summaryrefslogtreecommitdiffstats
path: root/src/crypto/bn/asm/armv4-mont.pl
blob: 4206fd8503bdf2e0587bcc7589bcfb593b6a9d46 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
#!/usr/bin/env perl

# ====================================================================
# Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
# project. The module is, however, dual licensed under OpenSSL and
# CRYPTOGAMS licenses depending on where you obtain it. For further
# details see http://www.openssl.org/~appro/cryptogams/.
# ====================================================================

# January 2007.

# Montgomery multiplication for ARMv4.
#
# Performance improvement naturally varies among CPU implementations
# and compilers. The code was observed to provide +65-35% improvement
# [depending on key length, less for longer keys] on ARM920T, and
# +115-80% on Intel IXP425. This is compared to pre-bn_mul_mont code
# base and compiler generated code with in-lined umull and even umlal
# instructions. The latter means that this code didn't really have an 
# "advantage" of utilizing some "secret" instruction.
#
# The code is interoperable with Thumb ISA and is rather compact, less
# than 1/2KB. Windows CE port would be trivial, as it's exclusively
# about decorations, ABI and instruction syntax are identical.

# November 2013
#
# Add NEON code path, which handles lengths divisible by 8. RSA/DSA
# performance improvement on Cortex-A8 is ~45-100% depending on key
# length, more for longer keys. On Cortex-A15 the span is ~10-105%.
# On Snapdragon S4 improvement was measured to vary from ~70% to
# incredible ~380%, yes, 4.8x faster, for RSA4096 sign. But this is
# rather because original integer-only code seems to perform
# suboptimally on S4. Situation on Cortex-A9 is unfortunately
# different. It's being looked into, but the trouble is that
# performance for vectors longer than 256 bits is actually couple
# of percent worse than for integer-only code. The code is chosen
# for execution on all NEON-capable processors, because gain on
# others outweighs the marginal loss on Cortex-A9.

$flavour = shift;
if ($flavour=~/^\w[\w\-]*\.\w+$/) { $output=$flavour; undef $flavour; }
else { while (($output=shift) && ($output!~/^\w[\w\-]*\.\w+$/)) {} }

if ($flavour && $flavour ne "void") {
    $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
    ( $xlate="${dir}arm-xlate.pl" and -f $xlate ) or
    ( $xlate="${dir}../../perlasm/arm-xlate.pl" and -f $xlate) or
    die "can't locate arm-xlate.pl";

    open STDOUT,"| \"$^X\" $xlate $flavour $output";
} else {
    open STDOUT,">$output";
}

$num="r0";	# starts as num argument, but holds &tp[num-1]
$ap="r1";
$bp="r2"; $bi="r2"; $rp="r2";
$np="r3";
$tp="r4";
$aj="r5";
$nj="r6";
$tj="r7";
$n0="r8";
###########	# r9 is reserved by ELF as platform specific, e.g. TLS pointer
$alo="r10";	# sl, gcc uses it to keep @GOT
$ahi="r11";	# fp
$nlo="r12";	# ip
###########	# r13 is stack pointer
$nhi="r14";	# lr
###########	# r15 is program counter

#### argument block layout relative to &tp[num-1], a.k.a. $num
$_rp="$num,#12*4";
# ap permanently resides in r1
$_bp="$num,#13*4";
# np permanently resides in r3
$_n0="$num,#14*4";
$_num="$num,#15*4";	$_bpend=$_num;

$code=<<___;
#include <openssl/arm_arch.h>

.text
.code	32

#if __ARM_MAX_ARCH__>=7
.align	5
.LOPENSSL_armcap:
.word	OPENSSL_armcap_P-.Lbn_mul_mont
#endif

