xref: /linux/lib/crypto/arm64/poly1305-armv8.pl (revision 03f76ddff5b04a808ae16c06418460151e2fdd4b) 
1#!/usr/bin/env perl
2# SPDX-License-Identifier: GPL-1.0+ OR BSD-3-Clause
3#
4# ====================================================================
5# Written by Andy Polyakov, @dot-asm, initially for the OpenSSL
6# project.
7# ====================================================================
8#
9# This module implements Poly1305 hash for ARMv8.
10#
11# June 2015
12#
13# Numbers are cycles per processed byte with poly1305_blocks alone.
14#
15#		IALU/gcc-4.9	NEON
16#
17# Apple A7	1.86/+5%	0.72
18# Cortex-A53	2.69/+58%	1.47
19# Cortex-A57	2.70/+7%	1.14
20# Denver	1.64/+50%	1.18(*)
21# X-Gene	2.13/+68%	2.27
22# Mongoose	1.77/+75%	1.12
23# Kryo		2.70/+55%	1.13
24# ThunderX2	1.17/+95%	1.36
25#
26# (*)	estimate based on resources availability is less than 1.0,
27#	i.e. measured result is worse than expected, presumably binary
28#	translator is not almighty;
29
30$flavour=shift;
31$output=shift;
32
33if ($flavour && $flavour ne "void") {
34    $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
35    ( $xlate="${dir}arm-xlate.pl" and -f $xlate ) or
36    ( $xlate="${dir}../../perlasm/arm-xlate.pl" and -f $xlate) or
37    die "can't locate arm-xlate.pl";
38
39    open STDOUT,"| \"$^X\" $xlate $flavour $output";
40} else {
41    open STDOUT,">$output";
42}
43
44my ($ctx,$inp,$len,$padbit) = map("x$_",(0..3));
45my ($mac,$nonce)=($inp,$len);
46
47my ($h0,$h1,$h2,$r0,$r1,$s1,$t0,$t1,$d0,$d1,$d2) = map("x$_",(4..14));
48
49$code.=<<___;
50#ifndef __KERNEL__
51# include "arm_arch.h"
52.extern	OPENSSL_armcap_P
53#else
54# define poly1305_init   poly1305_block_init
55# define poly1305_blocks poly1305_blocks_arm64
56#endif
57
58.text
59
60// forward "declarations" are required for Apple
61.globl	poly1305_blocks
62.globl	poly1305_emit
63
64.globl	poly1305_init
65.type	poly1305_init,%function
66.align	5
67poly1305_init:
68	cmp	$inp,xzr
69	stp	xzr,xzr,[$ctx]		// zero hash value
70	stp	xzr,xzr,[$ctx,#16]	// [along with is_base2_26]
71
72	csel	x0,xzr,x0,eq
73	b.eq	.Lno_key
74
75#ifndef	__KERNEL__
76	adrp	x17,OPENSSL_armcap_P
77	ldr	w17,[x17,#:lo12:OPENSSL_armcap_P]
78#endif
79
80	ldp	$r0,$r1,[$inp]		// load key
81	mov	$s1,#0xfffffffc0fffffff
82	movk	$s1,#0x0fff,lsl#48
83#ifdef	__AARCH64EB__
84	rev	$r0,$r0			// flip bytes
85	rev	$r1,$r1
86#endif
87	and	$r0,$r0,$s1		// &=0ffffffc0fffffff
88	and	$s1,$s1,#-4
89	and	$r1,$r1,$s1		// &=0ffffffc0ffffffc
90	mov	w#$s1,#-1
91	stp	$r0,$r1,[$ctx,#32]	// save key value
92	str	w#$s1,[$ctx,#48]	// impossible key power value
93
94#ifndef	__KERNEL__
95	tst	w17,#ARMV7_NEON
96
97	adr	$d0,.Lpoly1305_blocks
98	adr	$r0,.Lpoly1305_blocks_neon
99	adr	$d1,.Lpoly1305_emit
100
101	csel	$d0,$d0,$r0,eq
102
103# ifdef	__ILP32__
104	stp	w#$d0,w#$d1,[$len]
105# else
106	stp	$d0,$d1,[$len]
107# endif
108#endif
109	mov	x0,#1
110.Lno_key:
111	ret
112.size	poly1305_init,.-poly1305_init
113
114.type	poly1305_blocks,%function
115.align	5
116poly1305_blocks:
117.Lpoly1305_blocks:
118	ands	$len,$len,#-16
119	b.eq	.Lno_data
120
121	ldp	$h0,$h1,[$ctx]		// load hash value
122	ldp	$h2,x17,[$ctx,#16]	// [along with is_base2_26]
123	ldp	$r0,$r1,[$ctx,#32]	// load key value
124
125#ifdef	__AARCH64EB__
126	lsr	$d0,$h0,#32
127	mov	w#$d1,w#$h0
128	lsr	$d2,$h1,#32
129	mov	w15,w#$h1
130	lsr	x16,$h2,#32
131#else
132	mov	w#$d0,w#$h0
133	lsr	$d1,$h0,#32
134	mov	w#$d2,w#$h1
135	lsr	x15,$h1,#32
136	mov	w16,w#$h2
137#endif
138
139	add	$d0,$d0,$d1,lsl#26	// base 2^26 -> base 2^64
140	lsr	$d1,$d2,#12
141	adds	$d0,$d0,$d2,lsl#52
142	add	$d1,$d1,x15,lsl#14
143	adc	$d1,$d1,xzr
144	lsr	$d2,x16,#24
145	adds	$d1,$d1,x16,lsl#40
146	adc	$d2,$d2,xzr
147
148	cmp	x17,#0			// is_base2_26?
