1#! /usr/bin/env perl
2# Copyright 2016-2021 The OpenSSL Project Authors. All Rights Reserved.
3#
4# Licensed under the OpenSSL license (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#
17#			IALU(*)/gcc-4.4		NEON
18#
19# ARM11xx(ARMv6)	7.78/+100%		-
20# Cortex-A5		6.35/+130%		3.00
21# Cortex-A8		6.25/+115%		2.36
22# Cortex-A9		5.10/+95%		2.55
23# Cortex-A15		3.85/+85%		1.25(**)
24# Snapdragon S4		5.70/+100%		1.48(**)
25#
26# (*)	this is for -march=armv6, i.e. with bunch of ldrb loading data;
27# (**)	these are trade-off results, they can be improved by ~8% but at
28#	the cost of 15/12% regression on Cortex-A5/A7, it's even possible
29#	to improve Cortex-A9 result, but then A5/A7 loose more than 20%;
30
31$flavour = shift;
32if ($flavour=~/\w[\w\-]*\.\w+$/) { $output=$flavour; undef $flavour; }
33else { while (($output=shift) && ($output!~/\w[\w\-]*\.\w+$/)) {} }
34
35if ($flavour && $flavour ne "void") {
36    $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
37    ( $xlate="${dir}arm-xlate.pl" and -f $xlate ) or
38    ( $xlate="${dir}../../perlasm/arm-xlate.pl" and -f $xlate) or
39    die "can't locate arm-xlate.pl";
40
41    open STDOUT,"| \"$^X\" $xlate $flavour $output";
42} else {
43    open STDOUT,">$output";
44}
45
46($ctx,$inp,$len,$padbit)=map("r$_",(0..3));
47
48$code.=<<___;
49#include "arm_arch.h"
50
51.text
52#if defined(__thumb2__)
53.syntax	unified
54.thumb
55#else
56.code	32
57#endif
58
59.globl	poly1305_emit
60.globl	poly1305_blocks
61.globl	poly1305_init
62.type	poly1305_init,%function
63.align	5
64poly1305_init:
65.Lpoly1305_init:
66	stmdb	sp!,{r4-r11}
67
68	eor	r3,r3,r3
69	cmp	$inp,#0
70	str	r3,[$ctx,#0]		@ zero hash value
71	str	r3,[$ctx,#4]
72	str	r3,[$ctx,#8]
73	str	r3,[$ctx,#12]
74	str	r3,[$ctx,#16]
75	str	r3,[$ctx,#36]		@ is_base2_26
76	add	$ctx,$ctx,#20
77
78#ifdef	__thumb2__
79	it	eq
80#endif
81	moveq	r0,#0
82	beq	.Lno_key
83
84#if	__ARM_MAX_ARCH__>=7
85	adr	r11,.Lpoly1305_init
86	ldr	r12,.LOPENSSL_armcap
87#endif
88	ldrb	r4,[$inp,#0]
89	mov	r10,#0x0fffffff
90	ldrb	r5,[$inp,#1]
91	and	r3,r10,#-4		@ 0x0ffffffc
92	ldrb	r6,[$inp,#2]
93	ldrb	r7,[$inp,#3]
94	orr	r4,r4,r5,lsl#8
95	ldrb	r5,[$inp,#4]
96	orr	r4,r4,r6,lsl#16
97	ldrb	r6,[$inp,#5]
98	orr	r4,r4,r7,lsl#24
99	ldrb	r7,[$inp,#6]
100	and	r4,r4,r10
101
102#if	__ARM_MAX_ARCH__>=7
103	ldr	r12,[r11,r12]		@ OPENSSL_armcap_P
104# ifdef	__APPLE__
105	ldr	r12,[r12]
106# endif
107#endif
108	ldrb	r8,[$inp,#7]
109	orr	r5,r5,r6,lsl#8
110	ldrb	r6,[$inp,#8]
111	orr	r5,r5,r7,lsl#16
112	ldrb	r7,[$inp,#9]
113	orr	r5,r5,r8,lsl#24
114	ldrb	r8,[$inp,#10]
115	and	r5,r5,r3
116
117#if	__ARM_MAX_ARCH__>=7
118	tst	r12,#ARMV7_NEON		@ check for NEON
119# ifdef	__APPLE__
120	adr	r9,poly1305_blocks_neon
121	adr	r11,poly1305_blocks
122#  ifdef __thumb2__
123	it	ne
124#  endif
125	movne	r11,r9
126	adr	r12,poly1305_emit
127	adr	r10,poly1305_emit_neon
128#  ifdef __thumb2__
129	it	ne
130#  endif
131	movne	r12,r10
132# else
133#  ifdef __thumb2__
134	itete	eq
135#  endif
136	addeq	r12,r11,#(.Lpoly1305_emit-.Lpoly1305_init)
137	addne	r12,r11,#(.Lpoly1305_emit_neon-.Lpoly1305_init)
138	addeq	r11,r11,#(.Lpoly1305_blocks-.Lpoly1305_init)
139	addne	r11,r11,#(.Lpoly1305_blocks_neon-.Lpoly1305_init)
140# endif
141# ifdef	__thumb2__
142	orr	r12,r12,#1	@ thumb-ify address
143	orr	r11,r11,#1
144# endif
145#endif
146	ldrb	r9,[$inp,#11]
147	orr	r6,r6,r7,lsl#8
148	ldrb	r7,[$inp,#12]
149	orr	r6,r6,r8,lsl#16
150	ldrb	r8,[$inp,#13]
151	orr	r6,r6,r9,lsl#24
152	ldrb	r9,[$inp,#14]
153	and	r6,r6,r3
154
155	ldrb	r10,[$inp,#15]
156	orr	r7,r7,r8,lsl#8
157	str	r4,[$ctx,#0]
158	orr	r7,r7,r9,lsl#16
159	str	r5,[$ctx,#4]
160	orr	r7,r7,r10,lsl#24
161	str	r6,[$ctx,#8]
162	and	r7,r7,r3
163	str	r7,[$ctx,#12]
164#if	__ARM_MAX_ARCH__>=7
165	stmia	r2,{r11,r12}		@ fill functions table
166	mov	r0,#1
167#else
168	mov	r0,#0
169#endif
170.Lno_key:
171	ldmia	sp!,{r4-r11}
172#if	__ARM_ARCH__>=5
173	ret				@ bx	lr
174#else
175	tst	lr,#1
176	moveq	pc,lr			@ be binary compatible with V4, yet
177	bx	lr			@ interoperable with Thumb ISA:-)
178#endif
179.size	poly1305_init,.-poly1305_init