.global	bn_mul_mont
.hidden	bn_mul_mont
.type	bn_mul_mont,%function

.align	5
bn_mul_mont:
.Lbn_mul_mont:
	ldr	ip,[sp,#4]		@ load num
	stmdb	sp!,{r0,r2}		@ sp points at argument block
#if __ARM_MAX_ARCH__>=7
	tst	ip,#7
	bne	.Lialu
	adr	r0,bn_mul_mont
	ldr	r2,.LOPENSSL_armcap
	ldr	r0,[r0,r2]
#ifdef	__APPLE__
	ldr	r0,[r0]
#endif
	tst	r0,#1			@ NEON available?
	ldmia	sp, {r0,r2}
	beq	.Lialu
	add	sp,sp,#8
	b	bn_mul8x_mont_neon
.align	4
.Lialu:
#endif
	cmp	ip,#2
	mov	$num,ip			@ load num
	movlt	r0,#0
	addlt	sp,sp,#2*4
	blt	.Labrt

	stmdb	sp!,{r4-r12,lr}		@ save 10 registers

	mov	$num,$num,lsl#2		@ rescale $num for byte count
	sub	sp,sp,$num		@ alloca(4*num)
	sub	sp,sp,#4		@ +extra dword
	sub	$num,$num,#4		@ "num=num-1"
	add	$tp,$bp,$num		@ &bp[num-1]

	add	$num,sp,$num		@ $num to point at &tp[num-1]
	ldr	$n0,[$_n0]		@ &n0
	ldr	$bi,[$bp]		@ bp[0]
	ldr	$aj,[$ap],#4		@ ap[0],ap++
	ldr	$nj,[$np],#4		@ np[0],np++
	ldr	$n0,[$n0]		@ *n0
	str	$tp,[$_bpend]		@ save &bp[num]

	umull	$alo,$ahi,$aj,$bi	@ ap[0]*bp[0]
	str	$n0,[$_n0]		@ save n0 value
	mul	$n0,$alo,$n0		@ "tp[0]"*n0
	mov	$nlo,#0
	umlal	$alo,$nlo,$nj,$n0	@ np[0]*n0+"t[0]"
	mov	$tp,sp

.L1st:
	ldr	$aj,[$ap],#4		@ ap[j],ap++
	mov	$alo,$ahi
	ldr	$nj,[$np],#4		@ np[j],np++
	mov	$ahi,#0
	umlal	$alo,$ahi,$aj,$bi	@ ap[j]*bp[0]
	mov	$nhi,#0
	umlal	$nlo,$nhi,$nj,$n0	@ np[j]*n0
	adds	$nlo,$nlo,$alo
	str	$nlo,[$tp],#4		@ tp[j-1]=,tp++
	adc	$nlo,$nhi,#0
	cmp	$tp,$num
	bne	.L1st

	adds	$nlo,$nlo,$ahi
	ldr	$tp,[$_bp]		@ restore bp
	mov	$nhi,#0
	ldr	$n0,[$_n0]		@ restore n0
	adc	$nhi,$nhi,#0
	str	$nlo,[$num]		@ tp[num-1]=
	str	$nhi,[$num,#4]		@ tp[num]=

.Louter:
	sub	$tj,$num,sp		@ "original" $num-1 value
	sub	$ap,$ap,$tj		@ "rewind" ap to &ap[1]
	ldr	$bi,[$tp,#4]!		@ *(++bp)
	sub	$np,$np,$tj		@ "rewind" np to &np[1]
	ldr	$aj,[$ap,#-4]		@ ap[0]
	ldr	$alo,[sp]		@ tp[0]
	ldr	$nj,[$np,#-4]		@ np[0]
	ldr	$tj,[sp,#4]		@ tp[1]

	mov	$ahi,#0
	umlal	$alo,$ahi,$aj,$bi	@ ap[0]*bp[i]+tp[0]
	str	$tp,[$_bp]		@ save bp
	mul	$n0,$alo,$n0
	mov	$nlo,#0
	umlal	$alo,$nlo,$nj,$n0	@ np[0]*n0+"tp[0]"
	mov	$tp,sp

.Linner:
	ldr	$aj,[$ap],#4		@ ap[j],ap++
	adds	$alo,$ahi,$tj		@ +=tp[j]
	ldr	$nj,[$np],#4		@ np[j],np++
	mov	$ahi,#0
	umlal	$alo,$ahi,$aj,$bi	@ ap[j]*bp[i]
	mov	$nhi,#0
	umlal	$nlo,$nhi,$nj,$n0	@ np[j]*n0
	adc	$ahi,$ahi,#0
	ldr	$tj,[$tp,#8]		@ tp[j+1]
	adds	$nlo,$nlo,$alo
	str	$nlo,[$tp],#4		@ tp[j-1]=,tp++
	adc	$nlo,$nhi,#0
	cmp	$tp,$num
	bne	.Linner

