1#! /usr/bin/env perl
2# Copyright 2016 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# This module implements Poly1305 hash for ARMv8.
18#
19# June 2015
20#
21# Numbers are cycles per processed byte with poly1305_blocks alone.
22#
23#		IALU/gcc-4.9	NEON
24#
25# Apple A7	1.86/+5%	0.72
26# Cortex-A53	2.69/+58%	1.47
27# Cortex-A57	2.70/+7%	1.14
28# Denver	1.64/+50%	1.18(*)
29# X-Gene	2.13/+68%	2.27
30# Mongoose	1.77/+75%	1.12
31# Kryo		2.70/+55%	1.13
32#
33# (*)	estimate based on resources availability is less than 1.0,
34#	i.e. measured result is worse than expected, presumably binary
35#	translator is not almighty;
36
37$flavour=shift;
38$output=shift;
39
40$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
41( $xlate="${dir}arm-xlate.pl" and -f $xlate ) or
42( $xlate="${dir}../../perlasm/arm-xlate.pl" and -f $xlate) or
43die "can't locate arm-xlate.pl";
44
45open OUT,"| \"$^X\" $xlate $flavour $output";
46*STDOUT=*OUT;
47
48my ($ctx,$inp,$len,$padbit) = map("x$_",(0..3));
49my ($mac,$nonce)=($inp,$len);
50
51my ($h0,$h1,$h2,$r0,$r1,$s1,$t0,$t1,$d0,$d1,$d2) = map("x$_",(4..14));
52
53$code.=<<___;
54#include "arm_arch.h"
55
56.text
57
58// forward "declarations" are required for Apple
59.extern	OPENSSL_armcap_P
60.globl	poly1305_blocks
61.globl	poly1305_emit
62
63.globl	poly1305_init
64.type	poly1305_init,%function
65.align	5
66poly1305_init:
67	cmp	$inp,xzr
68	stp	xzr,xzr,[$ctx]		// zero hash value
69	stp	xzr,xzr,[$ctx,#16]	// [along with is_base2_26]
70
71	csel	x0,xzr,x0,eq
72	b.eq	.Lno_key
73
74#ifdef	__ILP32__
75	ldrsw	$t1,.LOPENSSL_armcap_P
76#else
77	ldr	$t1,.LOPENSSL_armcap_P
78#endif
79	adr	$t0,.LOPENSSL_armcap_P
80
81	ldp	$r0,$r1,[$inp]		// load key
82	mov	$s1,#0xfffffffc0fffffff
83	movk	$s1,#0x0fff,lsl#48
84	ldr	w17,[$t0,$t1]
85#ifdef	__ARMEB__
86	rev	$r0,$r0			// flip bytes
87	rev	$r1,$r1
88#endif
89	and	$r0,$r0,$s1		// &=0ffffffc0fffffff
90	and	$s1,$s1,#-4
91	and	$r1,$r1,$s1		// &=0ffffffc0ffffffc
92	stp	$r0,$r1,[$ctx,#32]	// save key value
93
94	tst	w17,#ARMV7_NEON
95
96	adr	$d0,poly1305_blocks
97	adr	$r0,poly1305_blocks_neon
98	adr	$d1,poly1305_emit
99	adr	$r1,poly1305_emit_neon
100
101	csel	$d0,$d0,$r0,eq
102	csel	$d1,$d1,$r1,eq
103
104#ifdef	__ILP32__
105	stp	w12,w13,[$len]
106#else
107	stp	$d0,$d1,[$len]
108#endif
109
110	mov	x0,#1
111.Lno_key:
112	ret
113.size	poly1305_init,.-poly1305_init
114
115.type	poly1305_blocks,%function
116.align	5
117poly1305_blocks:
118	ands	$len,$len,#-16
119	b.eq	.Lno_data
120
121	ldp	$h0,$h1,[$ctx]		// load hash value
122	ldp	$r0,$r1,[$ctx,#32]	// load key value
123	ldr	$h2,[$ctx,#16]
124	add	$s1,$r1,$r1,lsr#2	// s1 = r1 + (r1 >> 2)
125	b	.Loop
126
127.align	5
128.Loop:
129	ldp	$t0,$t1,[$inp],#16	// load input
130	sub	$len,$len,#16
131#ifdef	__ARMEB__
132	rev	$t0,$t0
133	rev	$t1,$t1
134#endif
135	adds	$h0,$h0,$t0		// accumulate input
136	adcs	$h1,$h1,$t1
137
138	mul	$d0,$h0,$r0		// h0*r0
139	adc	$h2,$h2,$padbit
140	umulh	$d1,$h0,$r0
141
142	mul	$t0,$h1,$s1		// h1*5*r1
143	umulh	$t1,$h1,$s1
144
145	adds	$d0,$d0,$t0
146	mul	$t0,$h0,$r1		// h0*r1
147	adc	$d1,$d1,$t1
148	umulh	$d2,$h0,$r1
149
150	adds	$d1,$d1,$t0
151	mul	$t0,$h1,$r0		// h1*r0
152	adc	$d2,$d2,xzr
153	umulh	$t1,$h1,$r0
154
155	adds	$d1,$d1,$t0
156	mul	$t0,$h2,$s1		// h2*5*r1
157	adc	$d2,$d2,$t1