149	add	$s1,$r1,$r1,lsr#2	// s1 = r1 + (r1 >> 2)
150	csel	$h0,$h0,$d0,eq		// choose between radixes
151	csel	$h1,$h1,$d1,eq
152	csel	$h2,$h2,$d2,eq
153
154.Loop:
155	ldp	$t0,$t1,[$inp],#16	// load input
156	sub	$len,$len,#16
157#ifdef	__AARCH64EB__
158	rev	$t0,$t0
159	rev	$t1,$t1
160#endif
161	adds	$h0,$h0,$t0		// accumulate input
162	adcs	$h1,$h1,$t1
163
164	mul	$d0,$h0,$r0		// h0*r0
165	adc	$h2,$h2,$padbit
166	umulh	$d1,$h0,$r0
167
168	mul	$t0,$h1,$s1		// h1*5*r1
169	umulh	$t1,$h1,$s1
170
171	adds	$d0,$d0,$t0
172	mul	$t0,$h0,$r1		// h0*r1
173	adc	$d1,$d1,$t1
174	umulh	$d2,$h0,$r1
175
176	adds	$d1,$d1,$t0
177	mul	$t0,$h1,$r0		// h1*r0
178	adc	$d2,$d2,xzr
179	umulh	$t1,$h1,$r0
180
181	adds	$d1,$d1,$t0
182	mul	$t0,$h2,$s1		// h2*5*r1
183	adc	$d2,$d2,$t1
184	mul	$t1,$h2,$r0		// h2*r0
185
186	adds	$d1,$d1,$t0
187	adc	$d2,$d2,$t1
188
189	and	$t0,$d2,#-4		// final reduction
190	and	$h2,$d2,#3
191	add	$t0,$t0,$d2,lsr#2
192	adds	$h0,$d0,$t0
193	adcs	$h1,$d1,xzr
194	adc	$h2,$h2,xzr
195
196	cbnz	$len,.Loop
197
198	stp	$h0,$h1,[$ctx]		// store hash value
199	stp	$h2,xzr,[$ctx,#16]	// [and clear is_base2_26]
200
201.Lno_data:
202	ret
203.size	poly1305_blocks,.-poly1305_blocks
204
205.type	poly1305_emit,%function
206.align	5
207poly1305_emit:
208.Lpoly1305_emit:
209	ldp	$h0,$h1,[$ctx]		// load hash base 2^64
210	ldp	$h2,$r0,[$ctx,#16]	// [along with is_base2_26]
211	ldp	$t0,$t1,[$nonce]	// load nonce
212
213#ifdef	__AARCH64EB__
214	lsr	$d0,$h0,#32
215	mov	w#$d1,w#$h0
216	lsr	$d2,$h1,#32
217	mov	w15,w#$h1
218	lsr	x16,$h2,#32
219#else
220	mov	w#$d0,w#$h0
221	lsr	$d1,$h0,#32
222	mov	w#$d2,w#$h1
223	lsr	x15,$h1,#32
224	mov	w16,w#$h2
225#endif
226
227	add	$d0,$d0,$d1,lsl#26	// base 2^26 -> base 2^64
228	lsr	$d1,$d2,#12
229	adds	$d0,$d0,$d2,lsl#52
230	add	$d1,$d1,x15,lsl#14
231	adc	$d1,$d1,xzr
232	lsr	$d2,x16,#24
233	adds	$d1,$d1,x16,lsl#40
234	adc	$d2,$d2,xzr
235
236	cmp	$r0,#0			// is_base2_26?