	adds	$nlo,$nlo,$ahi
	mov	$nhi,#0
	ldr	$tp,[$_bp]		@ restore bp
	adc	$nhi,$nhi,#0
	ldr	$n0,[$_n0]		@ restore n0
	adds	$nlo,$nlo,$tj
	ldr	$tj,[$_bpend]		@ restore &bp[num]
	adc	$nhi,$nhi,#0
	str	$nlo,[$num]		@ tp[num-1]=
	str	$nhi,[$num,#4]		@ tp[num]=

	cmp	$tp,$tj
	bne	.Louter

	ldr	$rp,[$_rp]		@ pull rp
	add	$num,$num,#4		@ $num to point at &tp[num]
	sub	$aj,$num,sp		@ "original" num value
	mov	$tp,sp			@ "rewind" $tp
	mov	$ap,$tp			@ "borrow" $ap
	sub	$np,$np,$aj		@ "rewind" $np to &np[0]

	subs	$tj,$tj,$tj		@ "clear" carry flag
.Lsub:	ldr	$tj,[$tp],#4
	ldr	$nj,[$np],#4
	sbcs	$tj,$tj,$nj		@ tp[j]-np[j]
	str	$tj,[$rp],#4		@ rp[j]=
	teq	$tp,$num		@ preserve carry
	bne	.Lsub
	sbcs	$nhi,$nhi,#0		@ upmost carry
	mov	$tp,sp			@ "rewind" $tp
	sub	$rp,$rp,$aj		@ "rewind" $rp

	and	$ap,$tp,$nhi
	bic	$np,$rp,$nhi
	orr	$ap,$ap,$np		@ ap=borrow?tp:rp

.Lcopy:	ldr	$tj,[$ap],#4		@ copy or in-place refresh
	str	sp,[$tp],#4		@ zap tp
	str	$tj,[$rp],#4
	cmp	$tp,$num
	bne	.Lcopy

	add	sp,$num,#4		@ skip over tp[num+1]
	ldmia	sp!,{r4-r12,lr}		@ restore registers
	add	sp,sp,#2*4		@ skip over {r0,r2}
	mov	r0,#1
.Labrt:
#if __ARM_ARCH__>=5
	ret				@ bx lr
#else
	tst	lr,#1
	moveq	pc,lr			@ be binary compatible with V4, yet
	bx	lr			@ interoperable with Thumb ISA:-)
#endif
.size	bn_mul_mont,.-bn_mul_mont
___
{
sub Dlo()   { shift=~m|q([1]?[0-9])|?"d".($1*2):"";     }
sub Dhi()   { shift=~m|q([1]?[0-9])|?"d".($1*2+1):"";   }

my ($A0,$A1,$A2,$A3)=map("d$_",(0..3));
my ($N0,$N1,$N2,$N3)=map("d$_",(4..7));
my ($Z,$Temp)=("q4","q5");
my ($A0xB,$A1xB,$A2xB,$A3xB,$A4xB,$A5xB,$A6xB,$A7xB)=map("q$_",(6..13));
my ($Bi,$Ni,$M0)=map("d$_",(28..31));
my $zero=&Dlo($Z);
my $temp=&Dlo($Temp);

my ($rptr,$aptr,$bptr,$nptr,$n0,$num)=map("r$_",(0..5));
my ($tinptr,$toutptr,$inner,$outer)=map("r$_",(6..9));

$code.=<<___;
#if __ARM_MAX_ARCH__>=7
.arch	armv7-a
.fpu	neon

.type	bn_mul8x_mont_neon,%function
.align	5
bn_mul8x_mont_neon:
	mov	ip,sp
	stmdb	sp!,{r4-r11}
	vstmdb	sp!,{d8-d15}		@ ABI specification says so
	ldmia	ip,{r4-r5}		@ load rest of parameter block

	sub		$toutptr,sp,#16
	vld1.32		{${Bi}[0]}, [$bptr,:32]!
	sub		$toutptr,$toutptr,$num,lsl#4
	vld1.32		{$A0-$A3},  [$aptr]!		@ can't specify :32 :-(
	and		$toutptr,$toutptr,#-64
	vld1.32		{${M0}[0]}, [$n0,:32]
	mov		sp,$toutptr			@ alloca
	veor		$zero,$zero,$zero
	subs		$inner,$num,#8
	vzip.16		$Bi,$zero