158	mul	$t1,$h2,$r0		// h2*r0
159
160	adds	$d1,$d1,$t0
161	adc	$d2,$d2,$t1
162
163	and	$t0,$d2,#-4		// final reduction
164	and	$h2,$d2,#3
165	add	$t0,$t0,$d2,lsr#2
166	adds	$h0,$d0,$t0
167	adcs	$h1,$d1,xzr
168	adc	$h2,$h2,xzr
169
170	cbnz	$len,.Loop
171
172	stp	$h0,$h1,[$ctx]		// store hash value
173	str	$h2,[$ctx,#16]
174
175.Lno_data:
176	ret
177.size	poly1305_blocks,.-poly1305_blocks
178
179.type	poly1305_emit,%function
180.align	5
181poly1305_emit:
182	ldp	$h0,$h1,[$ctx]		// load hash base 2^64
183	ldr	$h2,[$ctx,#16]
184	ldp	$t0,$t1,[$nonce]	// load nonce
185
186	adds	$d0,$h0,#5		// compare to modulus
187	adcs	$d1,$h1,xzr
188	adc	$d2,$h2,xzr
189
190	tst	$d2,#-4			// see if it's carried/borrowed
191
192	csel	$h0,$h0,$d0,eq
193	csel	$h1,$h1,$d1,eq
194
195#ifdef	__ARMEB__
196	ror	$t0,$t0,#32		// flip nonce words
197	ror	$t1,$t1,#32
198#endif
199	adds	$h0,$h0,$t0		// accumulate nonce
200	adc	$h1,$h1,$t1
201#ifdef	__ARMEB__
202	rev	$h0,$h0			// flip output bytes
203	rev	$h1,$h1
204#endif
205	stp	$h0,$h1,[$mac]		// write result
206
207	ret
208.size	poly1305_emit,.-poly1305_emit
209___
210my ($R0,$R1,$S1,$R2,$S2,$R3,$S3,$R4,$S4) = map("v$_.4s",(0..8));
211my ($IN01_0,$IN01_1,$IN01_2,$IN01_3,$IN01_4) = map("v$_.2s",(9..13));
212my ($IN23_0,$IN23_1,$IN23_2,$IN23_3,$IN23_4) = map("v$_.2s",(14..18));
213my ($ACC0,$ACC1,$ACC2,$ACC3,$ACC4) = map("v$_.2d",(19..23));
214my ($H0,$H1,$H2,$H3,$H4) = map("v$_.2s",(24..28));
215my ($T0,$T1,$MASK) = map("v$_",(29..31));
216
217my ($in2,$zeros)=("x16","x17");
218my $is_base2_26 = $zeros;		# borrow
219
220$code.=<<___;
221.type	poly1305_mult,%function
222.align	5
223poly1305_mult:
224	mul	$d0,$h0,$r0		// h0*r0
225	umulh	$d1,$h0,$r0
226
227	mul	$t0,$h1,$s1		// h1*5*r1
228	umulh	$t1,$h1,$s1
229
230	adds	$d0,$d0,$t0
231	mul	$t0,$h0,$r1		// h0*r1
232	adc	$d1,$d1,$t1
233	umulh	$d2,$h0,$r1
234
235	adds	$d1,$d1,$t0
236	mul	$t0,$h1,$r0		// h1*r0
237	adc	$d2,$d2,xzr
238	umulh	$t1,$h1,$r0
239
240	adds	$d1,$d1,$t0
241	mul	$t0,$h2,$s1		// h2*5*r1
242	adc	$d2,$d2,$t1
243	mul	$t1,$h2,$r0		// h2*r0
244
245	adds	$d1,$d1,$t0
246	adc	$d2,$d2,$t1
247
248	and	$t0,$d2,#-4		// final reduction
249	and	$h2,$d2,#3
250	add	$t0,$t0,$d2,lsr#2
251	adds	$h0,$d0,$t0
252	adcs	$h1,$d1,xzr
253	adc	$h2,$h2,xzr
254
255	ret
256.size	poly1305_mult,.-poly1305_mult
257
258.type	poly1305_splat,%function
259.align	5
260poly1305_splat:
261	and	x12,$h0,#0x03ffffff	// base 2^64 -> base 2^26
262	ubfx	x13,$h0,#26,#26
263	extr	x14,$h1,$h0,#52
264	and	x14,x14,#0x03ffffff
265	ubfx	x15,$h1,#14,#26
266	extr	x16,$h2,$h1,#40
267
268	str	w12,[$ctx,#16*0]	// r0
269	add	w12,w13,w13,lsl#2	// r1*5
270	str	w13,[$ctx,#16*1]	// r1
271	add	w13,w14,w14,lsl#2	// r2*5
272	str	w12,[$ctx,#16*2]	// s1
273	str	w14,[$ctx,#16*3]	// r2
274	add	w14,w15,w15,lsl#2	// r3*5
275	str	w13,[$ctx,#16*4]	// s2
276	str	w15,[$ctx,#16*5]	// r3
277	add	w15,w16,w16,lsl#2	// r4*5
278	str	w14,[$ctx,#16*6]	// s3
279	str	w16,[$ctx,#16*7]	// r4
280	str	w15,[$ctx,#16*8]	// s4
281
282	ret
283.size	poly1305_splat,.-poly1305_splat
284
285.type	poly1305_blocks_neon,%function
286.align	5
287poly1305_blocks_neon:
288	ldr	$is_base2_26,[$ctx,#24]
289	cmp	$len,#128
290	b.hs	.Lblocks_neon
291	cbz	$is_base2_26,poly1305_blocks
292
293.Lblocks_neon:
294	stp	x29,x30,[sp,#-80]!
