#$Boff+0]
420	ldr	$Thi,[sp,#$Boff+4]
421	ldr	$t0, [$ctx,#$Aoff+$lo]
422	ldr	$t1, [$ctx,#$Aoff+$hi]
423	ldr	$t2, [$ctx,#$Boff+$lo]
424	ldr	$t3, [$ctx,#$Boff+$hi]
425	adds	$t0,$Alo,$t0
426	str	$t0, [$ctx,#$Aoff+$lo]
427	adc	$t1,$Ahi,$t1
428	str	$t1, [$ctx,#$Aoff+$hi]
429	adds	$t2,$Tlo,$t2
430	str	$t2, [$ctx,#$Boff+$lo]
431	adc	$t3,$Thi,$t3
432	str	$t3, [$ctx,#$Boff+$hi]
433
434	ldr	$Alo,[sp,#$Coff+0]
435	ldr	$Ahi,[sp,#$Coff+4]
436	ldr	$Tlo,[sp,#$Doff+0]
437	ldr	$Thi,[sp,#$Doff+4]
438	ldr	$t0, [$ctx,#$Coff+$lo]
439	ldr	$t1, [$ctx,#$Coff+$hi]
440	ldr	$t2, [$ctx,#$Doff+$lo]
441	ldr	$t3, [$ctx,#$Doff+$hi]
442	adds	$t0,$Alo,$t0
443	str	$t0, [$ctx,#$Coff+$lo]
444	adc	$t1,$Ahi,$t1
445	str	$t1, [$ctx,#$Coff+$hi]
446	adds	$t2,$Tlo,$t2
447	str	$t2, [$ctx,#$Doff+$lo]
448	adc	$t3,$Thi,$t3
449	str	$t3, [$ctx,#$Doff+$hi]
450
451	ldr	$Tlo,[sp,#$Foff+0]
452	ldr	$Thi,[sp,#$Foff+4]
453	ldr	$t0, [$ctx,#$Eoff+$lo]
454	ldr	$t1, [$ctx,#$Eoff+$hi]
455	ldr	$t2, [$ctx,#$Foff+$lo]
456	ldr	$t3, [$ctx,#$Foff+$hi]
457	adds	$Elo,$Elo,$t0
458	str	$Elo,[$ctx,#$Eoff+$lo]
459	adc	$Ehi,$Ehi,$t1
460	str	$Ehi,[$ctx,#$Eoff+$hi]
461	adds	$t2,$Tlo,$t2
462	str	$t2, [$ctx,#$Foff+$lo]
463	adc	$t3,$Thi,$t3
464	str	$t3, [$ctx,#$Foff+$hi]
465
466	ldr	$Alo,[sp,#$Goff+0]
467	ldr	$Ahi,[sp,#$Goff+4]
468	ldr	$Tlo,[sp,#$Hoff+0]
469	ldr	$Thi,[sp,#$Hoff+4]
470	ldr	$t0, [$ctx,#$Goff+$lo]
471	ldr	$t1, [$ctx,#$Goff+$hi]
472	ldr	$t2, [$ctx,#$Hoff+$lo]
473	ldr	$t3, [$ctx,#$Hoff+$hi]
474	adds	$t0,$Alo,$t0
475	str	$t0, [$ctx,#$Goff+$lo]
476	adc	$t1,$Ahi,$t1
477	str	$t1, [$ctx,#$Goff+$hi]
478	adds	$t2,$Tlo,$t2
479	str	$t2, [$ctx,#$Hoff+$lo]
480	adc	$t3,$Thi,$t3
481	str	$t3, [$ctx,#$Hoff+$hi]
482
483	add	sp,sp,#640
484	sub	$Ktbl,$Ktbl,#640
485
486	teq	$inp,$len
487	bne	.Loop
488
489	add	sp,sp,#8*9		@ destroy frame
490#if __ARM_ARCH__>=5
491	ldmia	sp!,{r4-r12,pc}
492#else
493	ldmia	sp!,{r4-r12,lr}
494	tst	lr,#1
495	moveq	pc,lr			@ be binary compatible with V4, yet
496	bx	lr			@ interoperable with Thumb ISA:-)
497#endif
498.size	sha512_block_data_order,.-sha512_block_data_order
499___
500
501{
502my @Sigma0=(28,34,39);
503my @Sigma1=(14,18,41);
504my @sigma0=(1, 8, 7);
505my @sigma1=(19,61,6);
506
507my $Ktbl="r3";