	vmull.u32	$A0xB,$Bi,${A0}[0]
	vmull.u32	$A1xB,$Bi,${A0}[1]
	vmull.u32	$A2xB,$Bi,${A1}[0]
	vshl.i64	$temp,`&Dhi("$A0xB")`,#16
	vmull.u32	$A3xB,$Bi,${A1}[1]

	vadd.u64	$temp,$temp,`&Dlo("$A0xB")`
	veor		$zero,$zero,$zero
	vmul.u32	$Ni,$temp,$M0

	vmull.u32	$A4xB,$Bi,${A2}[0]
	 vld1.32	{$N0-$N3}, [$nptr]!
	vmull.u32	$A5xB,$Bi,${A2}[1]
	vmull.u32	$A6xB,$Bi,${A3}[0]
	vzip.16		$Ni,$zero
	vmull.u32	$A7xB,$Bi,${A3}[1]

	bne	.LNEON_1st

	@ special case for num=8, everything is in register bank...

	vmlal.u32	$A0xB,$Ni,${N0}[0]
	sub		$outer,$num,#1
	vmlal.u32	$A1xB,$Ni,${N0}[1]
	vmlal.u32	$A2xB,$Ni,${N1}[0]
	vmlal.u32	$A3xB,$Ni,${N1}[1]

	vmlal.u32	$A4xB,$Ni,${N2}[0]
	vmov		$Temp,$A0xB
	vmlal.u32	$A5xB,$Ni,${N2}[1]
	vmov		$A0xB,$A1xB
	vmlal.u32	$A6xB,$Ni,${N3}[0]
	vmov		$A1xB,$A2xB
	vmlal.u32	$A7xB,$Ni,${N3}[1]
	vmov		$A2xB,$A3xB
	vmov		$A3xB,$A4xB
	vshr.u64	$temp,$temp,#16
	vmov		$A4xB,$A5xB
	vmov		$A5xB,$A6xB
	vadd.u64	$temp,$temp,`&Dhi("$Temp")`
	vmov		$A6xB,$A7xB
	veor		$A7xB,$A7xB
	vshr.u64	$temp,$temp,#16

	b	.LNEON_outer8

.align	4
.LNEON_outer8:
	vld1.32		{${Bi}[0]}, [$bptr,:32]!
	veor		$zero,$zero,$zero
	vzip.16		$Bi,$zero
	vadd.u64	`&Dlo("$A0xB")`,`&Dlo("$A0xB")`,$temp

	vmlal.u32	$A0xB,$Bi,${A0}[0]
	vmlal.u32	$A1xB,$Bi,${A0}[1]
	vmlal.u32	$A2xB,$Bi,${A1}[0]
	vshl.i64	$temp,`&Dhi("$A0xB")`,#16
	vmlal.u32	$A3xB,$Bi,${A1}[1]

	vadd.u64	$temp,$temp,`&Dlo("$A0xB")`
	veor		$zero,$zero,$zero
	subs		$outer,$outer,#1
	vmul.u32	$Ni,$temp,$M0

	vmlal.u32	$A4xB,$Bi,${A2}[0]
	vmlal.u32	$A5xB,$Bi,${A2}[1]
	vmlal.u32	$A6xB,$Bi,${A3}[0]
	vzip.16		$Ni,$zero
	vmlal.u32	$A7xB,$Bi,${A3}[1]

	vmlal.u32	$A0xB,$Ni,${N0}[0]
	vmlal.u32	$A1xB,$Ni,${N0}[1]
	vmlal.u32	$A2xB,$Ni,${N1}[0]
	vmlal.u32	$A3xB,$Ni,${N1}[1]

	vmlal.u32	$A4xB,$Ni,${N2}[0]
	vmov		$Temp,$A0xB
	vmlal.u32	$A5xB,$Ni,${N2}[1]
	vmov		$A0xB,$A1xB
	vmlal.u32	$A6xB,$Ni,${N3}[0]
	vmov		$A1xB,$A2xB
	vmlal.u32	$A7xB,$Ni,${N3}[1]
	vmov		$A2xB,$A3xB
	vmov		$A3xB,$A4xB
	vshr.u64	$temp,$temp,#16
	vmov		$A4xB,$A5xB
	vmov		$A5xB,$A6xB
	vadd.u64	$temp,$temp,`&Dhi("$Temp")`
	vmov		$A6xB,$A7xB
	veor		$A7xB,$A7xB
	vshr.u64	$temp,$temp,#16