295	add	x29,sp,#0
296
297	ands	$len,$len,#-16
298	b.eq	.Lno_data_neon
299
300	cbz	$is_base2_26,.Lbase2_64_neon
301
302	ldp	w10,w11,[$ctx]		// load hash value base 2^26
303	ldp	w12,w13,[$ctx,#8]
304	ldr	w14,[$ctx,#16]
305
306	tst	$len,#31
307	b.eq	.Leven_neon
308
309	ldp	$r0,$r1,[$ctx,#32]	// load key value
310
311	add	$h0,x10,x11,lsl#26	// base 2^26 -> base 2^64
312	lsr	$h1,x12,#12
313	adds	$h0,$h0,x12,lsl#52
314	add	$h1,$h1,x13,lsl#14
315	adc	$h1,$h1,xzr
316	lsr	$h2,x14,#24
317	adds	$h1,$h1,x14,lsl#40
318	adc	$d2,$h2,xzr		// can be partially reduced...
319
320	ldp	$d0,$d1,[$inp],#16	// load input
321	sub	$len,$len,#16
322	add	$s1,$r1,$r1,lsr#2	// s1 = r1 + (r1 >> 2)
323
324	and	$t0,$d2,#-4		// ... so reduce
325	and	$h2,$d2,#3
326	add	$t0,$t0,$d2,lsr#2
327	adds	$h0,$h0,$t0
328	adcs	$h1,$h1,xzr
329	adc	$h2,$h2,xzr
330
331#ifdef	__ARMEB__
332	rev	$d0,$d0
333	rev	$d1,$d1
334#endif
335	adds	$h0,$h0,$d0		// accumulate input
336	adcs	$h1,$h1,$d1
337	adc	$h2,$h2,$padbit
338
339	bl	poly1305_mult
340	ldr	x30,[sp,#8]
341
342	cbz	$padbit,.Lstore_base2_64_neon
343
344	and	x10,$h0,#0x03ffffff	// base 2^64 -> base 2^26
345	ubfx	x11,$h0,#26,#26
346	extr	x12,$h1,$h0,#52
347	and	x12,x12,#0x03ffffff
348	ubfx	x13,$h1,#14,#26
349	extr	x14,$h2,$h1,#40
350
351	cbnz	$len,.Leven_neon
352
353	stp	w10,w11,[$ctx]		// store hash value base 2^26
354	stp	w12,w13,[$ctx,#8]
355	str	w14,[$ctx,#16]
356	b	.Lno_data_neon
357
358.align	4
359.Lstore_base2_64_neon:
360	stp	$h0,$h1,[$ctx]		// store hash value base 2^64
361	stp	$h2,xzr,[$ctx,#16]	// note that is_base2_26 is zeroed
362	b	.Lno_data_neon
363
364.align	4
365.Lbase2_64_neon:
366	ldp	$r0,$r1,[$ctx,#32]	// load key value
367
368	ldp	$h0,$h1,[$ctx]		// load hash value base 2^64
369	ldr	$h2,[$ctx,#16]
370
371	tst	$len,#31
372	b.eq	.Linit_neon
373
374	ldp	$d0,$d1,[$inp],#16	// load input
375	sub	$len,$len,#16
376	add	$s1,$r1,$r1,lsr#2	// s1 = r1 + (r1 >> 2)
377#ifdef	__ARMEB__
378	rev	$d0,$d0
379	rev	$d1,$d1
380#endif
381	adds	$h0,$h0,$d0		// accumulate input
382	adcs	$h1,$h1,$d1
383	adc	$h2,$h2,$padbit
384
385	bl	poly1305_mult
386
387.Linit_neon:
388	and	x10,$h0,#0x03ffffff	// base 2^64 -> base 2^26
389	ubfx	x11,$h0,#26,#26
390	extr	x12,$h1,$h0,#52
391	and	x12,x12,#0x03ffffff
392	ubfx	x13,$h1,#14,#26
393	extr	x14,$h2,$h1,#40
394
395	stp	d8,d9,[sp,#16]		// meet ABI requirements
396	stp	d10,d11,[sp,#32]
397	stp	d12,d13,[sp,#48]
398	stp	d14,d15,[sp,#64]
399
400	fmov	${H0},x10
401	fmov	${H1},x11
402	fmov	${H2},x12
403	fmov	${H3},x13
404	fmov	${H4},x14
405
406	////////////////////////////////// initialize r^n table
407	mov	$h0,$r0			// r^1
408	add	$s1,$r1,$r1,lsr#2	// s1 = r1 + (r1 >> 2)
409	mov	$h1,$r1
410	mov	$h2,xzr
411	add	$ctx,$ctx,#48+12
412	bl	poly1305_splat
413
414	bl	poly1305_mult		// r^2
415	sub	$ctx,$ctx,#4
416	bl	poly1305_splat
417
418	bl	poly1305_mult		// r^3
419	sub	$ctx,$ctx,#4
420	bl	poly1305_splat
421
422	bl	poly1305_mult		// r^4
423	sub	$ctx,$ctx,#4
424	bl	poly1305_splat
425	ldr	x30,[sp,#8]
426
427	add	$in2,$inp,#32
428	adr	$zeros,.Lzeros