180___
181{
182my ($h0,$h1,$h2,$h3,$h4,$r0,$r1,$r2,$r3)=map("r$_",(4..12));
183my ($s1,$s2,$s3)=($r1,$r2,$r3);
184
185$code.=<<___;
186.type	poly1305_blocks,%function
187.align	5
188poly1305_blocks:
189.Lpoly1305_blocks:
190	stmdb	sp!,{r3-r11,lr}
191
192	ands	$len,$len,#-16
193	beq	.Lno_data
194
195	cmp	$padbit,#0
196	add	$len,$len,$inp		@ end pointer
197	sub	sp,sp,#32
198
199	ldmia	$ctx,{$h0-$r3}		@ load context
200
201	str	$ctx,[sp,#12]		@ offload stuff
202	mov	lr,$inp
203	str	$len,[sp,#16]
204	str	$r1,[sp,#20]
205	str	$r2,[sp,#24]
206	str	$r3,[sp,#28]
207	b	.Loop
208
209.Loop:
210#if __ARM_ARCH__<7
211	ldrb	r0,[lr],#16		@ load input
212# ifdef	__thumb2__
213	it	hi
214# endif
215	addhi	$h4,$h4,#1		@ 1<<128
216	ldrb	r1,[lr,#-15]
217	ldrb	r2,[lr,#-14]
218	ldrb	r3,[lr,#-13]
219	orr	r1,r0,r1,lsl#8
220	ldrb	r0,[lr,#-12]
221	orr	r2,r1,r2,lsl#16
222	ldrb	r1,[lr,#-11]
223	orr	r3,r2,r3,lsl#24
224	ldrb	r2,[lr,#-10]
225	adds	$h0,$h0,r3		@ accumulate input
226
227	ldrb	r3,[lr,#-9]
228	orr	r1,r0,r1,lsl#8
229	ldrb	r0,[lr,#-8]
230	orr	r2,r1,r2,lsl#16
231	ldrb	r1,[lr,#-7]
232	orr	r3,r2,r3,lsl#24
233	ldrb	r2,[lr,#-6]
234	adcs	$h1,$h1,r3
235
236	ldrb	r3,[lr,#-5]
237	orr	r1,r0,r1,lsl#8
238	ldrb	r0,[lr,#-4]
239	orr	r2,r1,r2,lsl#16
240	ldrb	r1,[lr,#-3]
241	orr	r3,r2,r3,lsl#24
242	ldrb	r2,[lr,#-2]
243	adcs	$h2,$h2,r3
244
245	ldrb	r3,[lr,#-1]
246	orr	r1,r0,r1,lsl#8
247	str	lr,[sp,#8]		@ offload input pointer
248	orr	r2,r1,r2,lsl#16
249	add	$s1,$r1,$r1,lsr#2
250	orr	r3,r2,r3,lsl#24
251#else
252	ldr	r0,[lr],#16		@ load input
253# ifdef	__thumb2__
254	it	hi
255# endif
256	addhi	$h4,$h4,#1		@ padbit
257	ldr	r1,[lr,#-12]
258	ldr	r2,[lr,#-8]
259	ldr	r3,[lr,#-4]
260# ifdef	__ARMEB__
261	rev	r0,r0
262	rev	r1,r1
263	rev	r2,r2
264	rev	r3,r3
265# endif
266	adds	$h0,$h0,r0		@ accumulate input
267	str	lr,[sp,#8]		@ offload input pointer
268	adcs	$h1,$h1,r1
269	add	$s1,$r1,$r1,lsr#2
270	adcs	$h2,$h2,r2
271#endif
272	add	$s2,$r2,$r2,lsr#2
273	adcs	$h3,$h3,r3
274	add	$s3,$r3,$r3,lsr#2
275
276	umull	r2,r3,$h1,$r0
277	 adc	$h4,$h4,#0
278	umull	r0,r1,$h0,$r0
279	umlal	r2,r3,$h4,$s1
280	umlal	r0,r1,$h3,$s1
281	ldr	$r1,[sp,#20]		@ reload $r1
282	umlal	r2,r3,$h2,$s3
283	umlal	r0,r1,$h1,$s3
284	umlal	r2,r3,$h3,$s2
285	umlal	r0,r1,$h2,$s2
286	umlal	r2,r3,$h0,$r1
287	str	r0,[sp,#0]		@ future $h0
288	 mul	r0,$s2,$h4
289	ldr	$r2,[sp,#24]		@ reload $r2
290	adds	r2,r2,r1		@ d1+=d0>>32
291	 eor	r1,r1,r1
292	adc	lr,r3,#0		@ future $h2
293	str	r2,[sp,#4]		@ future $h1
294
295	mul	r2,$s3,$h4
296	eor	r3,r3,r3
297	umlal	r0,r1,$h3,$s3
298	ldr	$r3,[sp,#28]		@ reload $r3
299	umlal	r2,r3,$h3,$r0
300	umlal	r0,r1,$h2,$r0
301	umlal	r2,r3,$h2,$r1
302	umlal	r0,r1,$h1,$r1
303	umlal	r2,r3,$h1,$r2
304	umlal	r0,r1,$h0,$r2
305	umlal	r2,r3,$h0,$r3
306	ldr	$h0,[sp,#0]
307	mul	$h4,$r0,$h4
308	ldr	$h1,[sp,#4]
309
310	adds	$h2,lr,r0		@ d2+=d1>>32
311	ldr	lr,[sp,#8]		@ reload input pointer
312	adc	r1,r1,#0
313	adds	$h3,r2,r1		@ d3+=d2>>32
314	ldr	r0,[sp,#16]		@ reload end pointer
315	adc	r3,r3,#0
316	add	$h4,$h4,r3		@ h4+=d3>>32
317
318	and	r1,$h4,#-4
319	and	$h4,$h4,#3
320	add	r1,r1,r1,lsr#2		@ *=5
321	adds	$h0,$h0,r1
322	adcs	$h1,$h1,#0
323	adcs	$h2,$h2,#0
324	adcs	$h3,$h3,#0
325	adc	$h4,$h4,#0
326
327	cmp	r0,lr			@ done yet?
328	bhi	.Loop
329
330	ldr	$ctx,[sp,#12]
331	add	sp,sp,#32
332	stmia	$ctx,{$h0-$h4}		@ store the result
333
334.Lno_data:
335#if	__ARM_ARCH__>=5
336	ldmia	sp!,{r3-r11,pc}
337#else
338	ldmia	sp!,{r3-r11,lr}
339	tst	lr,#1
340	moveq	pc,lr			@ be binary compatible with V4, yet
341	bx	lr			@ interoperable with Thumb ISA:-)
342#endif
343.size	poly1305_blocks,.-poly1305_blocks
344___
345}
346{
347my ($ctx,$mac,$nonce)=map("r$_",(0..2));
348my ($h0,$h1,$h2,$h3,$h4,$g0,$g1,$g2,$g3)=map("r$_",(3..11));
349my $g4=$h4;
350
351$code.=<<___;
352.type	poly1305_emit,%function
353.align	5
354poly1305_emit:
355.Lpoly1305_emit:
356	stmdb	sp!,{r4-r11}
357.Lpoly1305_emit_enter:
358
359	ldmia	$ctx,{$h0-$h4}
360	adds	$g0,$h0,#5		@ compare to modulus
361	adcs	$g1,$h1,#0
362	adcs	$g2,$h2,#0
363	adcs	$g3,$h3,#0
364	adc	$g4,$h4,#0
365	tst	$g4,#4			@ did it carry/borrow?