	bne	.LNEON_outer8

	vadd.u64	`&Dlo("$A0xB")`,`&Dlo("$A0xB")`,$temp
	mov		$toutptr,sp
	vshr.u64	$temp,`&Dlo("$A0xB")`,#16
	mov		$inner,$num
	vadd.u64	`&Dhi("$A0xB")`,`&Dhi("$A0xB")`,$temp
	add		$tinptr,sp,#16
	vshr.u64	$temp,`&Dhi("$A0xB")`,#16
	vzip.16		`&Dlo("$A0xB")`,`&Dhi("$A0xB")`

	b	.LNEON_tail2

.align	4
.LNEON_1st:
	vmlal.u32	$A0xB,$Ni,${N0}[0]
	 vld1.32	{$A0-$A3}, [$aptr]!
	vmlal.u32	$A1xB,$Ni,${N0}[1]
	subs		$inner,$inner,#8
	vmlal.u32	$A2xB,$Ni,${N1}[0]
	vmlal.u32	$A3xB,$Ni,${N1}[1]

	vmlal.u32	$A4xB,$Ni,${N2}[0]
	 vld1.32	{$N0-$N1}, [$nptr]!
	vmlal.u32	$A5xB,$Ni,${N2}[1]
	 vst1.64	{$A0xB-$A1xB}, [$toutptr,:256]!
	vmlal.u32	$A6xB,$Ni,${N3}[0]
	vmlal.u32	$A7xB,$Ni,${N3}[1]
	 vst1.64	{$A2xB-$A3xB}, [$toutptr,:256]!

	vmull.u32	$A0xB,$Bi,${A0}[0]
	 vld1.32	{$N2-$N3}, [$nptr]!
	vmull.u32	$A1xB,$Bi,${A0}[1]
	 vst1.64	{$A4xB-$A5xB}, [$toutptr,:256]!
	vmull.u32	$A2xB,$Bi,${A1}[0]
	vmull.u32	$A3xB,$Bi,${A1}[1]
	 vst1.64	{$A6xB-$A7xB}, [$toutptr,:256]!

	vmull.u32	$A4xB,$Bi,${A2}[0]
	vmull.u32	$A5xB,$Bi,${A2}[1]
	vmull.u32	$A6xB,$Bi,${A3}[0]
	vmull.u32	$A7xB,$Bi,${A3}[1]

	bne	.LNEON_1st

	vmlal.u32	$A0xB,$Ni,${N0}[0]
	add		$tinptr,sp,#16
	vmlal.u32	$A1xB,$Ni,${N0}[1]
	sub		$aptr,$aptr,$num,lsl#2		@ rewind $aptr
	vmlal.u32	$A2xB,$Ni,${N1}[0]
	 vld1.64	{$Temp}, [sp,:128]
	vmlal.u32	$A3xB,$Ni,${N1}[1]
	sub		$outer,$num,#1

	vmlal.u32	$A4xB,$Ni,${N2}[0]
	vst1.64		{$A0xB-$A1xB}, [$toutptr,:256]!
	vmlal.u32	$A5xB,$Ni,${N2}[1]
	vshr.u64	$temp,$temp,#16
	 vld1.64	{$A0xB},       [$tinptr, :128]!
	vmlal.u32	$A6xB,$Ni,${N3}[0]
	vst1.64		{$A2xB-$A3xB}, [$toutptr,:256]!
	vmlal.u32	$A7xB,$Ni,${N3}[1]

	vst1.64		{$A4xB-$A5xB}, [$toutptr,:256]!
	vadd.u64	$temp,$temp,`&Dhi("$Temp")`
	veor		$Z,$Z,$Z
	vst1.64		{$A6xB-$A7xB}, [$toutptr,:256]!
	 vld1.64	{$A1xB-$A2xB}, [$tinptr, :256]!
	vst1.64		{$Z},          [$toutptr,:128]
	vshr.u64	$temp,$temp,#16

	b		.LNEON_outer

.align	4
.LNEON_outer:
	vld1.32		{${Bi}[0]}, [$bptr,:32]!
	sub		$nptr,$nptr,$num,lsl#2		@ rewind $nptr
	vld1.32		{$A0-$A3},  [$aptr]!
	veor		$zero,$zero,$zero
	mov		$toutptr,sp
	vzip.16		$Bi,$zero
	sub		$inner,$num,#8
	vadd.u64	`&Dlo("$A0xB")`,`&Dlo("$A0xB")`,$temp