429	subs	$len,$len,#64
430	csel	$in2,$zeros,$in2,lo
431
432	mov	x4,#1
433	str	x4,[$ctx,#-24]		// set is_base2_26
434	sub	$ctx,$ctx,#48		// restore original $ctx
435	b	.Ldo_neon
436
437.align	4
438.Leven_neon:
439	add	$in2,$inp,#32
440	adr	$zeros,.Lzeros
441	subs	$len,$len,#64
442	csel	$in2,$zeros,$in2,lo
443
444	stp	d8,d9,[sp,#16]		// meet ABI requirements
445	stp	d10,d11,[sp,#32]
446	stp	d12,d13,[sp,#48]
447	stp	d14,d15,[sp,#64]
448
449	fmov	${H0},x10
450	fmov	${H1},x11
451	fmov	${H2},x12
452	fmov	${H3},x13
453	fmov	${H4},x14
454
455.Ldo_neon:
456	ldp	x8,x12,[$in2],#16	// inp[2:3] (or zero)
457	ldp	x9,x13,[$in2],#48
458
459	lsl	$padbit,$padbit,#24
460	add	x15,$ctx,#48
461
462#ifdef	__ARMEB__
463	rev	x8,x8
464	rev	x12,x12
465	rev	x9,x9
466	rev	x13,x13
467#endif
468	and	x4,x8,#0x03ffffff	// base 2^64 -> base 2^26
469	and	x5,x9,#0x03ffffff
470	ubfx	x6,x8,#26,#26
471	ubfx	x7,x9,#26,#26
472	add	x4,x4,x5,lsl#32		// bfi	x4,x5,#32,#32
473	extr	x8,x12,x8,#52
474	extr	x9,x13,x9,#52
475	add	x6,x6,x7,lsl#32		// bfi	x6,x7,#32,#32
476	fmov	$IN23_0,x4
477	and	x8,x8,#0x03ffffff
478	and	x9,x9,#0x03ffffff
479	ubfx	x10,x12,#14,#26
480	ubfx	x11,x13,#14,#26
481	add	x12,$padbit,x12,lsr#40
482	add	x13,$padbit,x13,lsr#40
483	add	x8,x8,x9,lsl#32		// bfi	x8,x9,#32,#32
484	fmov	$IN23_1,x6
485	add	x10,x10,x11,lsl#32	// bfi	x10,x11,#32,#32
486	add	x12,x12,x13,lsl#32	// bfi	x12,x13,#32,#32
487	fmov	$IN23_2,x8
488	fmov	$IN23_3,x10
489	fmov	$IN23_4,x12
490
491	ldp	x8,x12,[$inp],#16	// inp[0:1]
492	ldp	x9,x13,[$inp],#48
493
494	ld1	{$R0,$R1,$S1,$R2},[x15],#64
495	ld1	{$S2,$R3,$S3,$R4},[x15],#64
496	ld1	{$S4},[x15]
497
498#ifdef	__ARMEB__
499	rev	x8,x8
500	rev	x12,x12
501	rev	x9,x9
502	rev	x13,x13
503#endif
504	and	x4,x8,#0x03ffffff	// base 2^64 -> base 2^26
505	and	x5,x9,#0x03ffffff
506	ubfx	x6,x8,#26,#26
507	ubfx	x7,x9,#26,#26
508	add	x4,x4,x5,lsl#32		// bfi	x4,x5,#32,#32
509	extr	x8,x12,x8,#52
510	extr	x9,x13,x9,#52
511	add	x6,x6,x7,lsl#32		// bfi	x6,x7,#32,#32
512	fmov	$IN01_0,x4
513	and	x8,x8,#0x03ffffff
514	and	x9,x9,#0x03ffffff
515	ubfx	x10,x12,#14,#26
516	ubfx	x11,x13,#14,#26
517	add	x12,$padbit,x12,lsr#40
518	add	x13,$padbit,x13,lsr#40
519	add	x8,x8,x9,lsl#32		// bfi	x8,x9,#32,#32
520	fmov	$IN01_1,x6
521	add	x10,x10,x11,lsl#32	// bfi	x10,x11,#32,#32
522	add	x12,x12,x13,lsl#32	// bfi	x12,x13,#32,#32
523	movi	$MASK.2d,#-1
524	fmov	$IN01_2,x8
525	fmov	$IN01_3,x10
526	fmov	$IN01_4,x12
527	ushr	$MASK.2d,$MASK.2d,#38
528
529	b.ls	.Lskip_loop
530
531.align	4
532.Loop_neon:
533	////////////////////////////////////////////////////////////////
534	// ((inp[0]*r^4+inp[2]*r^2+inp[4])*r^4+inp[6]*r^2
535	// ((inp[1]*r^4+inp[3]*r^2+inp[5])*r^3+inp[7]*r
536	//   \___________________/
537	// ((inp[0]*r^4+inp[2]*r^2+inp[4])*r^4+inp[6]*r^2+inp[8])*r^2
538	// ((inp[1]*r^4+inp[3]*r^2+inp[5])*r^4+inp[7]*r^2+inp[9])*r
539	//   \___________________/ \____________________/
540	//
541	// Note that we start with inp[2:3]*r^2. This is because it
542	// doesn't depend on reduction in previous iteration.