366
367#ifdef	__thumb2__
368	it	ne
369#endif
370	movne	$h0,$g0
371	ldr	$g0,[$nonce,#0]
372#ifdef	__thumb2__
373	it	ne
374#endif
375	movne	$h1,$g1
376	ldr	$g1,[$nonce,#4]
377#ifdef	__thumb2__
378	it	ne
379#endif
380	movne	$h2,$g2
381	ldr	$g2,[$nonce,#8]
382#ifdef	__thumb2__
383	it	ne
384#endif
385	movne	$h3,$g3
386	ldr	$g3,[$nonce,#12]
387
388	adds	$h0,$h0,$g0
389	adcs	$h1,$h1,$g1
390	adcs	$h2,$h2,$g2
391	adc	$h3,$h3,$g3
392
393#if __ARM_ARCH__>=7
394# ifdef __ARMEB__
395	rev	$h0,$h0
396	rev	$h1,$h1
397	rev	$h2,$h2
398	rev	$h3,$h3
399# endif
400	str	$h0,[$mac,#0]
401	str	$h1,[$mac,#4]
402	str	$h2,[$mac,#8]
403	str	$h3,[$mac,#12]
404#else
405	strb	$h0,[$mac,#0]
406	mov	$h0,$h0,lsr#8
407	strb	$h1,[$mac,#4]
408	mov	$h1,$h1,lsr#8
409	strb	$h2,[$mac,#8]
410	mov	$h2,$h2,lsr#8
411	strb	$h3,[$mac,#12]
412	mov	$h3,$h3,lsr#8
413
414	strb	$h0,[$mac,#1]
415	mov	$h0,$h0,lsr#8
416	strb	$h1,[$mac,#5]
417	mov	$h1,$h1,lsr#8
418	strb	$h2,[$mac,#9]
419	mov	$h2,$h2,lsr#8
420	strb	$h3,[$mac,#13]
421	mov	$h3,$h3,lsr#8
422
423	strb	$h0,[$mac,#2]
424	mov	$h0,$h0,lsr#8
425	strb	$h1,[$mac,#6]
426	mov	$h1,$h1,lsr#8
427	strb	$h2,[$mac,#10]
428	mov	$h2,$h2,lsr#8
429	strb	$h3,[$mac,#14]
430	mov	$h3,$h3,lsr#8
431
432	strb	$h0,[$mac,#3]
433	strb	$h1,[$mac,#7]
434	strb	$h2,[$mac,#11]
435	strb	$h3,[$mac,#15]
436#endif
437	ldmia	sp!,{r4-r11}
438#if	__ARM_ARCH__>=5
439	ret				@ bx	lr
440#else
441	tst	lr,#1
442	moveq	pc,lr			@ be binary compatible with V4, yet
443	bx	lr			@ interoperable with Thumb ISA:-)
444#endif
445.size	poly1305_emit,.-poly1305_emit
446___
447{
448my ($R0,$R1,$S1,$R2,$S2,$R3,$S3,$R4,$S4) = map("d$_",(0..9));
449my ($D0,$D1,$D2,$D3,$D4, $H0,$H1,$H2,$H3,$H4) = map("q$_",(5..14));
450my ($T0,$T1,$MASK) = map("q$_",(15,4,0));
451
452my ($in2,$zeros,$tbl0,$tbl1) = map("r$_",(4..7));
453
454$code.=<<___;
455#if	__ARM_MAX_ARCH__>=7
456.fpu	neon
457
458.type	poly1305_init_neon,%function
459.align	5
460poly1305_init_neon:
461	ldr	r4,[$ctx,#20]		@ load key base 2^32
462	ldr	r5,[$ctx,#24]
463	ldr	r6,[$ctx,#28]
464	ldr	r7,[$ctx,#32]
465
466	and	r2,r4,#0x03ffffff	@ base 2^32 -> base 2^26
467	mov	r3,r4,lsr#26
468	mov	r4,r5,lsr#20
469	orr	r3,r3,r5,lsl#6
470	mov	r5,r6,lsr#14
471	orr	r4,r4,r6,lsl#12
472	mov	r6,r7,lsr#8
473	orr	r5,r5,r7,lsl#18
474	and	r3,r3,#0x03ffffff
475	and	r4,r4,#0x03ffffff
476	and	r5,r5,#0x03ffffff
477
478	vdup.32	$R0,r2			@ r^1 in both lanes
479	add	r2,r3,r3,lsl#2		@ *5
480	vdup.32	$R1,r3
481	add	r3,r4,r4,lsl#2
482	vdup.32	$S1,r2
483	vdup.32	$R2,r4
484	add	r4,r5,r5,lsl#2
485	vdup.32	$S2,r3
486	vdup.32	$R3,r5
487	add	r5,r6,r6,lsl#2
488	vdup.32	$S3,r4
489	vdup.32	$R4,r6
490	vdup.32	$S4,r5
491
492	mov	$zeros,#2		@ counter
493
494.Lsquare_neon:
495	@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@
496	@ d0 = h0*r0 + h4*5*r1 + h3*5*r2 + h2*5*r3 + h1*5*r4
497	@ d1 = h1*r0 + h0*r1   + h4*5*r2 + h3*5*r3 + h2*5*r4
498	@ d2 = h2*r0 + h1*r1   + h0*r2   + h4*5*r3 + h3*5*r4
499	@ d3 = h3*r0 + h2*r1   + h1*r2   + h0*r3   + h4*5*r4
500	@ d4 = h4*r0 + h3*r1   + h2*r2   + h1*r3   + h0*r4
501
502	vmull.u32	$D0,$R0,${R0}[1]
503	vmull.u32	$D1,$R1,${R0}[1]
504	vmull.u32	$D2,$R2,${R0}[1]
505	vmull.u32	$D3,$R3,${R0}[1]
506	vmull.u32	$D4,$R4,${R0}[1]
507
508	vmlal.u32	$D0,$R4,${S1}[1]
509	vmlal.u32	$D1,$R0,${R1}[1]
510	vmlal.u32	$D2,$R1,${R1}[1]
511	vmlal.u32	$D3,$R2,${R1}[1]
512	vmlal.u32	$D4,$R3,${R1}[1]
513
514	vmlal.u32	$D0,$R3,${S2}[1]
515	vmlal.u32	$D1,$R4,${S2}[1]
516	vmlal.u32	$D3,$R1,${R2}[1]
517	vmlal.u32	$D2,$R0,${R2}[1]
518	vmlal.u32	$D4,$R2,${R2}[1]
519
520	vmlal.u32	$D0,$R2,${S3}[1]
521	vmlal.u32	$D3,$R0,${R3}[1]
522	vmlal.u32	$D1,$R3,${S3}[1]
523	vmlal.u32	$D2,$R4,${S3}[1]
524	vmlal.u32	$D4,$R1,${R3}[1]
525
526	vmlal.u32	$D3,$R4,${S4}[1]
527	vmlal.u32	$D0,$R1,${S4}[1]
528	vmlal.u32	$D1,$R2,${S4}[1]
529	vmlal.u32	$D2,$R3,${S4}[1]
530	vmlal.u32	$D4,$R0,${R4}[1]
531
532	@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@
533	@ lazy reduction as discussed in "NEON crypto" by D.J. Bernstein
534	@ and P. Schwabe
535	@
536	@ H0>>+H1>>+H2>>+H3>>+H4
537	@ H3>>+H4>>*5+H0>>+H1
538	@
539	@ Trivia.
540	@
541	@ Result of multiplication of n-bit number by m-bit number is
542	@ n+m bits wide. However! Even though 2^n is a n+1-bit number,
543	@ m-bit number multiplied by 2^n is still n+m bits wide.
544	@
545	@ Sum of two n-bit numbers is n+1 bits wide, sum of three - n+2,
546	@ and so is sum of four. Sum of 2^m n-m-bit numbers and n-bit
547	@ one is n+1 bits wide.