237	csel	$h0,$h0,$d0,eq		// choose between radixes
238	csel	$h1,$h1,$d1,eq
239	csel	$h2,$h2,$d2,eq
240
241	adds	$d0,$h0,#5		// compare to modulus
242	adcs	$d1,$h1,xzr
243	adc	$d2,$h2,xzr
244
245	tst	$d2,#-4			// see if it's carried/borrowed
246
247	csel	$h0,$h0,$d0,eq
248	csel	$h1,$h1,$d1,eq
249
250#ifdef	__AARCH64EB__
251	ror	$t0,$t0,#32		// flip nonce words
252	ror	$t1,$t1,#32
253#endif
254	adds	$h0,$h0,$t0		// accumulate nonce
255	adc	$h1,$h1,$t1
256#ifdef	__AARCH64EB__
257	rev	$h0,$h0			// flip output bytes
258	rev	$h1,$h1
259#endif
260	stp	$h0,$h1,[$mac]		// write result
261
262	ret
263.size	poly1305_emit,.-poly1305_emit
264___
265my ($R0,$R1,$S1,$R2,$S2,$R3,$S3,$R4,$S4) = map("v$_.4s",(0..8));
266my ($IN01_0,$IN01_1,$IN01_2,$IN01_3,$IN01_4) = map("v$_.2s",(9..13));
267my ($IN23_0,$IN23_1,$IN23_2,$IN23_3,$IN23_4) = map("v$_.2s",(14..18));
268my ($ACC0,$ACC1,$ACC2,$ACC3,$ACC4) = map("v$_.2d",(19..23));
269my ($H0,$H1,$H2,$H3,$H4) = map("v$_.2s",(24..28));
270my ($T0,$T1,$MASK) = map("v$_",(29..31));
271
272my ($in2,$zeros)=("x16","x17");
273my $is_base2_26 = $zeros;		# borrow
274
275$code.=<<___;
276.type	poly1305_mult,%function
277.align	5
278poly1305_mult:
279	mul	$d0,$h0,$r0		// h0*r0
280	umulh	$d1,$h0,$r0
281
282	mul	$t0,$h1,$s1		// h1*5*r1
283	umulh	$t1,$h1,$s1
284
285	adds	$d0,$d0,$t0
286	mul	$t0,$h0,$r1		// h0*r1
287	adc	$d1,$d1,$t1
288	umulh	$d2,$h0,$r1
289
290	adds	$d1,$d1,$t0
291	mul	$t0,$h1,$r0		// h1*r0
292	adc	$d2,$d2,xzr
293	umulh	$t1,$h1,$r0
294
295	adds	$d1,$d1,$t0
296	mul	$t0,$h2,$s1		// h2*5*r1
297	adc	$d2,$d2,$t1
298	mul	$t1,$h2,$r0		// h2*r0
299
300	adds	$d1,$d1,$t0
301	adc	$d2,$d2,$t1
302
303	and	$t0,$d2,#-4		// final reduction
304	and	$h2,$d2,#3
305	add	$t0,$t0,$d2,lsr#2
306	adds	$h0,$d0,$t0
307	adcs	$h1,$d1,xzr
308	adc	$h2,$h2,xzr
309
310	ret
311.size	poly1305_mult,.-poly1305_mult
312
313.type	poly1305_splat,%function
314.align	4
315poly1305_splat:
316	and	x12,$h0,#0x03ffffff	// base 2^64 -> base 2^26
317	ubfx	x13,$h0,#26,#26
318	extr	x14,$h1,$h0,#52
319	and	x14,x14,#0x03ffffff
320	ubfx	x15,$h1,#14,#26
321	extr	x16,$h2,$h1,#40
322
323	str	w12,[$ctx,#16*0]	// r0
324	add	w12,w13,w13,lsl#2	// r1*5
325	str	w13,[$ctx,#16*1]	// r1
326	add	w13,w14,w14,lsl#2	// r2*5
327	str	w12,[$ctx,#16*2]	// s1
328	str	w14,[$ctx,#16*3]	// r2
329	add	w14,w15,w15,lsl#2	// r3*5
330	str	w13,[$ctx,#16*4]	// s2
331	str	w15,[$ctx,#16*5]	// r3
332	add	w15,w16,w16,lsl#2	// r4*5
333	str	w14,[$ctx,#16*6]	// s3
334	str	w16,[$ctx,#16*7]	// r4
335	str	w15,[$ctx,#16*8]	// s4
336
337	ret
338.size	poly1305_splat,.-poly1305_splat
339
340#ifdef	__KERNEL__
341.globl	poly1305_blocks_neon
342#endif
343.type	poly1305_blocks_neon,%function
344.align	5
345poly1305_blocks_neon:
346.Lpoly1305_blocks_neon:
347	ldr	$is_base2_26,[$ctx,#24]
348	cmp	$len,#128
349	b.lo	.Lpoly1305_blocks
350
351	.inst	0xd503233f		// paciasp
352	stp	x29,x30,[sp,#-80]!
353	add	x29,sp,#0
354
355	stp	d8,d9,[sp,#16]		// meet ABI requirements
356	stp	d10,d11,[sp,#32]
357	stp	d12,d13,[sp,#48]
358	stp	d14,d15,[sp,#64]
359
360	cbz	$is_base2_26,.Lbase2_64_neon
361
362	ldp	w10,w11,[$ctx]		// load hash value base 2^26
363	ldp	w12,w13,[$ctx,#8]
364	ldr	w14,[$ctx,#16]
365
366	tst	$len,#31
367	b.eq	.Leven_neon
368
369	ldp	$r0,$r1,[$ctx,#32]	// load key value
370
371	add	$h0,x10,x11,lsl#26	// base 2^26 -> base 2^64
372	lsr	$h1,x12,#12
373	adds	$h0,$h0,x12,lsl#52
374	add	$h1,$h1,x13,lsl#14
375	adc	$h1,$h1,xzr
376	lsr	$h2,x14,#24
377	adds	$h1,$h1,x14,lsl#40
378	adc	$d2,$h2,xzr		// can be partially reduced...