	vmlal.u32	$A0xB,$Bi,${A0}[0]
	 vld1.64	{$A3xB-$A4xB},[$tinptr,:256]!
	vmlal.u32	$A1xB,$Bi,${A0}[1]
	vmlal.u32	$A2xB,$Bi,${A1}[0]
	 vld1.64	{$A5xB-$A6xB},[$tinptr,:256]!
	vmlal.u32	$A3xB,$Bi,${A1}[1]

	vshl.i64	$temp,`&Dhi("$A0xB")`,#16
	veor		$zero,$zero,$zero
	vadd.u64	$temp,$temp,`&Dlo("$A0xB")`
	 vld1.64	{$A7xB},[$tinptr,:128]!
	vmul.u32	$Ni,$temp,$M0

	vmlal.u32	$A4xB,$Bi,${A2}[0]
	 vld1.32	{$N0-$N3}, [$nptr]!
	vmlal.u32	$A5xB,$Bi,${A2}[1]
	vmlal.u32	$A6xB,$Bi,${A3}[0]
	vzip.16		$Ni,$zero
	vmlal.u32	$A7xB,$Bi,${A3}[1]

.LNEON_inner:
	vmlal.u32	$A0xB,$Ni,${N0}[0]
	 vld1.32	{$A0-$A3}, [$aptr]!
	vmlal.u32	$A1xB,$Ni,${N0}[1]
	 subs		$inner,$inner,#8
	vmlal.u32	$A2xB,$Ni,${N1}[0]
	vmlal.u32	$A3xB,$Ni,${N1}[1]
	vst1.64		{$A0xB-$A1xB}, [$toutptr,:256]!

	vmlal.u32	$A4xB,$Ni,${N2}[0]
	 vld1.64	{$A0xB},       [$tinptr, :128]!
	vmlal.u32	$A5xB,$Ni,${N2}[1]
	vst1.64		{$A2xB-$A3xB}, [$toutptr,:256]!
	vmlal.u32	$A6xB,$Ni,${N3}[0]
	 vld1.64	{$A1xB-$A2xB}, [$tinptr, :256]!
	vmlal.u32	$A7xB,$Ni,${N3}[1]
	vst1.64		{$A4xB-$A5xB}, [$toutptr,:256]!

	vmlal.u32	$A0xB,$Bi,${A0}[0]
	 vld1.64	{$A3xB-$A4xB}, [$tinptr, :256]!
	vmlal.u32	$A1xB,$Bi,${A0}[1]
	vst1.64		{$A6xB-$A7xB}, [$toutptr,:256]!
	vmlal.u32	$A2xB,$Bi,${A1}[0]
	 vld1.64	{$A5xB-$A6xB}, [$tinptr, :256]!
	vmlal.u32	$A3xB,$Bi,${A1}[1]
	 vld1.32	{$N0-$N3}, [$nptr]!

	vmlal.u32	$A4xB,$Bi,${A2}[0]
	 vld1.64	{$A7xB},       [$tinptr, :128]!
	vmlal.u32	$A5xB,$Bi,${A2}[1]
	vmlal.u32	$A6xB,$Bi,${A3}[0]
	vmlal.u32	$A7xB,$Bi,${A3}[1]

	bne	.LNEON_inner

	vmlal.u32	$A0xB,$Ni,${N0}[0]
	add		$tinptr,sp,#16
	vmlal.u32	$A1xB,$Ni,${N0}[1]
	sub		$aptr,$aptr,$num,lsl#2		@ rewind $aptr
	vmlal.u32	$A2xB,$Ni,${N1}[0]
	 vld1.64	{$Temp}, [sp,:128]
	vmlal.u32	$A3xB,$Ni,${N1}[1]
	subs		$outer,$outer,#1

	vmlal.u32	$A4xB,$Ni,${N2}[0]
	vst1.64		{$A0xB-$A1xB}, [$toutptr,:256]!
	vmlal.u32	$A5xB,$Ni,${N2}[1]
	 vld1.64	{$A0xB},       [$tinptr, :128]!
	vshr.u64	$temp,$temp,#16
	vst1.64		{$A2xB-$A3xB}, [$toutptr,:256]!
	vmlal.u32	$A6xB,$Ni,${N3}[0]
	 vld1.64	{$A1xB-$A2xB}, [$tinptr, :256]!
	vmlal.u32	$A7xB,$Ni,${N3}[1]