548	@
549	@ >>+ denotes Hnext += Hn>>26, Hn &= 0x3ffffff. This means that
550	@ H0, H2, H3 are guaranteed to be 26 bits wide, while H1 and H4
551	@ can be 27. However! In cases when their width exceeds 26 bits
552	@ they are limited by 2^26+2^6. This in turn means that *sum*
553	@ of the products with these values can still be viewed as sum
554	@ of 52-bit numbers as long as the amount of addends is not a
555	@ power of 2. For example,
556	@
557	@ H4 = H4*R0 + H3*R1 + H2*R2 + H1*R3 + H0 * R4,
558	@
559	@ which can't be larger than 5 * (2^26 + 2^6) * (2^26 + 2^6), or
560	@ 5 * (2^52 + 2*2^32 + 2^12), which in turn is smaller than
561	@ 8 * (2^52) or 2^55. However, the value is then multiplied by
562	@ by 5, so we should be looking at 5 * 5 * (2^52 + 2^33 + 2^12),
563	@ which is less than 32 * (2^52) or 2^57. And when processing
564	@ data we are looking at triple as many addends...
565	@
566	@ In key setup procedure pre-reduced H0 is limited by 5*4+1 and
567	@ 5*H4 - by 5*5 52-bit addends, or 57 bits. But when hashing the
568	@ input H0 is limited by (5*4+1)*3 addends, or 58 bits, while
569	@ 5*H4 by 5*5*3, or 59[!] bits. How is this relevant? vmlal.u32
570	@ instruction accepts 2x32-bit input and writes 2x64-bit result.
571	@ This means that result of reduction have to be compressed upon
572	@ loop wrap-around. This can be done in the process of reduction
573	@ to minimize amount of instructions [as well as amount of
574	@ 128-bit instructions, which benefits low-end processors], but
575	@ one has to watch for H2 (which is narrower than H0) and 5*H4
576	@ not being wider than 58 bits, so that result of right shift
577	@ by 26 bits fits in 32 bits. This is also useful on x86,
578	@ because it allows to use paddd in place for paddq, which
579	@ benefits Atom, where paddq is ridiculously slow.
580
581	vshr.u64	$T0,$D3,#26
582	vmovn.i64	$D3#lo,$D3
583	 vshr.u64	$T1,$D0,#26
584	 vmovn.i64	$D0#lo,$D0
585	vadd.i64	$D4,$D4,$T0		@ h3 -> h4
586	vbic.i32	$D3#lo,#0xfc000000	@ &=0x03ffffff
587	 vadd.i64	$D1,$D1,$T1		@ h0 -> h1
588	 vbic.i32	$D0#lo,#0xfc000000
589
590	vshrn.u64	$T0#lo,$D4,#26
591	vmovn.i64	$D4#lo,$D4
592	 vshr.u64	$T1,$D1,#26
593	 vmovn.i64	$D1#lo,$D1
594	 vadd.i64	$D2,$D2,$T1		@ h1 -> h2
595	vbic.i32	$D4#lo,#0xfc000000
596	 vbic.i32	$D1#lo,#0xfc000000
597
598	vadd.i32	$D0#lo,$D0#lo,$T0#lo
599	vshl.u32	$T0#lo,$T0#lo,#2
600	 vshrn.u64	$T1#lo,$D2,#26
601	 vmovn.i64	$D2#lo,$D2
602	vadd.i32	$D0#lo,$D0#lo,$T0#lo	@ h4 -> h0
603	 vadd.i32	$D3#lo,$D3#lo,$T1#lo	@ h2 -> h3
604	 vbic.i32	$D2#lo,#0xfc000000
605
606	vshr.u32	$T0#lo,$D0#lo,#26
607	vbic.i32	$D0#lo,#0xfc000000
608	 vshr.u32	$T1#lo,$D3#lo,#26
609	 vbic.i32	$D3#lo,#0xfc000000
610	vadd.i32	$D1#lo,$D1#lo,$T0#lo	@ h0 -> h1
611	 vadd.i32	$D4#lo,$D4#lo,$T1#lo	@ h3 -> h4
612
613	subs		$zeros,$zeros,#1
614	beq		.Lsquare_break_neon
615
616	add		$tbl0,$ctx,#(48+0*9*4)
617	add		$tbl1,$ctx,#(48+1*9*4)
618
619	vtrn.32		$R0,$D0#lo		@ r^2:r^1
620	vtrn.32		$R2,$D2#lo
621	vtrn.32		$R3,$D3#lo
622	vtrn.32		$R1,$D1#lo
623	vtrn.32		$R4,$D4#lo
624
625	vshl.u32	$S2,$R2,#2		@ *5
626	vshl.u32	$S3,$R3,#2
627	vshl.u32	$S1,$R1,#2
628	vshl.u32	$S4,$R4,#2
629	vadd.i32	$S2,$S2,$R2
630	vadd.i32	$S1,$S1,$R1
631	vadd.i32	$S3,$S3,$R3
632	vadd.i32	$S4,$S4,$R4
633
634	vst4.32		{${R0}[0],${R1}[0],${S1}[0],${R2}[0]},[$tbl0]!
635	vst4.32		{${R0}[1],${R1}[1],${S1}[1],${R2}[1]},[$tbl1]!
636	vst4.32		{${S2}[0],${R3}[0],${S3}[0],${R4}[0]},[$tbl0]!
637	vst4.32		{${S2}[1],${R3}[1],${S3}[1],${R4}[1]},[$tbl1]!
638	vst1.32		{${S4}[0]},[$tbl0,:32]
639	vst1.32		{${S4}[1]},[$tbl1,:32]
640
641	b		.Lsquare_neon
642
643.align	4
644.Lsquare_break_neon:
645	add		$tbl0,$ctx,#(48+2*4*9)
646	add		$tbl1,$ctx,#(48+3*4*9)
647
648	vmov		$R0,$D0#lo		@ r^4:r^3
649	vshl.u32	$S1,$D1#lo,#2		@ *5
650	vmov		$R1,$D1#lo
651	vshl.u32	$S2,$D2#lo,#2
652	vmov		$R2,$D2#lo
653	vshl.u32	$S3,$D3#lo,#2
654	vmov		$R3,$D3#lo
655	vshl.u32	$S4,$D4#lo,#2
656	vmov		$R4,$D4#lo
657	vadd.i32	$S1,$S1,$D1#lo
658	vadd.i32	$S2,$S2,$D2#lo
659	vadd.i32	$S3,$S3,$D3#lo
660	vadd.i32	$S4,$S4,$D4#lo
661
662	vst4.32		{${R0}[0],${R1}[0],${S1}[0],${R2}[0]},[$tbl0]!
663	vst4.32		{${R0}[1],${R1}[1],${S1}[1],${R2}[1]},[$tbl1]!
664	vst4.32		{${S2}[0],${R3}[0],${S3}[0],${R4}[0]},[$tbl0]!
665	vst4.32		{${S2}[1],${R3}[1],${S3}[1],${R4}[1]},[$tbl1]!
666	vst1.32		{${S4}[0]},[$tbl0]
667	vst1.32		{${S4}[1]},[$tbl1]
668
669	ret				@ bx	lr
670.size	poly1305_init_neon,.-poly1305_init_neon
671
672.type	poly1305_blocks_neon,%function
673.align	5
674poly1305_blocks_neon:
675.Lpoly1305_blocks_neon:
676	ldr	ip,[$ctx,#36]		@ is_base2_26
677	ands	$len,$len,#-16
678	beq	.Lno_data_neon
679
680	cmp	$len,#64
681	bhs	.Lenter_neon
682	tst	ip,ip			@ is_base2_26?