543	////////////////////////////////////////////////////////////////
544	// d4 = h0*r4 + h1*r3   + h2*r2   + h3*r1   + h4*r0
545	// d3 = h0*r3 + h1*r2   + h2*r1   + h3*r0   + h4*5*r4
546	// d2 = h0*r2 + h1*r1   + h2*r0   + h3*5*r4 + h4*5*r3
547	// d1 = h0*r1 + h1*r0   + h2*5*r4 + h3*5*r3 + h4*5*r2
548	// d0 = h0*r0 + h1*5*r4 + h2*5*r3 + h3*5*r2 + h4*5*r1
549
550	subs	$len,$len,#64
551	umull	$ACC4,$IN23_0,${R4}[2]
552	csel	$in2,$zeros,$in2,lo
553	umull	$ACC3,$IN23_0,${R3}[2]
554	umull	$ACC2,$IN23_0,${R2}[2]
555	 ldp	x8,x12,[$in2],#16	// inp[2:3] (or zero)
556	umull	$ACC1,$IN23_0,${R1}[2]
557	 ldp	x9,x13,[$in2],#48
558	umull	$ACC0,$IN23_0,${R0}[2]
559#ifdef	__ARMEB__
560	 rev	x8,x8
561	 rev	x12,x12
562	 rev	x9,x9
563	 rev	x13,x13
564#endif
565
566	umlal	$ACC4,$IN23_1,${R3}[2]
567	 and	x4,x8,#0x03ffffff	// base 2^64 -> base 2^26
568	umlal	$ACC3,$IN23_1,${R2}[2]
569	 and	x5,x9,#0x03ffffff
570	umlal	$ACC2,$IN23_1,${R1}[2]
571	 ubfx	x6,x8,#26,#26
572	umlal	$ACC1,$IN23_1,${R0}[2]
573	 ubfx	x7,x9,#26,#26
574	umlal	$ACC0,$IN23_1,${S4}[2]
575	 add	x4,x4,x5,lsl#32		// bfi	x4,x5,#32,#32
576
577	umlal	$ACC4,$IN23_2,${R2}[2]
578	 extr	x8,x12,x8,#52
579	umlal	$ACC3,$IN23_2,${R1}[2]
580	 extr	x9,x13,x9,#52
581	umlal	$ACC2,$IN23_2,${R0}[2]
582	 add	x6,x6,x7,lsl#32		// bfi	x6,x7,#32,#32
583	umlal	$ACC1,$IN23_2,${S4}[2]
584	 fmov	$IN23_0,x4
585	umlal	$ACC0,$IN23_2,${S3}[2]
586	 and	x8,x8,#0x03ffffff
587
588	umlal	$ACC4,$IN23_3,${R1}[2]
589	 and	x9,x9,#0x03ffffff
590	umlal	$ACC3,$IN23_3,${R0}[2]
591	 ubfx	x10,x12,#14,#26
592	umlal	$ACC2,$IN23_3,${S4}[2]
593	 ubfx	x11,x13,#14,#26
594	umlal	$ACC1,$IN23_3,${S3}[2]
595	 add	x8,x8,x9,lsl#32		// bfi	x8,x9,#32,#32
596	umlal	$ACC0,$IN23_3,${S2}[2]
597	 fmov	$IN23_1,x6
598
599	add	$IN01_2,$IN01_2,$H2
600	 add	x12,$padbit,x12,lsr#40
601	umlal	$ACC4,$IN23_4,${R0}[2]
602	 add	x13,$padbit,x13,lsr#40
603	umlal	$ACC3,$IN23_4,${S4}[2]
604	 add	x10,x10,x11,lsl#32	// bfi	x10,x11,#32,#32
605	umlal	$ACC2,$IN23_4,${S3}[2]
606	 add	x12,x12,x13,lsl#32	// bfi	x12,x13,#32,#32
607	umlal	$ACC1,$IN23_4,${S2}[2]
608	 fmov	$IN23_2,x8
609	umlal	$ACC0,$IN23_4,${S1}[2]
610	 fmov	$IN23_3,x10
611
612	////////////////////////////////////////////////////////////////
613	// (hash+inp[0:1])*r^4 and accumulate
614
615	add	$IN01_0,$IN01_0,$H0
616	 fmov	$IN23_4,x12
617	umlal	$ACC3,$IN01_2,${R1}[0]
618	 ldp	x8,x12,[$inp],#16	// inp[0:1]
619	umlal	$ACC0,$IN01_2,${S3}[0]
620	 ldp	x9,x13,[$inp],#48
621	umlal	$ACC4,$IN01_2,${R2}[0]
622	umlal	$ACC1,$IN01_2,${S4}[0]
623	umlal	$ACC2,$IN01_2,${R0}[0]
624#ifdef	__ARMEB__
625	 rev	x8,x8
626	 rev	x12,x12
627	 rev	x9,x9
628	 rev	x13,x13
629#endif
630
631	add	$IN01_1,$IN01_1,$H1
632	umlal	$ACC3,$IN01_0,${R3}[0]
633	umlal	$ACC4,$IN01_0,${R4}[0]
634	 and	x4,x8,#0x03ffffff	// base 2^64 -> base 2^26