	vst1.64		{$A4xB-$A5xB}, [$toutptr,:256]!
	vadd.u64	$temp,$temp,`&Dhi("$Temp")`
	vst1.64		{$A6xB-$A7xB}, [$toutptr,:256]!
	vshr.u64	$temp,$temp,#16

	bne	.LNEON_outer

	mov		$toutptr,sp
	mov		$inner,$num

.LNEON_tail:
	vadd.u64	`&Dlo("$A0xB")`,`&Dlo("$A0xB")`,$temp
	vld1.64		{$A3xB-$A4xB}, [$tinptr, :256]!
	vshr.u64	$temp,`&Dlo("$A0xB")`,#16
	vadd.u64	`&Dhi("$A0xB")`,`&Dhi("$A0xB")`,$temp
	vld1.64		{$A5xB-$A6xB}, [$tinptr, :256]!
	vshr.u64	$temp,`&Dhi("$A0xB")`,#16
	vld1.64		{$A7xB},       [$tinptr, :128]!
	vzip.16		`&Dlo("$A0xB")`,`&Dhi("$A0xB")`

.LNEON_tail2:
	vadd.u64	`&Dlo("$A1xB")`,`&Dlo("$A1xB")`,$temp
	vst1.32		{`&Dlo("$A0xB")`[0]}, [$toutptr, :32]!
	vshr.u64	$temp,`&Dlo("$A1xB")`,#16
	vadd.u64	`&Dhi("$A1xB")`,`&Dhi("$A1xB")`,$temp
	vshr.u64	$temp,`&Dhi("$A1xB")`,#16
	vzip.16		`&Dlo("$A1xB")`,`&Dhi("$A1xB")`

	vadd.u64	`&Dlo("$A2xB")`,`&Dlo("$A2xB")`,$temp
	vst1.32		{`&Dlo("$A1xB")`[0]}, [$toutptr, :32]!
	vshr.u64	$temp,`&Dlo("$A2xB")`,#16
	vadd.u64	`&Dhi("$A2xB")`,`&Dhi("$A2xB")`,$temp
	vshr.u64	$temp,`&Dhi("$A2xB")`,#16
	vzip.16		`&Dlo("$A2xB")`,`&Dhi("$A2xB")`

	vadd.u64	`&Dlo("$A3xB")`,`&Dlo("$A3xB")`,$temp
	vst1.32		{`&Dlo("$A2xB")`[0]}, [$toutptr, :32]!
	vshr.u64	$temp,`&Dlo("$A3xB")`,#16
	vadd.u64	`&Dhi("$A3xB")`,`&Dhi("$A3xB")`,$temp
	vshr.u64	$temp,`&Dhi("$A3xB")`,#16
	vzip.16		`&Dlo("$A3xB")`,`&Dhi("$A3xB")`

	vadd.u64	`&Dlo("$A4xB")`,`&Dlo("$A4xB")`,$temp
	vst1.32		{`&Dlo("$A3xB")`[0]}, [$toutptr, :32]!
	vshr.u64	$temp,`&Dlo("$A4xB")`,#16
	vadd.u64	`&Dhi("$A4xB")`,`&Dhi("$A4xB")`,$temp
	vshr.u64	$temp,`&Dhi("$A4xB")`,#16
	vzip.16		`&Dlo("$A4xB")`,`&Dhi("$A4xB")`

	vadd.u64	`&Dlo("$A5xB")`,`&Dlo("$A5xB")`,$temp
	vst1.32		{`&Dlo("$A4xB")`[0]}, [$toutptr, :32]!
	vshr.u64	$temp,`&Dlo("$A5xB")`,#16
	vadd.u64	`&Dhi("$A5xB")`,`&Dhi("$A5xB")`,$temp
	vshr.u64	$temp,`&Dhi("$A5xB")`,#16
	vzip.16		`&Dlo("$A5xB")`,`&Dhi("$A5xB")`