683	beq	.Lpoly1305_blocks
684
685.Lenter_neon:
686	stmdb	sp!,{r4-r7}
687	vstmdb	sp!,{d8-d15}		@ ABI specification says so
688
689	tst	ip,ip			@ is_base2_26?
690	bne	.Lbase2_26_neon
691
692	stmdb	sp!,{r1-r3,lr}
693	bl	poly1305_init_neon
694
695	ldr	r4,[$ctx,#0]		@ load hash value base 2^32
696	ldr	r5,[$ctx,#4]
697	ldr	r6,[$ctx,#8]
698	ldr	r7,[$ctx,#12]
699	ldr	ip,[$ctx,#16]
700
701	and	r2,r4,#0x03ffffff	@ base 2^32 -> base 2^26
702	mov	r3,r4,lsr#26
703	 veor	$D0#lo,$D0#lo,$D0#lo
704	mov	r4,r5,lsr#20
705	orr	r3,r3,r5,lsl#6
706	 veor	$D1#lo,$D1#lo,$D1#lo
707	mov	r5,r6,lsr#14
708	orr	r4,r4,r6,lsl#12
709	 veor	$D2#lo,$D2#lo,$D2#lo
710	mov	r6,r7,lsr#8
711	orr	r5,r5,r7,lsl#18
712	 veor	$D3#lo,$D3#lo,$D3#lo
713	and	r3,r3,#0x03ffffff
714	orr	r6,r6,ip,lsl#24
715	 veor	$D4#lo,$D4#lo,$D4#lo
716	and	r4,r4,#0x03ffffff
717	mov	r1,#1
718	and	r5,r5,#0x03ffffff
719	str	r1,[$ctx,#36]		@ is_base2_26
720
721	vmov.32	$D0#lo[0],r2
722	vmov.32	$D1#lo[0],r3
723	vmov.32	$D2#lo[0],r4
724	vmov.32	$D3#lo[0],r5
725	vmov.32	$D4#lo[0],r6
726	adr	$zeros,.Lzeros
727
728	ldmia	sp!,{r1-r3,lr}
729	b	.Lbase2_32_neon
730
731.align	4
732.Lbase2_26_neon:
733	@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@
734	@ load hash value
735
736	veor		$D0#lo,$D0#lo,$D0#lo
737	veor		$D1#lo,$D1#lo,$D1#lo
738	veor		$D2#lo,$D2#lo,$D2#lo
739	veor		$D3#lo,$D3#lo,$D3#lo
740	veor		$D4#lo,$D4#lo,$D4#lo
741	vld4.32		{$D0#lo[0],$D1#lo[0],$D2#lo[0],$D3#lo[0]},[$ctx]!
742	adr		$zeros,.Lzeros
743	vld1.32		{$D4#lo[0]},[$ctx]
744	sub		$ctx,$ctx,#16		@ rewind
745
746.Lbase2_32_neon:
747	add		$in2,$inp,#32
748	mov		$padbit,$padbit,lsl#24
749	tst		$len,#31
750	beq		.Leven
751
752	vld4.32		{$H0#lo[0],$H1#lo[0],$H2#lo[0],$H3#lo[0]},[$inp]!
753	vmov.32		$H4#lo[0],$padbit
754	sub		$len,$len,#16
755	add		$in2,$inp,#32
756
757# ifdef	__ARMEB__
758	vrev32.8	$H0,$H0
759	vrev32.8	$H3,$H3
760	vrev32.8	$H1,$H1
761	vrev32.8	$H2,$H2
762# endif
763	vsri.u32	$H4#lo,$H3#lo,#8	@ base 2^32 -> base 2^26
764	vshl.u32	$H3#lo,$H3#lo,#18
765
766	vsri.u32	$H3#lo,$H2#lo,#14
767	vshl.u32	$H2#lo,$H2#lo,#12
768	vadd.i32	$H4#hi,$H4#lo,$D4#lo	@ add hash value and move to #hi
769
770	vbic.i32	$H3#lo,#0xfc000000
771	vsri.u32	$H2#lo,$H1#lo,#20
772	vshl.u32	$H1#lo,$H1#lo,#6
773
774	vbic.i32	$H2#lo,#0xfc000000
775	vsri.u32	$H1#lo,$H0#lo,#26
776	vadd.i32	$H3#hi,$H3#lo,$D3#lo
777
778	vbic.i32	$H0#lo,#0xfc000000
779	vbic.i32	$H1#lo,#0xfc000000
780	vadd.i32	$H2#hi,$H2#lo,$D2#lo
781
782	vadd.i32	$H0#hi,$H0#lo,$D0#lo
783	vadd.i32	$H1#hi,$H1#lo,$D1#lo
784
785	mov		$tbl1,$zeros
786	add		$tbl0,$ctx,#48
787
788	cmp		$len,$len
789	b		.Long_tail
790
791.align	4
792.Leven:
793	subs		$len,$len,#64
794	it		lo
795	movlo		$in2,$zeros
796
797	vmov.i32	$H4,#1<<24		@ padbit, yes, always
798	vld4.32		{$H0#lo,$H1#lo,$H2#lo,$H3#lo},[$inp]	@ inp[0:1]
799	add		$inp,$inp,#64
800	vld4.32		{$H0#hi,$H1#hi,$H2#hi,$H3#hi},[$in2]	@ inp[2:3] (or 0)
801	add		$in2,$in2,#64
802	itt		hi
803	addhi		$tbl1,$ctx,#(48+1*9*4)
804	addhi		$tbl0,$ctx,#(48+3*9*4)
805
806# ifdef	__ARMEB__
807	vrev32.8	$H0,$H0
808	vrev32.8	$H3,$H3
809	vrev32.8	$H1,$H1
810	vrev32.8	$H2,$H2
811# endif
812	vsri.u32	$H4,$H3,#8		@ base 2^32 -> base 2^26
813	vshl.u32	$H3,$H3,#18
814
815	vsri.u32	$H3,$H2,#14
816	vshl.u32	$H2,$H2,#12
817
818	vbic.i32	$H3,#0xfc000000
819	vsri.u32	$H2,$H1,#20
820	vshl.u32	$H1,$H1,#6
821
822	vbic.i32	$H2,#0xfc000000
823	vsri.u32	$H1,$H0,#26
824
825	vbic.i32	$H0,#0xfc000000
826	vbic.i32	$H1,#0xfc000000
827
828	bls		.Lskip_loop
829
830	vld4.32		{${R0}[1],${R1}[1],${S1}[1],${R2}[1]},[$tbl1]!	@ load r^2
831	vld4.32		{${R0}[0],${R1}[0],${S1}[0],${R2}[0]},[$tbl0]!	@ load r^4
832	vld4.32		{${S2}[1],${R3}[1],${S3}[1],${R4}[1]},[$tbl1]!
833	vld4.32		{${S2}[0],${R3}[0],${S3}[0],${R4}[0]},[$tbl0]!