635	umlal	$ACC2,$IN01_0,${R2}[0]
636	 and	x5,x9,#0x03ffffff
637	umlal	$ACC0,$IN01_0,${R0}[0]
638	 ubfx	x6,x8,#26,#26
639	umlal	$ACC1,$IN01_0,${R1}[0]
640	 ubfx	x7,x9,#26,#26
641
642	add	$IN01_3,$IN01_3,$H3
643	 add	x4,x4,x5,lsl#32		// bfi	x4,x5,#32,#32
644	umlal	$ACC3,$IN01_1,${R2}[0]
645	 extr	x8,x12,x8,#52
646	umlal	$ACC4,$IN01_1,${R3}[0]
647	 extr	x9,x13,x9,#52
648	umlal	$ACC0,$IN01_1,${S4}[0]
649	 add	x6,x6,x7,lsl#32		// bfi	x6,x7,#32,#32
650	umlal	$ACC2,$IN01_1,${R1}[0]
651	 fmov	$IN01_0,x4
652	umlal	$ACC1,$IN01_1,${R0}[0]
653	 and	x8,x8,#0x03ffffff
654
655	add	$IN01_4,$IN01_4,$H4
656	 and	x9,x9,#0x03ffffff
657	umlal	$ACC3,$IN01_3,${R0}[0]
658	 ubfx	x10,x12,#14,#26
659	umlal	$ACC0,$IN01_3,${S2}[0]
660	 ubfx	x11,x13,#14,#26
661	umlal	$ACC4,$IN01_3,${R1}[0]
662	 add	x8,x8,x9,lsl#32		// bfi	x8,x9,#32,#32
663	umlal	$ACC1,$IN01_3,${S3}[0]
664	 fmov	$IN01_1,x6
665	umlal	$ACC2,$IN01_3,${S4}[0]
666	 add	x12,$padbit,x12,lsr#40
667
668	umlal	$ACC3,$IN01_4,${S4}[0]
669	 add	x13,$padbit,x13,lsr#40
670	umlal	$ACC0,$IN01_4,${S1}[0]
671	 add	x10,x10,x11,lsl#32	// bfi	x10,x11,#32,#32
672	umlal	$ACC4,$IN01_4,${R0}[0]
673	 add	x12,x12,x13,lsl#32	// bfi	x12,x13,#32,#32
674	umlal	$ACC1,$IN01_4,${S2}[0]
675	 fmov	$IN01_2,x8
676	umlal	$ACC2,$IN01_4,${S3}[0]
677	 fmov	$IN01_3,x10
678	 fmov	$IN01_4,x12
679
680	/////////////////////////////////////////////////////////////////
681	// lazy reduction as discussed in "NEON crypto" by D.J. Bernstein
682	// and P. Schwabe
683	//
684	// [see discussion in poly1305-armv4 module]
685
686	ushr	$T0.2d,$ACC3,#26
687	xtn	$H3,$ACC3
688	 ushr	$T1.2d,$ACC0,#26
689	 and	$ACC0,$ACC0,$MASK.2d
690	add	$ACC4,$ACC4,$T0.2d	// h3 -> h4
691	bic	$H3,#0xfc,lsl#24	// &=0x03ffffff
692	 add	$ACC1,$ACC1,$T1.2d	// h0 -> h1
693
694	ushr	$T0.2d,$ACC4,#26
695	xtn	$H4,$ACC4
696	 ushr	$T1.2d,$ACC1,#26
697	 xtn	$H1,$ACC1
698	bic	$H4,#0xfc,lsl#24
699	 add	$ACC2,$ACC2,$T1.2d	// h1 -> h2
700
701	add	$ACC0,$ACC0,$T0.2d
702	shl	$T0.2d,$T0.2d,#2
703	 shrn	$T1.2s,$ACC2,#26
704	 xtn	$H2,$ACC2
705	add	$ACC0,$ACC0,$T0.2d	// h4 -> h0
706	 bic	$H1,#0xfc,lsl#24
707	 add	$H3,$H3,$T1.2s		// h2 -> h3
708	 bic	$H2,#0xfc,lsl#24
709
710	shrn	$T0.2s,$ACC0,#26
711	xtn	$H0,$ACC0
712	 ushr	$T1.2s,$H3,#26
713	 bic	$H3,#0xfc,lsl#24
714	 bic	$H0,#0xfc,lsl#24
715	add	$H1,$H1,$T0.2s		// h0 -> h1
716	 add	$H4,$H4,$T1.2s		// h3 -> h4
717
718	b.hi	.Loop_neon
719
720.Lskip_loop:
721	dup	$IN23_2,${IN23_2}[0]
722	add	$IN01_2,$IN01_2,$H2
723
724	////////////////////////////////////////////////////////////////
725	// multiply (inp[0:1]+hash) or inp[2:3] by r^2:r^1
726
727	adds	$len,$len,#32
728	b.ne	.Long_tail
729
730	dup	$IN23_2,${IN01_2}[0]
731	add	$IN23_0,$IN01_0,$H0
732	add	$IN23_3,$IN01_3,$H3
733	add	$IN23_1,$IN01_1,$H1