379
380	ldp	$d0,$d1,[$inp],#16	// load input
381	sub	$len,$len,#16
382	add	$s1,$r1,$r1,lsr#2	// s1 = r1 + (r1 >> 2)
383
384#ifdef	__AARCH64EB__
385	rev	$d0,$d0
386	rev	$d1,$d1
387#endif
388	adds	$h0,$h0,$d0		// accumulate input
389	adcs	$h1,$h1,$d1
390	adc	$h2,$h2,$padbit
391
392	bl	poly1305_mult
393
394	and	x10,$h0,#0x03ffffff	// base 2^64 -> base 2^26
395	ubfx	x11,$h0,#26,#26
396	extr	x12,$h1,$h0,#52
397	and	x12,x12,#0x03ffffff
398	ubfx	x13,$h1,#14,#26
399	extr	x14,$h2,$h1,#40
400
401	b	.Leven_neon
402
403.align	4
404.Lbase2_64_neon:
405	ldp	$r0,$r1,[$ctx,#32]	// load key value
406
407	ldp	$h0,$h1,[$ctx]		// load hash value base 2^64
408	ldr	$h2,[$ctx,#16]
409
410	tst	$len,#31
411	b.eq	.Linit_neon
412
413	ldp	$d0,$d1,[$inp],#16	// load input
414	sub	$len,$len,#16
415	add	$s1,$r1,$r1,lsr#2	// s1 = r1 + (r1 >> 2)
416#ifdef	__AARCH64EB__
417	rev	$d0,$d0
418	rev	$d1,$d1
419#endif
420	adds	$h0,$h0,$d0		// accumulate input
421	adcs	$h1,$h1,$d1
422	adc	$h2,$h2,$padbit
423
424	bl	poly1305_mult
425
426.Linit_neon:
427	ldr	w17,[$ctx,#48]		// first table element
428	and	x10,$h0,#0x03ffffff	// base 2^64 -> base 2^26
429	ubfx	x11,$h0,#26,#26
430	extr	x12,$h1,$h0,#52
431	and	x12,x12,#0x03ffffff
432	ubfx	x13,$h1,#14,#26
433	extr	x14,$h2,$h1,#40
434
435	cmp	w17,#-1			// is value impossible?
436	b.ne	.Leven_neon
437
438	fmov	${H0},x10
439	fmov	${H1},x11
440	fmov	${H2},x12
441	fmov	${H3},x13
442	fmov	${H4},x14
443
444	////////////////////////////////// initialize r^n table
445	mov	$h0,$r0			// r^1
446	add	$s1,$r1,$r1,lsr#2	// s1 = r1 + (r1 >> 2)
447	mov	$h1,$r1
448	mov	$h2,xzr
449	add	$ctx,$ctx,#48+12
450	bl	poly1305_splat
451
452	bl	poly1305_mult		// r^2
453	sub	$ctx,$ctx,#4
454	bl	poly1305_splat
455
456	bl	poly1305_mult		// r^3
457	sub	$ctx,$ctx,#4
458	bl	poly1305_splat
459
460	bl	poly1305_mult		// r^4
461	sub	$ctx,$ctx,#4
462	bl	poly1305_splat
463	sub	$ctx,$ctx,#48		// restore original $ctx
464	b	.Ldo_neon
465
466.align	4
467.Leven_neon:
468	fmov	${H0},x10
469	fmov	${H1},x11
470	fmov	${H2},x12
471	fmov	${H3},x13
472	fmov	${H4},x14
473
474.Ldo_neon:
475	ldp	x8,x12,[$inp,#32]	// inp[2:3]
476	subs	$len,$len,#64
477	ldp	x9,x13,[$inp,#48]
478	add	$in2,$inp,#96
479	adrp	$zeros,.Lzeros
480	add	$zeros,$zeros,#:lo12:.Lzeros
481
482	lsl	$padbit,$padbit,#24
483	add	x15,$ctx,#48
484
485#ifdef	__AARCH64EB__
486	rev	x8,x8
487	rev	x12,x12
488	rev	x9,x9
489	rev	x13,x13
490#endif
491	and	x4,x8,#0x03ffffff	// base 2^64 -> base 2^26
492	and	x5,x9,#0x03ffffff
493	ubfx	x6,x8,#26,#26
494	ubfx	x7,x9,#26,#26
495	add	x4,x4,x5,lsl#32		// bfi	x4,x5,#32,#32
496	extr	x8,x12,x8,#52
497	extr	x9,x13,x9,#52
498	add	x6,x6,x7,lsl#32		// bfi	x6,x7,#32,#32
499	fmov	$IN23_0,x4
500	and	x8,x8,#0x03ffffff
501	and	x9,x9,#0x03ffffff
502	ubfx	x10,x12,#14,#26
503	ubfx	x11,x13,#14,#26
504	add	x12,$padbit,x12,lsr#40
505	add	x13,$padbit,x13,lsr#40
506	add	x8,x8,x9,lsl#32		// bfi	x8,x9,#32,#32
507	fmov	$IN23_1,x6