	vadd.u64	`&Dlo("$A6xB")`,`&Dlo("$A6xB")`,$temp
	vst1.32		{`&Dlo("$A5xB")`[0]}, [$toutptr, :32]!
	vshr.u64	$temp,`&Dlo("$A6xB")`,#16
	vadd.u64	`&Dhi("$A6xB")`,`&Dhi("$A6xB")`,$temp
	vld1.64		{$A0xB}, [$tinptr, :128]!
	vshr.u64	$temp,`&Dhi("$A6xB")`,#16
	vzip.16		`&Dlo("$A6xB")`,`&Dhi("$A6xB")`

	vadd.u64	`&Dlo("$A7xB")`,`&Dlo("$A7xB")`,$temp
	vst1.32		{`&Dlo("$A6xB")`[0]}, [$toutptr, :32]!
	vshr.u64	$temp,`&Dlo("$A7xB")`,#16
	vadd.u64	`&Dhi("$A7xB")`,`&Dhi("$A7xB")`,$temp
	vld1.64		{$A1xB-$A2xB},	[$tinptr, :256]!
	vshr.u64	$temp,`&Dhi("$A7xB")`,#16
	vzip.16		`&Dlo("$A7xB")`,`&Dhi("$A7xB")`
	subs		$inner,$inner,#8
	vst1.32		{`&Dlo("$A7xB")`[0]}, [$toutptr, :32]!

	bne	.LNEON_tail

	vst1.32	{${temp}[0]}, [$toutptr, :32]		@ top-most bit
	sub	$nptr,$nptr,$num,lsl#2			@ rewind $nptr
	subs	$aptr,sp,#0				@ clear carry flag
	add	$bptr,sp,$num,lsl#2

.LNEON_sub:
	ldmia	$aptr!, {r4-r7}
	ldmia	$nptr!, {r8-r11}
	sbcs	r8, r4,r8
	sbcs	r9, r5,r9
	sbcs	r10,r6,r10
	sbcs	r11,r7,r11
	teq	$aptr,$bptr				@ preserves carry
	stmia	$rptr!, {r8-r11}
	bne	.LNEON_sub

	ldr	r10, [$aptr]				@ load top-most bit
	veor	q0,q0,q0
	sub	r11,$bptr,sp				@ this is num*4
	veor	q1,q1,q1
	mov	$aptr,sp
	sub	$rptr,$rptr,r11				@ rewind $rptr
	mov	$nptr,$bptr				@ second 3/4th of frame
	sbcs	r10,r10,#0				@ result is carry flag

.LNEON_copy_n_zap:
	ldmia	$aptr!, {r4-r7}
	ldmia	$rptr,  {r8-r11}
	movcc	r8, r4
	vst1.64	{q0-q1}, [$nptr,:256]!			@ wipe
	movcc	r9, r5
	movcc	r10,r6
	vst1.64	{q0-q1}, [$nptr,:256]!			@ wipe
	movcc	r11,r7
	ldmia	$aptr, {r4-r7}
	stmia	$rptr!, {r8-r11}
	sub	$aptr,$aptr,#16
	ldmia	$rptr, {r8-r11}
	movcc	r8, r4
	vst1.64	{q0-q1}, [$aptr,:256]!			@ wipe
	movcc	r9, r5
	movcc	r10,r6
	vst1.64	{q0-q1}, [$nptr,:256]!			@ wipe
	movcc	r11,r7
	teq	$aptr,$bptr				@ preserves carry
	stmia	$rptr!, {r8-r11}
	bne	.LNEON_copy_n_zap

	sub	sp,ip,#96
        vldmia  sp!,{d8-d15}
        ldmia   sp!,{r4-r11}
	ret						@ bx lr
.size	bn_mul8x_mont_neon,.-bn_mul8x_mont_neon
#endif
___
}
$code.=<<___;
.asciz	"Montgomery multiplication for ARMv4/NEON, CRYPTOGAMS by <appro\@openssl.org>"
.align	2
#if __ARM_MAX_ARCH__>=7
.comm	OPENSSL_armcap_P,4,4
.hidden	OPENSSL_armcap_P
#endif
___

$code =~ s/\`([^\`]*)\`/eval $1/gem;
$code =~ s/\bbx\s+lr\b/.word\t0xe12fff1e/gm;	# make it possible to compile with -march=armv4
$code =~ s/\bret\b/bx	lr/gm;
print $code;
close STDOUT;