734	add	$IN23_4,$IN01_4,$H4
735
736.Long_tail:
737	dup	$IN23_0,${IN23_0}[0]
738	umull2	$ACC0,$IN23_2,${S3}
739	umull2	$ACC3,$IN23_2,${R1}
740	umull2	$ACC4,$IN23_2,${R2}
741	umull2	$ACC2,$IN23_2,${R0}
742	umull2	$ACC1,$IN23_2,${S4}
743
744	dup	$IN23_1,${IN23_1}[0]
745	umlal2	$ACC0,$IN23_0,${R0}
746	umlal2	$ACC2,$IN23_0,${R2}
747	umlal2	$ACC3,$IN23_0,${R3}
748	umlal2	$ACC4,$IN23_0,${R4}
749	umlal2	$ACC1,$IN23_0,${R1}
750
751	dup	$IN23_3,${IN23_3}[0]
752	umlal2	$ACC0,$IN23_1,${S4}
753	umlal2	$ACC3,$IN23_1,${R2}
754	umlal2	$ACC2,$IN23_1,${R1}
755	umlal2	$ACC4,$IN23_1,${R3}
756	umlal2	$ACC1,$IN23_1,${R0}
757
758	dup	$IN23_4,${IN23_4}[0]
759	umlal2	$ACC3,$IN23_3,${R0}
760	umlal2	$ACC4,$IN23_3,${R1}
761	umlal2	$ACC0,$IN23_3,${S2}
762	umlal2	$ACC1,$IN23_3,${S3}
763	umlal2	$ACC2,$IN23_3,${S4}
764
765	umlal2	$ACC3,$IN23_4,${S4}
766	umlal2	$ACC0,$IN23_4,${S1}
767	umlal2	$ACC4,$IN23_4,${R0}
768	umlal2	$ACC1,$IN23_4,${S2}
769	umlal2	$ACC2,$IN23_4,${S3}
770
771	b.eq	.Lshort_tail
772
773	////////////////////////////////////////////////////////////////
774	// (hash+inp[0:1])*r^4:r^3 and accumulate
775
776	add	$IN01_0,$IN01_0,$H0
777	umlal	$ACC3,$IN01_2,${R1}
778	umlal	$ACC0,$IN01_2,${S3}
779	umlal	$ACC4,$IN01_2,${R2}
780	umlal	$ACC1,$IN01_2,${S4}
781	umlal	$ACC2,$IN01_2,${R0}
782
783	add	$IN01_1,$IN01_1,$H1
784	umlal	$ACC3,$IN01_0,${R3}
785	umlal	$ACC0,$IN01_0,${R0}
786	umlal	$ACC4,$IN01_0,${R4}
787	umlal	$ACC1,$IN01_0,${R1}
788	umlal	$ACC2,$IN01_0,${R2}
789
790	add	$IN01_3,$IN01_3,$H3
791	umlal	$ACC3,$IN01_1,${R2}
792	umlal	$ACC0,$IN01_1,${S4}
793	umlal	$ACC4,$IN01_1,${R3}
794	umlal	$ACC1,$IN01_1,${R0}
795	umlal	$ACC2,$IN01_1,${R1}
796
797	add	$IN01_4,$IN01_4,$H4
798	umlal	$ACC3,$IN01_3,${R0}
799	umlal	$ACC0,$IN01_3,${S2}
800	umlal	$ACC4,$IN01_3,${R1}
801	umlal	$ACC1,$IN01_3,${S3}
802	umlal	$ACC2,$IN01_3,${S4}
803
804	umlal	$ACC3,$IN01_4,${S4}
805	umlal	$ACC0,$IN01_4,${S1}
806	umlal	$ACC4,$IN01_4,${R0}
807	umlal	$ACC1,$IN01_4,${S2}
808	umlal	$ACC2,$IN01_4,${S3}
809
810.Lshort_tail:
811	////////////////////////////////////////////////////////////////
812	// horizontal add
813
814	addp	$ACC3,$ACC3,$ACC3
815	 ldp	d8,d9,[sp,#16]		// meet ABI requirements
816	addp	$ACC0,$ACC0,$ACC0
817	 ldp	d10,d11,[sp,#32]
818	addp	$ACC4,$ACC4,$ACC4
819	 ldp	d12,d13,[sp,#48]
820	addp	$ACC1,$ACC1,$ACC1
821	 ldp	d14,d15,[sp,#64]
822	addp	$ACC2,$ACC2,$ACC2
823
824	////////////////////////////////////////////////////////////////
825	// lazy reduction, but without narrowing
826
827	ushr	$T0.2d,$ACC3,#26
828	and	$ACC3,$ACC3,$MASK.2d
829	 ushr	$T1.2d,$ACC0,#26
830	 and	$ACC0,$ACC0,$MASK.2d
831
832	add	$ACC4,$ACC4,$T0.2d	// h3 -> h4
833	 add	$ACC1,$ACC1,$T1.2d	// h0 -> h1
834
835	ushr	$T0.2d,$ACC4,#26
836	and	$ACC4,$ACC4,$MASK.2d
837	 ushr	$T1.2d,$ACC1,#26