508	add	x10,x10,x11,lsl#32	// bfi	x10,x11,#32,#32
509	add	x12,x12,x13,lsl#32	// bfi	x12,x13,#32,#32
510	fmov	$IN23_2,x8
511	fmov	$IN23_3,x10
512	fmov	$IN23_4,x12
513
514	ldp	x8,x12,[$inp],#16	// inp[0:1]
515	ldp	x9,x13,[$inp],#48
516
517	ld1	{$R0,$R1,$S1,$R2},[x15],#64
518	ld1	{$S2,$R3,$S3,$R4},[x15],#64
519	ld1	{$S4},[x15]
520
521#ifdef	__AARCH64EB__
522	rev	x8,x8
523	rev	x12,x12
524	rev	x9,x9
525	rev	x13,x13
526#endif
527	and	x4,x8,#0x03ffffff	// base 2^64 -> base 2^26
528	and	x5,x9,#0x03ffffff
529	ubfx	x6,x8,#26,#26
530	ubfx	x7,x9,#26,#26
531	add	x4,x4,x5,lsl#32		// bfi	x4,x5,#32,#32
532	extr	x8,x12,x8,#52
533	extr	x9,x13,x9,#52
534	add	x6,x6,x7,lsl#32		// bfi	x6,x7,#32,#32
535	fmov	$IN01_0,x4
536	and	x8,x8,#0x03ffffff
537	and	x9,x9,#0x03ffffff
538	ubfx	x10,x12,#14,#26
539	ubfx	x11,x13,#14,#26
540	add	x12,$padbit,x12,lsr#40
541	add	x13,$padbit,x13,lsr#40
542	add	x8,x8,x9,lsl#32		// bfi	x8,x9,#32,#32
543	fmov	$IN01_1,x6
544	add	x10,x10,x11,lsl#32	// bfi	x10,x11,#32,#32
545	add	x12,x12,x13,lsl#32	// bfi	x12,x13,#32,#32
546	movi	$MASK.2d,#-1
547	fmov	$IN01_2,x8
548	fmov	$IN01_3,x10
549	fmov	$IN01_4,x12
550	ushr	$MASK.2d,$MASK.2d,#38
551
552	b.ls	.Lskip_loop
553
554.align	4
555.Loop_neon:
556	////////////////////////////////////////////////////////////////
557	// ((inp[0]*r^4+inp[2]*r^2+inp[4])*r^4+inp[6]*r^2
558	// ((inp[1]*r^4+inp[3]*r^2+inp[5])*r^3+inp[7]*r
559	//   \___________________/
560	// ((inp[0]*r^4+inp[2]*r^2+inp[4])*r^4+inp[6]*r^2+inp[8])*r^2
561	// ((inp[1]*r^4+inp[3]*r^2+inp[5])*r^4+inp[7]*r^2+inp[9])*r
562	//   \___________________/ \____________________/
563	//
564	// Note that we start with inp[2:3]*r^2. This is because it
565	// doesn't depend on reduction in previous iteration.
566	////////////////////////////////////////////////////////////////
567	// d4 = h0*r4 + h1*r3   + h2*r2   + h3*r1   + h4*r0
568	// d3 = h0*r3 + h1*r2   + h2*r1   + h3*r0   + h4*5*r4
569	// d2 = h0*r2 + h1*r1   + h2*r0   + h3*5*r4 + h4*5*r3
570	// d1 = h0*r1 + h1*r0   + h2*5*r4 + h3*5*r3 + h4*5*r2
571	// d0 = h0*r0 + h1*5*r4 + h2*5*r3 + h3*5*r2 + h4*5*r1
572
573	subs	$len,$len,#64
574	umull	$ACC4,$IN23_0,${R4}[2]
575	csel	$in2,$zeros,$in2,lo
576	umull	$ACC3,$IN23_0,${R3}[2]
577	umull	$ACC2,$IN23_0,${R2}[2]
578	 ldp	x8,x12,[$in2],#16	// inp[2:3] (or zero)
579	umull	$ACC1,$IN23_0,${R1}[2]
580	 ldp	x9,x13,[$in2],#48
581	umull	$ACC0,$IN23_0,${R0}[2]
582#ifdef	__AARCH64EB__
583	 rev	x8,x8
584	 rev	x12,x12
585	 rev	x9,x9
586	 rev	x13,x13
587#endif
588
589	umlal	$ACC4,$IN23_1,${R3}[2]
590	 and	x4,x8,#0x03ffffff	// base 2^64 -> base 2^26
591	umlal	$ACC3,$IN23_1,${R2}[2]
592	 and	x5,x9,#0x03ffffff
593	umlal	$ACC2,$IN23_1,${R1}[2]
594	 ubfx	x6,x8,#26,#26
595	umlal	$ACC1,$IN23_1,${R0}[2]
596	 ubfx	x7,x9,#26,#26
597	umlal	$ACC0,$IN23_1,${S4}[2]
598	 add	x4,x4,x5,lsl#32		// bfi	x4,x5,#32,#32
599
600	umlal	$ACC4,$IN23_2,${R2}[2]
601	 extr	x8,x12,x8,#52
602	umlal	$ACC3,$IN23_2,${R1}[2]