834	b		.Loop_neon
835
836.align	5
837.Loop_neon:
838	@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@
839	@ ((inp[0]*r^4+inp[2]*r^2+inp[4])*r^4+inp[6]*r^2
840	@ ((inp[1]*r^4+inp[3]*r^2+inp[5])*r^3+inp[7]*r
841	@   \___________________/
842	@ ((inp[0]*r^4+inp[2]*r^2+inp[4])*r^4+inp[6]*r^2+inp[8])*r^2
843	@ ((inp[1]*r^4+inp[3]*r^2+inp[5])*r^4+inp[7]*r^2+inp[9])*r
844	@   \___________________/ \____________________/
845	@
846	@ Note that we start with inp[2:3]*r^2. This is because it
847	@ doesn't depend on reduction in previous iteration.
848	@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@
849	@ d4 = h4*r0 + h3*r1   + h2*r2   + h1*r3   + h0*r4
850	@ d3 = h3*r0 + h2*r1   + h1*r2   + h0*r3   + h4*5*r4
851	@ d2 = h2*r0 + h1*r1   + h0*r2   + h4*5*r3 + h3*5*r4
852	@ d1 = h1*r0 + h0*r1   + h4*5*r2 + h3*5*r3 + h2*5*r4
853	@ d0 = h0*r0 + h4*5*r1 + h3*5*r2 + h2*5*r3 + h1*5*r4
854
855	@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@
856	@ inp[2:3]*r^2
857
858	vadd.i32	$H2#lo,$H2#lo,$D2#lo	@ accumulate inp[0:1]
859	vmull.u32	$D2,$H2#hi,${R0}[1]
860	vadd.i32	$H0#lo,$H0#lo,$D0#lo
861	vmull.u32	$D0,$H0#hi,${R0}[1]
862	vadd.i32	$H3#lo,$H3#lo,$D3#lo
863	vmull.u32	$D3,$H3#hi,${R0}[1]
864	vmlal.u32	$D2,$H1#hi,${R1}[1]
865	vadd.i32	$H1#lo,$H1#lo,$D1#lo
866	vmull.u32	$D1,$H1#hi,${R0}[1]
867
868	vadd.i32	$H4#lo,$H4#lo,$D4#lo
869	vmull.u32	$D4,$H4#hi,${R0}[1]
870	subs		$len,$len,#64
871	vmlal.u32	$D0,$H4#hi,${S1}[1]
872	it		lo
873	movlo		$in2,$zeros
874	vmlal.u32	$D3,$H2#hi,${R1}[1]
875	vld1.32		${S4}[1],[$tbl1,:32]
876	vmlal.u32	$D1,$H0#hi,${R1}[1]
877	vmlal.u32	$D4,$H3#hi,${R1}[1]
878
879	vmlal.u32	$D0,$H3#hi,${S2}[1]
880	vmlal.u32	$D3,$H1#hi,${R2}[1]
881	vmlal.u32	$D4,$H2#hi,${R2}[1]
882	vmlal.u32	$D1,$H4#hi,${S2}[1]
883	vmlal.u32	$D2,$H0#hi,${R2}[1]
884
885	vmlal.u32	$D3,$H0#hi,${R3}[1]
886	vmlal.u32	$D0,$H2#hi,${S3}[1]
887	vmlal.u32	$D4,$H1#hi,${R3}[1]
888	vmlal.u32	$D1,$H3#hi,${S3}[1]
889	vmlal.u32	$D2,$H4#hi,${S3}[1]
890
891	vmlal.u32	$D3,$H4#hi,${S4}[1]
892	vmlal.u32	$D0,$H1#hi,${S4}[1]
893	vmlal.u32	$D4,$H0#hi,${R4}[1]
894	vmlal.u32	$D1,$H2#hi,${S4}[1]
895	vmlal.u32	$D2,$H3#hi,${S4}[1]
896
897	vld4.32		{$H0#hi,$H1#hi,$H2#hi,$H3#hi},[$in2]	@ inp[2:3] (or 0)
898	add		$in2,$in2,#64
899
900	@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@
901	@ (hash+inp[0:1])*r^4 and accumulate
902
903	vmlal.u32	$D3,$H3#lo,${R0}[0]
904	vmlal.u32	$D0,$H0#lo,${R0}[0]
905	vmlal.u32	$D4,$H4#lo,${R0}[0]
906	vmlal.u32	$D1,$H1#lo,${R0}[0]
907	vmlal.u32	$D2,$H2#lo,${R0}[0]
908	vld1.32		${S4}[0],[$tbl0,:32]
909
910	vmlal.u32	$D3,$H2#lo,${R1}[0]
911	vmlal.u32	$D0,$H4#lo,${S1}[0]
912	vmlal.u32	$D4,$H3#lo,${R1}[0]
913	vmlal.u32	$D1,$H0#lo,${R1}[0]
914	vmlal.u32	$D2,$H1#lo,${R1}[0]
915
916	vmlal.u32	$D3,$H1#lo,${R2}[0]
917	vmlal.u32	$D0,$H3#lo,${S2}[0]
918	vmlal.u32	$D4,$H2#lo,${R2}[0]
919	vmlal.u32	$D1,$H4#lo,${S2}[0]
920	vmlal.u32	$D2,$H0#lo,${R2}[0]
921
922	vmlal.u32	$D3,$H0#lo,${R3}[0]
923	vmlal.u32	$D0,$H2#lo,${S3}[0]
924	vmlal.u32	$D4,$H1#lo,${R3}[0]
925	vmlal.u32	$D1,$H3#lo,${S3}[0]
926	vmlal.u32	$D3,$H4#lo,${S4}[0]
927
928	vmlal.u32	$D2,$H4#lo,${S3}[0]
929	vmlal.u32	$D0,$H1#lo,${S4}[0]
930	vmlal.u32	$D4,$H0#lo,${R4}[0]
931	vmov.i32	$H4,#1<<24		@ padbit, yes, always
932	vmlal.u32	$D1,$H2#lo,${S4}[0]
933	vmlal.u32	$D2,$H3#lo,${S4}[0]
934
935	vld4.32		{$H0#lo,$H1#lo,$H2#lo,$H3#lo},[$inp]	@ inp[0:1]
936	add		$inp,$inp,#64
937# ifdef	__ARMEB__
938	vrev32.8	$H0,$H0
939	vrev32.8	$H1,$H1
940	vrev32.8	$H2,$H2
941	vrev32.8	$H3,$H3
942# endif
943
944	@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@
945	@ lazy reduction interleaved with base 2^32 -> base 2^26 of
946	@ inp[0:3] previously loaded to $H0-$H3 and smashed to $H0-$H4.