838	 and	$ACC1,$ACC1,$MASK.2d
839	 add	$ACC2,$ACC2,$T1.2d	// h1 -> h2
840
841	add	$ACC0,$ACC0,$T0.2d
842	shl	$T0.2d,$T0.2d,#2
843	 ushr	$T1.2d,$ACC2,#26
844	 and	$ACC2,$ACC2,$MASK.2d
845	add	$ACC0,$ACC0,$T0.2d	// h4 -> h0
846	 add	$ACC3,$ACC3,$T1.2d	// h2 -> h3
847
848	ushr	$T0.2d,$ACC0,#26
849	and	$ACC0,$ACC0,$MASK.2d
850	 ushr	$T1.2d,$ACC3,#26
851	 and	$ACC3,$ACC3,$MASK.2d
852	add	$ACC1,$ACC1,$T0.2d	// h0 -> h1
853	 add	$ACC4,$ACC4,$T1.2d	// h3 -> h4
854
855	////////////////////////////////////////////////////////////////
856	// write the result, can be partially reduced
857
858	st4	{$ACC0,$ACC1,$ACC2,$ACC3}[0],[$ctx],#16
859	st1	{$ACC4}[0],[$ctx]
860
861.Lno_data_neon:
862	ldr	x29,[sp],#80
863	ret
864.size	poly1305_blocks_neon,.-poly1305_blocks_neon
865
866.type	poly1305_emit_neon,%function
867.align	5
868poly1305_emit_neon:
869	ldr	$is_base2_26,[$ctx,#24]
870	cbz	$is_base2_26,poly1305_emit
871
872	ldp	w10,w11,[$ctx]		// load hash value base 2^26
873	ldp	w12,w13,[$ctx,#8]
874	ldr	w14,[$ctx,#16]
875
876	add	$h0,x10,x11,lsl#26	// base 2^26 -> base 2^64
877	lsr	$h1,x12,#12
878	adds	$h0,$h0,x12,lsl#52
879	add	$h1,$h1,x13,lsl#14
880	adc	$h1,$h1,xzr
881	lsr	$h2,x14,#24
882	adds	$h1,$h1,x14,lsl#40
883	adc	$h2,$h2,xzr		// can be partially reduced...
884
885	ldp	$t0,$t1,[$nonce]	// load nonce
886
887	and	$d0,$h2,#-4		// ... so reduce
888	add	$d0,$d0,$h2,lsr#2
889	and	$h2,$h2,#3
890	adds	$h0,$h0,$d0
891	adcs	$h1,$h1,xzr
892	adc	$h2,$h2,xzr
893
894	adds	$d0,$h0,#5		// compare to modulus
895	adcs	$d1,$h1,xzr
896	adc	$d2,$h2,xzr
897
898	tst	$d2,#-4			// see if it's carried/borrowed
899
900	csel	$h0,$h0,$d0,eq
901	csel	$h1,$h1,$d1,eq
902
903#ifdef	__ARMEB__
904	ror	$t0,$t0,#32		// flip nonce words
905	ror	$t1,$t1,#32
906#endif
907	adds	$h0,$h0,$t0		// accumulate nonce
908	adc	$h1,$h1,$t1
909#ifdef	__ARMEB__
910	rev	$h0,$h0			// flip output bytes
911	rev	$h1,$h1
912#endif
913	stp	$h0,$h1,[$mac]		// write result
914
915	ret
916.size	poly1305_emit_neon,.-poly1305_emit_neon
917
918.align	5
919.Lzeros:
920.long	0,0,0,0,0,0,0,0
921.LOPENSSL_armcap_P:
922#ifdef	__ILP32__
923.long	OPENSSL_armcap_P-.
924#else
925.quad	OPENSSL_armcap_P-.
926#endif
927.asciz	"Poly1305 for ARMv8, CRYPTOGAMS by <appro\@openssl.org>"
928.align	2
929___
930
931foreach (split("\n",$code)) {
932	s/\b(shrn\s+v[0-9]+)\.[24]d/$1.2s/			or
933	s/\b(fmov\s+)v([0-9]+)[^,]*,\s*x([0-9]+)/$1d$2,x$3/	or
934	(m/\bdup\b/ and (s/\.[24]s/.2d/g or 1))			or
935	(m/\b(eor|and)/ and (s/\.[248][sdh]/.16b/g or 1))	or
936	(m/\bum(ul|la)l\b/ and (s/\.4s/.2s/g or 1))		or
937	(m/\bum(ul|la)l2\b/ and (s/\.2s/.4s/g or 1))		or
938	(m/\bst[1-4]\s+{[^}]+}\[/ and (s/\.[24]d/.s/g or 1));
939
940	s/\.[124]([sd])\[/.$1\[/;
941
942	print $_,"\n";
943}
944close STDOUT;
945