603	 extr	x9,x13,x9,#52
604	umlal	$ACC2,$IN23_2,${R0}[2]
605	 add	x6,x6,x7,lsl#32		// bfi	x6,x7,#32,#32
606	umlal	$ACC1,$IN23_2,${S4}[2]
607	 fmov	$IN23_0,x4
608	umlal	$ACC0,$IN23_2,${S3}[2]
609	 and	x8,x8,#0x03ffffff
610
611	umlal	$ACC4,$IN23_3,${R1}[2]
612	 and	x9,x9,#0x03ffffff
613	umlal	$ACC3,$IN23_3,${R0}[2]
614	 ubfx	x10,x12,#14,#26
615	umlal	$ACC2,$IN23_3,${S4}[2]
616	 ubfx	x11,x13,#14,#26
617	umlal	$ACC1,$IN23_3,${S3}[2]
618	 add	x8,x8,x9,lsl#32		// bfi	x8,x9,#32,#32
619	umlal	$ACC0,$IN23_3,${S2}[2]
620	 fmov	$IN23_1,x6
621
622	add	$IN01_2,$IN01_2,$H2
623	 add	x12,$padbit,x12,lsr#40
624	umlal	$ACC4,$IN23_4,${R0}[2]
625	 add	x13,$padbit,x13,lsr#40
626	umlal	$ACC3,$IN23_4,${S4}[2]
627	 add	x10,x10,x11,lsl#32	// bfi	x10,x11,#32,#32
628	umlal	$ACC2,$IN23_4,${S3}[2]
629	 add	x12,x12,x13,lsl#32	// bfi	x12,x13,#32,#32
630	umlal	$ACC1,$IN23_4,${S2}[2]
631	 fmov	$IN23_2,x8
632	umlal	$ACC0,$IN23_4,${S1}[2]
633	 fmov	$IN23_3,x10
634
635	////////////////////////////////////////////////////////////////
636	// (hash+inp[0:1])*r^4 and accumulate
637
638	add	$IN01_0,$IN01_0,$H0
639	 fmov	$IN23_4,x12
640	umlal	$ACC3,$IN01_2,${R1}[0]
641	 ldp	x8,x12,[$inp],#16	// inp[0:1]
642	umlal	$ACC0,$IN01_2,${S3}[0]
643	 ldp	x9,x13,[$inp],#48
644	umlal	$ACC4,$IN01_2,${R2}[0]
645	umlal	$ACC1,$IN01_2,${S4}[0]
646	umlal	$ACC2,$IN01_2,${R0}[0]
647#ifdef	__AARCH64EB__
648	 rev	x8,x8
649	 rev	x12,x12
650	 rev	x9,x9
651	 rev	x13,x13
652#endif
653
654	add	$IN01_1,$IN01_1,$H1
655	umlal	$ACC3,$IN01_0,${R3}[0]
656	umlal	$ACC4,$IN01_0,${R4}[0]
657	 and	x4,x8,#0x03ffffff	// base 2^64 -> base 2^26
658	umlal	$ACC2,$IN01_0,${R2}[0]
659	 and	x5,x9,#0x03ffffff
660	umlal	$ACC0,$IN01_0,${R0}[0]
661	 ubfx	x6,x8,#26,#26
662	umlal	$ACC1,$IN01_0,${R1}[0]
663	 ubfx	x7,x9,#26,#26
664
665	add	$IN01_3,$IN01_3,$H3
666	 add	x4,x4,x5,lsl#32		// bfi	x4,x5,#32,#32
667	umlal	$ACC3,$IN01_1,${R2}[0]
668	 extr	x8,x12,x8,#52
669	umlal	$ACC4,$IN01_1,${R3}[0]
670	 extr	x9,x13,x9,#52
671	umlal	$ACC0,$IN01_1,${S4}[0]
672	 add	x6,x6,x7,lsl#32		// bfi	x6,x7,#32,#32
673	umlal	$ACC2,$IN01_1,${R1}[0]
674	 fmov	$IN01_0,x4
675	umlal	$ACC1,$IN01_1,${R0}[0]
676	 and	x8,x8,#0x03ffffff
677
678	add	$IN01_4,$IN01_4,$H4
679	 and	x9,x9,#0x03ffffff
680	umlal	$ACC3,$IN01_3,${R0}[0]
681	 ubfx	x10,x12,#14,#26
682	umlal	$ACC0,$IN01_3,${S2}[0]
683	 ubfx	x11,x13,#14,#26
684	umlal	$ACC4,$IN01_3,${R1}[0]
685	 add	x8,x8,x9,lsl#32		// bfi	x8,x9,#32,#32
686	umlal	$ACC1,$IN01_3,${S3}[0]
687	 fmov	$IN01_1,x6
688	umlal	$ACC2,$IN01_3,${S4}[0]
689	 add	x12,$padbit,x12,lsr#40
690
691	umlal	$ACC3,$IN01_4,${S4}[0]
692	 add	x13,$padbit,x13,lsr#40
693	umlal	$ACC0,$IN01_4,${S1}[0]
694	 add	x10,x10,x11,lsl#32	// bfi	x10,x11,#32,#32
695	umlal	$ACC4,$IN01_4,${R0}[0]
696	 add	x12,x12,x13,lsl#32	// bfi	x12,x13,#32,#32
697	umlal	$ACC1,$IN01_4,${S2}[0]
698	 fmov	$IN01_2,x8
699	umlal	$ACC2,$IN01_4,${S3}[0]
700	 fmov	$IN01_3,x10
701	 fmov	$IN01_4,x12
702