947
948	vshr.u64	$T0,$D3,#26
949	vmovn.i64	$D3#lo,$D3
950	 vshr.u64	$T1,$D0,#26
951	 vmovn.i64	$D0#lo,$D0
952	vadd.i64	$D4,$D4,$T0		@ h3 -> h4
953	vbic.i32	$D3#lo,#0xfc000000
954	  vsri.u32	$H4,$H3,#8		@ base 2^32 -> base 2^26
955	 vadd.i64	$D1,$D1,$T1		@ h0 -> h1
956	  vshl.u32	$H3,$H3,#18
957	 vbic.i32	$D0#lo,#0xfc000000
958
959	vshrn.u64	$T0#lo,$D4,#26
960	vmovn.i64	$D4#lo,$D4
961	 vshr.u64	$T1,$D1,#26
962	 vmovn.i64	$D1#lo,$D1
963	 vadd.i64	$D2,$D2,$T1		@ h1 -> h2
964	  vsri.u32	$H3,$H2,#14
965	vbic.i32	$D4#lo,#0xfc000000
966	  vshl.u32	$H2,$H2,#12
967	 vbic.i32	$D1#lo,#0xfc000000
968
969	vadd.i32	$D0#lo,$D0#lo,$T0#lo
970	vshl.u32	$T0#lo,$T0#lo,#2
971	  vbic.i32	$H3,#0xfc000000
972	 vshrn.u64	$T1#lo,$D2,#26
973	 vmovn.i64	$D2#lo,$D2
974	vaddl.u32	$D0,$D0#lo,$T0#lo	@ h4 -> h0 [widen for a sec]
975	  vsri.u32	$H2,$H1,#20
976	 vadd.i32	$D3#lo,$D3#lo,$T1#lo	@ h2 -> h3
977	  vshl.u32	$H1,$H1,#6
978	 vbic.i32	$D2#lo,#0xfc000000
979	  vbic.i32	$H2,#0xfc000000
980
981	vshrn.u64	$T0#lo,$D0,#26		@ re-narrow
982	vmovn.i64	$D0#lo,$D0
983	  vsri.u32	$H1,$H0,#26
984	  vbic.i32	$H0,#0xfc000000
985	 vshr.u32	$T1#lo,$D3#lo,#26
986	 vbic.i32	$D3#lo,#0xfc000000
987	vbic.i32	$D0#lo,#0xfc000000
988	vadd.i32	$D1#lo,$D1#lo,$T0#lo	@ h0 -> h1
989	 vadd.i32	$D4#lo,$D4#lo,$T1#lo	@ h3 -> h4
990	  vbic.i32	$H1,#0xfc000000
991
992	bhi		.Loop_neon
993
994.Lskip_loop:
995	@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@
996	@ multiply (inp[0:1]+hash) or inp[2:3] by r^2:r^1
997
998	add		$tbl1,$ctx,#(48+0*9*4)
999	add		$tbl0,$ctx,#(48+1*9*4)
1000	adds		$len,$len,#32
1001	it		ne
1002	movne		$len,#0
1003	bne		.Long_tail
1004
1005	vadd.i32	$H2#hi,$H2#lo,$D2#lo	@ add hash value and move to #hi
1006	vadd.i32	$H0#hi,$H0#lo,$D0#lo
1007	vadd.i32	$H3#hi,$H3#lo,$D3#lo
1008	vadd.i32	$H1#hi,$H1#lo,$D1#lo
1009	vadd.i32	$H4#hi,$H4#lo,$D4#lo
1010
1011.Long_tail:
1012	vld4.32		{${R0}[1],${R1}[1],${S1}[1],${R2}[1]},[$tbl1]!	@ load r^1
1013	vld4.32		{${R0}[0],${R1}[0],${S1}[0],${R2}[0]},[$tbl0]!	@ load r^2
1014
1015	vadd.i32	$H2#lo,$H2#lo,$D2#lo	@ can be redundant
1016	vmull.u32	$D2,$H2#hi,$R0
1017	vadd.i32	$H0#lo,$H0#lo,$D0#lo
1018	vmull.u32	$D0,$H0#hi,$R0
1019	vadd.i32	$H3#lo,$H3#lo,$D3#lo
1020	vmull.u32	$D3,$H3#hi,$R0
1021	vadd.i32	$H1#lo,$H1#lo,$D1#lo
1022	vmull.u32	$D1,$H1#hi,$R0
1023	vadd.i32	$H4#lo,$H4#lo,$D4#lo
1024	vmull.u32	$D4,$H4#hi,$R0
1025
1026	vmlal.u32	$D0,$H4#hi,$S1
1027	vld4.32		{${S2}[1],${R3}[1],${S3}[1],${R4}[1]},[$tbl1]!
1028	vmlal.u32	$D3,$H2#hi,$R1
1029	vld4.32		{${S2}[0],${R3}[0],${S3}[0],${R4}[0]},[$tbl0]!
1030	vmlal.u32	$D1,$H0#hi,$R1
1031	vmlal.u32	$D4,$H3#hi,$R1
1032	vmlal.u32	$D2,$H1#hi,$R1
1033
1034	vmlal.u32	$D3,$H1#hi,$R2
1035	vld1.32		${S4}[1],[$tbl1,:32]
1036	vmlal.u32	$D0,$H3#hi,$S2
1037	vld1.32		${S4}[0],[$tbl0,:32]
1038	vmlal.u32	$D4,$H2#hi,$R2
1039	vmlal.u32	$D1,$H4#hi,$S2
1040	vmlal.u32	$D2,$H0#hi,$R2
1041
1042	vmlal.u32	$D3,$H0#hi,$R3
1043	 it		ne
1044	 addne		$tbl1,$ctx,#(48+2*9*4)
1045	vmlal.u32	$D0,$H2#hi,$S3
1046	 it		ne
1047	 addne		$tbl0,$ctx,#(48+3*9*4)
1048	vmlal.u32	$D4,$H1#hi,$R3
1049	vmlal.u32	$D1,$H3#hi,$S3
1050	vmlal.u32	$D2,$H4#hi,$S3
1051
1052	vmlal.u32	$D3,$H4#hi,$S4
1053	 vorn		$MASK,$MASK,$MASK	@ all-ones, can be redundant
1054	vmlal.u32	$D0,$H1#hi,$S4
1055	 vshr.u64	$MASK,$MASK,#38
1056	vmlal.u32	$D4,$H0#hi,$R4
1057	vmlal.u32	$D1,$H2#hi,$S4
1058	vmlal.u32	$D2,$H3#hi,$S4
1059
1060	beq		.Lshort_tail
1061
1062	@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@
1063	@ (hash+inp[0:1])*r^4:r^3 and accumulate
1064
1065	vld4.32		{${R0}[1],${R1}[1],${S1}[1],${R2}[1]},[$tbl1]!	@ load r^3
1066	vld4.32		{${R0}[0],${R1}[0],${S1}[0],${R2}[0]},[$tbl0]!	@ load r^4
1067
1068	vmlal.u32	$D2,$H2#lo,$R0
1069	vmlal.u32	$D0,$H0#lo,$R0
1070	vmlal.u32	$D3,$H3#lo,$R0
1071	vmlal.u32	$D1,$H1#lo,$R0
1072	vmlal.u32	$D4,$H4#lo,$R0
1073
1074	vmlal.u32	$D0,$H4#lo,$S1
1075	vld4.32		{${S2}[1],${R3}[1],${S3}[1],${R4}[1]},[$tbl1]!
1076	vmlal.u32	$D3,$H2#lo,$R1
1077	vld4.32		{${S2}[0],${R3}[0],${S3}[0],${R4}[0]},[$tbl0]!