508my $cnt="r12";	# volatile register known as ip, intra-procedure-call scratch
509
510my @X=map("d$_",(0..15));
511my @V=($A,$B,$C,$D,$E,$F,$G,$H)=map("d$_",(16..23));
512
513sub NEON_00_15() {
514my $i=shift;
515my ($a,$b,$c,$d,$e,$f,$g,$h)=@_;
516my ($t0,$t1,$t2,$T1,$K,$Ch,$Maj)=map("d$_",(24..31));	# temps
517
518$code.=<<___ if ($i<16 || $i&1);
519	vshr.u64	$t0,$e,#@Sigma1[0]	@ $i
520#if $i<16
521	vld1.64		{@X[$i%16]},[$inp]!	@ handles unaligned
522#endif
523	vshr.u64	$t1,$e,#@Sigma1[1]
524#if $i>0
525	 vadd.i64	$a,$Maj			@ h+=Maj from the past
526#endif
527	vshr.u64	$t2,$e,#@Sigma1[2]
528___
529$code.=<<___;
530	vld1.64		{$K},[$Ktbl,:64]!	@ K[i++]
531	vsli.64		$t0,$e,#`64-@Sigma1[0]`
532	vsli.64		$t1,$e,#`64-@Sigma1[1]`
533	vmov		$Ch,$e
534	vsli.64		$t2,$e,#`64-@Sigma1[2]`
535#if $i<16 && defined(__ARMEL__)
536	vrev64.8	@X[$i],@X[$i]
537#endif
538	veor		$t1,$t0
539	vbsl		$Ch,$f,$g		@ Ch(e,f,g)
540	vshr.u64	$t0,$a,#@Sigma0[0]
541	veor		$t2,$t1			@ Sigma1(e)
542	vadd.i64	$T1,$Ch,$h
543	vshr.u64	$t1,$a,#@Sigma0[1]
544	vsli.64		$t0,$a,#`64-@Sigma0[0]`
545	vadd.i64	$T1,$t2
546	vshr.u64	$t2,$a,#@Sigma0[2]
547	vadd.i64	$K,@X[$i%16]
548	vsli.64		$t1,$a,#`64-@Sigma0[1]`
549	veor		$Maj,$a,$b
550	vsli.64		$t2,$a,#`64-@Sigma0[2]`
551	veor		$h,$t0,$t1
552	vadd.i64	$T1,$K
553	vbsl		$Maj,$c,$b		@ Maj(a,b,c)
554	veor		$h,$t2			@ Sigma0(a)
555	vadd.i64	$d,$T1
556	vadd.i64	$Maj,$T1
557	@ vadd.i64	$h,$Maj
558___
559}
560
561sub NEON_16_79() {
562my $i=shift;
563
564if ($i&1)	{ &NEON_00_15($i,@_); return; }
565
566# 2x-vectorized, therefore runs every 2nd round
567my @X=map("q$_",(0..7));			# view @X as 128-bit vector
568my ($t0,$t1,$s0,$s1) = map("q$_",(12..15));	# temps
569my ($d0,$d1,$d2) = map("d$_",(24..26));		# temps from NEON_00_15
570my $e=@_[4];					# $e from NEON_00_15
571$i /= 2;
572$code.=<<___;
573	vshr.u64	$t0,@X[($i+7)%8],#@sigma1[0]
574	vshr.u64	$t1,@X[($i+7)%8],#@sigma1[1]
575	 vadd.i64	@_[0],d30			@ h+=Maj from the past
576	vshr.u64	$s1,@X[($i+7)%8],#@sigma1[2]
577	vsli.64		$t0,@X[($i+7)%8],#`64-@sigma1[0]`
578	vext.8		$s0,@X[$i%8],@X[($i+1)%8],#8	@ X[i+1]
579	vsli.64		$t1,@X[($i+7)%8],#`64-@sigma1[1]`
580	veor		$s1,$t0
581	vshr.u64	$t0,$s0,#@sigma0[0]
582	veor		$s1,$t1				@ sigma1(X[i+14])
583	vshr.u64	$t1,$s0,#@sigma0[1]