703	/////////////////////////////////////////////////////////////////
704	// lazy reduction as discussed in "NEON crypto" by D.J. Bernstein
705	// and P. Schwabe
706	//
707	// [see discussion in poly1305-armv4 module]
708
709	ushr	$T0.2d,$ACC3,#26
710	xtn	$H3,$ACC3
711	 ushr	$T1.2d,$ACC0,#26
712	 and	$ACC0,$ACC0,$MASK.2d
713	add	$ACC4,$ACC4,$T0.2d	// h3 -> h4
714	bic	$H3,#0xfc,lsl#24	// &=0x03ffffff
715	 add	$ACC1,$ACC1,$T1.2d	// h0 -> h1
716
717	ushr	$T0.2d,$ACC4,#26
718	xtn	$H4,$ACC4
719	 ushr	$T1.2d,$ACC1,#26
720	 xtn	$H1,$ACC1
721	bic	$H4,#0xfc,lsl#24
722	 add	$ACC2,$ACC2,$T1.2d	// h1 -> h2
723
724	add	$ACC0,$ACC0,$T0.2d
725	shl	$T0.2d,$T0.2d,#2
726	 shrn	$T1.2s,$ACC2,#26
727	 xtn	$H2,$ACC2
728	add	$ACC0,$ACC0,$T0.2d	// h4 -> h0
729	 bic	$H1,#0xfc,lsl#24
730	 add	$H3,$H3,$T1.2s		// h2 -> h3
731	 bic	$H2,#0xfc,lsl#24
732
733	shrn	$T0.2s,$ACC0,#26
734	xtn	$H0,$ACC0
735	 ushr	$T1.2s,$H3,#26
736	 bic	$H3,#0xfc,lsl#24
737	 bic	$H0,#0xfc,lsl#24
738	add	$H1,$H1,$T0.2s		// h0 -> h1
739	 add	$H4,$H4,$T1.2s		// h3 -> h4
740
741	b.hi	.Loop_neon
742
743.Lskip_loop:
744	dup	$IN23_2,${IN23_2}[0]
745	add	$IN01_2,$IN01_2,$H2
746
747	////////////////////////////////////////////////////////////////
748	// multiply (inp[0:1]+hash) or inp[2:3] by r^2:r^1
749
750	adds	$len,$len,#32
751	b.ne	.Long_tail
752
753	dup	$IN23_2,${IN01_2}[0]
754	add	$IN23_0,$IN01_0,$H0
755	add	$IN23_3,$IN01_3,$H3
756	add	$IN23_1,$IN01_1,$H1
757	add	$IN23_4,$IN01_4,$H4
758
759.Long_tail:
760	dup	$IN23_0,${IN23_0}[0]
761	umull2	$ACC0,$IN23_2,${S3}
762	umull2	$ACC3,$IN23_2,${R1}
763	umull2	$ACC4,$IN23_2,${R2}
764	umull2	$ACC2,$IN23_2,${R0}
765	umull2	$ACC1,$IN23_2,${S4}
766
767	dup	$IN23_1,${IN23_1}[0]
768	umlal2	$ACC0,$IN23_0,${R0}
769	umlal2	$ACC2,$IN23_0,${R2}
770	umlal2	$ACC3,$IN23_0,${R3}
771	umlal2	$ACC4,$IN23_0,${R4}
772	umlal2	$ACC1,$IN23_0,${R1}
773
774	dup	$IN23_3,${IN23_3}[0]
775	umlal2	$ACC0,$IN23_1,${S4}
776	umlal2	$ACC3,$IN23_1,${R2}
777	umlal2	$ACC2,$IN23_1,${R1}
778	umlal2	$ACC4,$IN23_1,${R3}
779	umlal2	$ACC1,$IN23_1,${R0}
780
781	dup	$IN23_4,${IN23_4}[0]
782	umlal2	$ACC3,$IN23_3,${R0}
783	umlal2	$ACC4,$IN23_3,${R1}
784	umlal2	$ACC0,$IN23_3,${S2}
785	umlal2	$ACC1,$IN23_3,${S3}
786	umlal2	$ACC2,$IN23_3,${S4}
787
788	umlal2	$ACC3,$IN23_4,${S4}
789	umlal2	$ACC0,$IN23_4,${S1}
790	umlal2	$ACC4,$IN23_4,${R0}
791	umlal2	$ACC1,$IN23_4,${S2}
792	umlal2	$ACC2,$IN23_4,${S3}
793
794	b.eq	.Lshort_tail
795
796	////////////////////////////////////////////////////////////////
797	// (hash+inp[0:1])*r^4:r^3 and accumulate
798
799	add	$IN01_0,$IN01_0,$H0
800	umlal	$ACC3,$IN01_2,${R1}
801	umlal	$ACC0,$IN01_2,${S3}
802	umlal	$ACC4,$IN01_2,${R2}
803	umlal	$ACC1,$IN01_2,${S4}
804	umlal	$ACC2,$IN01_2,${R0}
805
806	add	$IN01_1,$IN01_1,$H1
807	umlal	$ACC3,$IN01_0,${R3}
808	umlal	$ACC0,$IN01_0,${R0}
809	umlal	$ACC4,$IN01_0,${R4}
810	umlal	$ACC1,$IN01_0,${R1}
811	umlal	$ACC2,$IN01_0,${R2}
812
813	add	$IN01_3,$IN01_3,$H3