1078	vmlal.u32	$D1,$H0#lo,$R1
1079	vmlal.u32	$D4,$H3#lo,$R1
1080	vmlal.u32	$D2,$H1#lo,$R1
1081
1082	vmlal.u32	$D3,$H1#lo,$R2
1083	vld1.32		${S4}[1],[$tbl1,:32]
1084	vmlal.u32	$D0,$H3#lo,$S2
1085	vld1.32		${S4}[0],[$tbl0,:32]
1086	vmlal.u32	$D4,$H2#lo,$R2
1087	vmlal.u32	$D1,$H4#lo,$S2
1088	vmlal.u32	$D2,$H0#lo,$R2
1089
1090	vmlal.u32	$D3,$H0#lo,$R3
1091	vmlal.u32	$D0,$H2#lo,$S3
1092	vmlal.u32	$D4,$H1#lo,$R3
1093	vmlal.u32	$D1,$H3#lo,$S3
1094	vmlal.u32	$D2,$H4#lo,$S3
1095
1096	vmlal.u32	$D3,$H4#lo,$S4
1097	 vorn		$MASK,$MASK,$MASK	@ all-ones
1098	vmlal.u32	$D0,$H1#lo,$S4
1099	 vshr.u64	$MASK,$MASK,#38
1100	vmlal.u32	$D4,$H0#lo,$R4
1101	vmlal.u32	$D1,$H2#lo,$S4
1102	vmlal.u32	$D2,$H3#lo,$S4
1103
1104.Lshort_tail:
1105	@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@
1106	@ horizontal addition
1107
1108	vadd.i64	$D3#lo,$D3#lo,$D3#hi
1109	vadd.i64	$D0#lo,$D0#lo,$D0#hi
1110	vadd.i64	$D4#lo,$D4#lo,$D4#hi
1111	vadd.i64	$D1#lo,$D1#lo,$D1#hi
1112	vadd.i64	$D2#lo,$D2#lo,$D2#hi
1113
1114	@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@
1115	@ lazy reduction, but without narrowing
1116
1117	vshr.u64	$T0,$D3,#26
1118	vand.i64	$D3,$D3,$MASK
1119	 vshr.u64	$T1,$D0,#26
1120	 vand.i64	$D0,$D0,$MASK
1121	vadd.i64	$D4,$D4,$T0		@ h3 -> h4
1122	 vadd.i64	$D1,$D1,$T1		@ h0 -> h1
1123
1124	vshr.u64	$T0,$D4,#26
1125	vand.i64	$D4,$D4,$MASK
1126	 vshr.u64	$T1,$D1,#26
1127	 vand.i64	$D1,$D1,$MASK
1128	 vadd.i64	$D2,$D2,$T1		@ h1 -> h2
1129
1130	vadd.i64	$D0,$D0,$T0
1131	vshl.u64	$T0,$T0,#2
1132	 vshr.u64	$T1,$D2,#26
1133	 vand.i64	$D2,$D2,$MASK
1134	vadd.i64	$D0,$D0,$T0		@ h4 -> h0
1135	 vadd.i64	$D3,$D3,$T1		@ h2 -> h3
1136
1137	vshr.u64	$T0,$D0,#26
1138	vand.i64	$D0,$D0,$MASK
1139	 vshr.u64	$T1,$D3,#26
1140	 vand.i64	$D3,$D3,$MASK
1141	vadd.i64	$D1,$D1,$T0		@ h0 -> h1
1142	 vadd.i64	$D4,$D4,$T1		@ h3 -> h4
1143
1144	cmp		$len,#0
1145	bne		.Leven
1146
1147	@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@
1148	@ store hash value
1149
1150	vst4.32		{$D0#lo[0],$D1#lo[0],$D2#lo[0],$D3#lo[0]},[$ctx]!
1151	vst1.32		{$D4#lo[0]},[$ctx]
1152
1153	vldmia	sp!,{d8-d15}			@ epilogue
1154	ldmia	sp!,{r4-r7}
1155.Lno_data_neon:
1156	ret					@ bx	lr
1157.size	poly1305_blocks_neon,.-poly1305_blocks_neon
1158
1159.type	poly1305_emit_neon,%function
1160.align	5
1161poly1305_emit_neon:
1162.Lpoly1305_emit_neon:
1163	ldr	ip,[$ctx,#36]		@ is_base2_26
1164
1165	stmdb	sp!,{r4-r11}
1166
1167	tst	ip,ip
1168	beq	.Lpoly1305_emit_enter
1169
1170	ldmia	$ctx,{$h0-$h4}
1171	eor	$g0,$g0,$g0
1172
1173	adds	$h0,$h0,$h1,lsl#26	@ base 2^26 -> base 2^32
1174	mov	$h1,$h1,lsr#6
1175	adcs	$h1,$h1,$h2,lsl#20
1176	mov	$h2,$h2,lsr#12
1177	adcs	$h2,$h2,$h3,lsl#14
1178	mov	$h3,$h3,lsr#18
1179	adcs	$h3,$h3,$h4,lsl#8
1180	adc	$h4,$g0,$h4,lsr#24	@ can be partially reduced ...
1181
1182	and	$g0,$h4,#-4		@ ... so reduce
1183	and	$h4,$h3,#3
1184	add	$g0,$g0,$g0,lsr#2	@ *= 5
1185	adds	$h0,$h0,$g0
1186	adcs	$h1,$h1,#0
1187	adcs	$h2,$h2,#0
1188	adcs	$h3,$h3,#0
1189	adc	$h4,$h4,#0
1190
1191	adds	$g0,$h0,#5		@ compare to modulus
1192	adcs	$g1,$h1,#0
1193	adcs	$g2,$h2,#0
1194	adcs	$g3,$h3,#0
1195	adc	$g4,$h4,#0
1196	tst	$g4,#4			@ did it carry/borrow?
1197
1198	it	ne
1199	movne	$h0,$g0
1200	ldr	$g0,[$nonce,#0]
1201	it	ne
1202	movne	$h1,$g1
1203	ldr	$g1,[$nonce,#4]
1204	it	ne
1205	movne	$h2,$g2
1206	ldr	$g2,[$nonce,#8]
1207	it	ne
1208	movne	$h3,$g3
1209	ldr	$g3,[$nonce,#12]
1210
1211	adds	$h0,$h0,$g0		@ accumulate nonce
1212	adcs	$h1,$h1,$g1
1213	adcs	$h2,$h2,$g2
1214	adc	$h3,$h3,$g3
1215
1216# ifdef __ARMEB__
1217	rev	$h0,$h0
1218	rev	$h1,$h1
1219	rev	$h2,$h2
1220	rev	$h3,$h3
1221# endif
1222	str	$h0,[$mac,#0]		@ store the result
1223	str	$h1,[$mac,#4]
1224	str	$h2,[$mac,#8]
1225	str	$h3,[$mac,#12]
1226
1227	ldmia	sp!,{r4-r11}
1228	ret				@ bx	lr
1229.size	poly1305_emit_neon,.-poly1305_emit_neon
1230
1231.align	5
1232.Lzeros:
1233.long	0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0
1234.LOPENSSL_armcap:
1235.word	OPENSSL_armcap_P-.Lpoly1305_init
1236#endif
1237___
1238}	}
1239$code.=<<___;
1240.asciz	"Poly1305 for ARMv4/NEON, CRYPTOGAMS by <appro\@openssl.org>"
1241.align	2
1242#if	__ARM_MAX_ARCH__>=7
1243.comm   OPENSSL_armcap_P,4,4
1244#endif
1245___
1246
1247foreach (split("\n",$code)) {
1248	s/\`([^\`]*)\`/eval $1/geo;
1249
1250	s/\bq([0-9]+)#(lo|hi)/sprintf "d%d",2*$1+($2 eq "hi")/geo	or
1251	s/\bret\b/bx	lr/go						or
1252	s/\bbx\s+lr\b/.word\t0xe12fff1e/go;	# make it possible to compile with -march=armv4
1253
1254	print $_,"\n";
1255}
1256close STDOUT or die "error closing STDOUT: $!"; # enforce flush
1257