584	vadd.i64	@X[$i%8],$s1
585	vshr.u64	$s1,$s0,#@sigma0[2]
586	vsli.64		$t0,$s0,#`64-@sigma0[0]`
587	vsli.64		$t1,$s0,#`64-@sigma0[1]`
588	vext.8		$s0,@X[($i+4)%8],@X[($i+5)%8],#8	@ X[i+9]
589	veor		$s1,$t0
590	vshr.u64	$d0,$e,#@Sigma1[0]		@ from NEON_00_15
591	vadd.i64	@X[$i%8],$s0
592	vshr.u64	$d1,$e,#@Sigma1[1]		@ from NEON_00_15
593	veor		$s1,$t1				@ sigma0(X[i+1])
594	vshr.u64	$d2,$e,#@Sigma1[2]		@ from NEON_00_15
595	vadd.i64	@X[$i%8],$s1
596___
597	&NEON_00_15(2*$i,@_);
598}
599
600$code.=<<___;
601#if __ARM_MAX_ARCH__>=7
602.arch	armv7-a
603.fpu	neon
604
605.global	sha512_block_data_order_neon
606.type	sha512_block_data_order_neon,%function
607.align	4
608sha512_block_data_order_neon:
609.LNEON:
610	dmb				@ errata #451034 on early Cortex A8
611	add	$len,$inp,$len,lsl#7	@ len to point at the end of inp
612	adr	$Ktbl,K512
613	VFP_ABI_PUSH
614	vldmia	$ctx,{$A-$H}		@ load context
615.Loop_neon:
616___
617for($i=0;$i<16;$i++)	{ &NEON_00_15($i,@V); unshift(@V,pop(@V)); }
618$code.=<<___;
619	mov		$cnt,#4
620.L16_79_neon:
621	subs		$cnt,#1
622___
623for(;$i<32;$i++)	{ &NEON_16_79($i,@V); unshift(@V,pop(@V)); }
624$code.=<<___;
625	bne		.L16_79_neon
626
627	 vadd.i64	$A,d30		@ h+=Maj from the past
628	vldmia		$ctx,{d24-d31}	@ load context to temp
629	vadd.i64	q8,q12		@ vectorized accumulate
630	vadd.i64	q9,q13
631	vadd.i64	q10,q14
632	vadd.i64	q11,q15
633	vstmia		$ctx,{$A-$H}	@ save context
634	teq		$inp,$len
635	sub		$Ktbl,#640	@ rewind K512
636	bne		.Loop_neon
637
638	VFP_ABI_POP
639	ret				@ bx lr
640.size	sha512_block_data_order_neon,.-sha512_block_data_order_neon
641#endif
642___
643}
644$code.=<<___;
645.asciz	"SHA512 block transform for ARMv4/NEON, CRYPTOGAMS by <appro\@openssl.org>"
646.align	2
647#if __ARM_MAX_ARCH__>=7 && !defined(__KERNEL__)
648.comm	OPENSSL_armcap_P,4,4
649.hidden	OPENSSL_armcap_P
650#endif
651___
652
653$code =~ s/\`([^\`]*)\`/eval $1/gem;
654$code =~ s/\bbx\s+lr\b/.word\t0xe12fff1e/gm;	# make it possible to compile with -march=armv4
655$code =~ s/\bret\b/bx	lr/gm;
656
657open SELF,$0;
658while(<SELF>) {
659	next if (/^#!/);
660	last if (!s/^#/@/ and !/^$/);
661	print;
662}
663close SELF;
664
665print $code;
666close STDOUT; # enforce flush
667