814	umlal	$ACC3,$IN01_1,${R2}
815	umlal	$ACC0,$IN01_1,${S4}
816	umlal	$ACC4,$IN01_1,${R3}
817	umlal	$ACC1,$IN01_1,${R0}
818	umlal	$ACC2,$IN01_1,${R1}
819
820	add	$IN01_4,$IN01_4,$H4
821	umlal	$ACC3,$IN01_3,${R0}
822	umlal	$ACC0,$IN01_3,${S2}
823	umlal	$ACC4,$IN01_3,${R1}
824	umlal	$ACC1,$IN01_3,${S3}
825	umlal	$ACC2,$IN01_3,${S4}
826
827	umlal	$ACC3,$IN01_4,${S4}
828	umlal	$ACC0,$IN01_4,${S1}
829	umlal	$ACC4,$IN01_4,${R0}
830	umlal	$ACC1,$IN01_4,${S2}
831	umlal	$ACC2,$IN01_4,${S3}
832
833.Lshort_tail:
834	////////////////////////////////////////////////////////////////
835	// horizontal add
836
837	addp	$ACC3,$ACC3,$ACC3
838	 ldp	d8,d9,[sp,#16]		// meet ABI requirements
839	addp	$ACC0,$ACC0,$ACC0
840	 ldp	d10,d11,[sp,#32]
841	addp	$ACC4,$ACC4,$ACC4
842	 ldp	d12,d13,[sp,#48]
843	addp	$ACC1,$ACC1,$ACC1
844	 ldp	d14,d15,[sp,#64]
845	addp	$ACC2,$ACC2,$ACC2
846	 ldr	x30,[sp,#8]
847
848	////////////////////////////////////////////////////////////////
849	// lazy reduction, but without narrowing
850
851	ushr	$T0.2d,$ACC3,#26
852	and	$ACC3,$ACC3,$MASK.2d
853	 ushr	$T1.2d,$ACC0,#26
854	 and	$ACC0,$ACC0,$MASK.2d
855
856	add	$ACC4,$ACC4,$T0.2d	// h3 -> h4
857	 add	$ACC1,$ACC1,$T1.2d	// h0 -> h1
858
859	ushr	$T0.2d,$ACC4,#26
860	and	$ACC4,$ACC4,$MASK.2d
861	 ushr	$T1.2d,$ACC1,#26
862	 and	$ACC1,$ACC1,$MASK.2d
863	 add	$ACC2,$ACC2,$T1.2d	// h1 -> h2
864
865	add	$ACC0,$ACC0,$T0.2d
866	shl	$T0.2d,$T0.2d,#2
867	 ushr	$T1.2d,$ACC2,#26
868	 and	$ACC2,$ACC2,$MASK.2d
869	add	$ACC0,$ACC0,$T0.2d	// h4 -> h0
870	 add	$ACC3,$ACC3,$T1.2d	// h2 -> h3
871
872	ushr	$T0.2d,$ACC0,#26
873	and	$ACC0,$ACC0,$MASK.2d
874	 ushr	$T1.2d,$ACC3,#26
875	 and	$ACC3,$ACC3,$MASK.2d
876	add	$ACC1,$ACC1,$T0.2d	// h0 -> h1
877	 add	$ACC4,$ACC4,$T1.2d	// h3 -> h4
878
879	////////////////////////////////////////////////////////////////
880	// write the result, can be partially reduced
881
882	st4	{$ACC0,$ACC1,$ACC2,$ACC3}[0],[$ctx],#16
883	mov	x4,#1
884	st1	{$ACC4}[0],[$ctx]
885	str	x4,[$ctx,#8]		// set is_base2_26
886
887	ldr	x29,[sp],#80
888	 .inst	0xd50323bf		// autiasp
889	ret
890.size	poly1305_blocks_neon,.-poly1305_blocks_neon
891
892.pushsection .rodata
893.align	5
894.Lzeros:
895.long	0,0,0,0,0,0,0,0
896.asciz	"Poly1305 for ARMv8, CRYPTOGAMS by \@dot-asm"
897.popsection
898
899.align	2
900#if !defined(__KERNEL__) && !defined(_WIN64)
901.comm	OPENSSL_armcap_P,4,4
902.hidden	OPENSSL_armcap_P
903#endif
904___
905
906foreach (split("\n",$code)) {
907	s/\b(shrn\s+v[0-9]+)\.[24]d/$1.2s/			or
908	s/\b(fmov\s+)v([0-9]+)[^,]*,\s*x([0-9]+)/$1d$2,x$3/	or
909	(m/\bdup\b/ and (s/\.[24]s/.2d/g or 1))			or
910	(m/\b(eor|and)/ and (s/\.[248][sdh]/.16b/g or 1))	or
911	(m/\bum(ul|la)l\b/ and (s/\.4s/.2s/g or 1))		or
912	(m/\bum(ul|la)l2\b/ and (s/\.2s/.4s/g or 1))		or
913	(m/\bst[1-4]\s+{[^}]+}\[/ and (s/\.[24]d/.s/g or 1));
914
915	s/\.[124]([sd])\[/.$1\[/;
916	s/w#x([0-9]+)/w$1/g;
917
918	print $_,"\n";
919}
920close STDOUT;
921