2 # Copyright 2016 The OpenSSL Project Authors. All Rights Reserved.
4 # Licensed under the Apache License 2.0 (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
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 # ====================================================================
17 # This module implements Poly1305 hash for ARMv8.
21 # Numbers are cycles per processed byte with poly1305_blocks alone.
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
32 # ThunderX2 1.17/+95% 1.36
34 # (*) estimate based on resources availability is less than 1.0,
35 # i.e. measured result is worse than expected, presumably binary
36 # translator is not almighty;
38 # $output is the last argument if it looks like a file (it has an extension)
39 # $flavour is the first argument if it doesn't look like a file
40 $output = $#ARGV >= 0 && $ARGV[$#ARGV] =~ m|\.\w+$| ? pop : undef;
41 $flavour = $#ARGV >= 0 && $ARGV[0] !~ m|\.| ? shift : undef;
43 $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
44 ( $xlate="${dir}arm-xlate.pl" and -f $xlate ) or
45 ( $xlate="${dir}../../perlasm/arm-xlate.pl" and -f $xlate) or
46 die "can't locate arm-xlate.pl";
48 open OUT,"| \"$^X\" $xlate $flavour \"$output\""
49 or die "can't call $xlate: $!";
52 my ($ctx,$inp,$len,$padbit) = map("x$_",(0..3));
53 my ($mac,$nonce)=($inp,$len);
55 my ($h0,$h1,$h2,$r0,$r1,$s1,$t0,$t1,$d0,$d1,$d2) = map("x$_",(4..14));
62 // forward "declarations" are required for Apple
63 .extern OPENSSL_armcap_P
64 .globl poly1305_blocks
68 .type poly1305_init,%function
72 stp xzr,xzr,[$ctx] // zero hash value
73 stp xzr,xzr,[$ctx,#16] // [along with is_base2_26]
78 adrp x17,OPENSSL_armcap_P
79 ldr w17,[x17,#:lo12:OPENSSL_armcap_P]
81 ldp $r0,$r1,[$inp] // load key
82 mov $s1,#0xfffffffc0fffffff
83 movk $s1,#0x0fff,lsl#48
85 rev $r0,$r0 // flip bytes
88 and $r0,$r0,$s1 // &=0ffffffc0fffffff
90 and $r1,$r1,$s1 // &=0ffffffc0ffffffc
91 stp $r0,$r1,[$ctx,#32] // save key value
95 adr $d0,.Lpoly1305_blocks
96 adr $r0,.Lpoly1305_blocks_neon
97 adr $d1,.Lpoly1305_emit
98 adr $r1,.Lpoly1305_emit_neon
112 .size poly1305_init,.-poly1305_init
114 .type poly1305_blocks,%function
121 ldp $h0,$h1,[$ctx] // load hash value
122 ldp $r0,$r1,[$ctx,#32] // load key value
124 add $s1,$r1,$r1,lsr#2 // s1 = r1 + (r1 >> 2)
129 ldp $t0,$t1,[$inp],#16 // load input
135 adds $h0,$h0,$t0 // accumulate input
138 mul $d0,$h0,$r0 // h0*r0
142 mul $t0,$h1,$s1 // h1*5*r1
146 mul $t0,$h0,$r1 // h0*r1
151 mul $t0,$h1,$r0 // h1*r0
156 mul $t0,$h2,$s1 // h2*5*r1
158 mul $t1,$h2,$r0 // h2*r0
163 and $t0,$d2,#-4 // final reduction
165 add $t0,$t0,$d2,lsr#2
172 stp $h0,$h1,[$ctx] // store hash value
177 .size poly1305_blocks,.-poly1305_blocks
179 .type poly1305_emit,%function
183 ldp $h0,$h1,[$ctx] // load hash base 2^64
185 ldp $t0,$t1,[$nonce] // load nonce
187 adds $d0,$h0,#5 // compare to modulus
191 tst $d2,#-4 // see if it's carried/borrowed
197 ror $t0,$t0,#32 // flip nonce words
200 adds $h0,$h0,$t0 // accumulate nonce
203 rev $h0,$h0 // flip output bytes
206 stp $h0,$h1,[$mac] // write result
209 .size poly1305_emit,.-poly1305_emit
211 my ($R0,$R1,$S1,$R2,$S2,$R3,$S3,$R4,$S4) = map("v$_.4s",(0..8));
212 my ($IN01_0,$IN01_1,$IN01_2,$IN01_3,$IN01_4) = map("v$_.2s",(9..13));
213 my ($IN23_0,$IN23_1,$IN23_2,$IN23_3,$IN23_4) = map("v$_.2s",(14..18));
214 my ($ACC0,$ACC1,$ACC2,$ACC3,$ACC4) = map("v$_.2d",(19..23));
215 my ($H0,$H1,$H2,$H3,$H4) = map("v$_.2s",(24..28));
216 my ($T0,$T1,$MASK) = map("v$_",(29..31));
218 my ($in2,$zeros)=("x16","x17");
219 my $is_base2_26 = $zeros; # borrow
222 .type poly1305_mult,%function
225 mul $d0,$h0,$r0 // h0*r0
228 mul $t0,$h1,$s1 // h1*5*r1
232 mul $t0,$h0,$r1 // h0*r1
237 mul $t0,$h1,$r0 // h1*r0
242 mul $t0,$h2,$s1 // h2*5*r1
244 mul $t1,$h2,$r0 // h2*r0
249 and $t0,$d2,#-4 // final reduction
251 add $t0,$t0,$d2,lsr#2
257 .size poly1305_mult,.-poly1305_mult
259 .type poly1305_splat,%function
262 and x12,$h0,#0x03ffffff // base 2^64 -> base 2^26
265 and x14,x14,#0x03ffffff
269 str w12,[$ctx,#16*0] // r0
270 add w12,w13,w13,lsl#2 // r1*5
271 str w13,[$ctx,#16*1] // r1
272 add w13,w14,w14,lsl#2 // r2*5
273 str w12,[$ctx,#16*2] // s1
274 str w14,[$ctx,#16*3] // r2
275 add w14,w15,w15,lsl#2 // r3*5
276 str w13,[$ctx,#16*4] // s2
277 str w15,[$ctx,#16*5] // r3
278 add w15,w16,w16,lsl#2 // r4*5
279 str w14,[$ctx,#16*6] // s3
280 str w16,[$ctx,#16*7] // r4
281 str w15,[$ctx,#16*8] // s4
284 .size poly1305_splat,.-poly1305_splat
286 .type poly1305_blocks_neon,%function
288 poly1305_blocks_neon:
289 .Lpoly1305_blocks_neon:
290 ldr $is_base2_26,[$ctx,#24]
293 cbz $is_base2_26,.Lpoly1305_blocks
296 .inst 0xd503233f // paciasp
297 stp x29,x30,[sp,#-80]!
303 cbz $is_base2_26,.Lbase2_64_neon
305 ldp w10,w11,[$ctx] // load hash value base 2^26
306 ldp w12,w13,[$ctx,#8]
312 ldp $r0,$r1,[$ctx,#32] // load key value
314 add $h0,x10,x11,lsl#26 // base 2^26 -> base 2^64
316 adds $h0,$h0,x12,lsl#52
317 add $h1,$h1,x13,lsl#14
320 adds $h1,$h1,x14,lsl#40
321 adc $d2,$h2,xzr // can be partially reduced...
323 ldp $d0,$d1,[$inp],#16 // load input
325 add $s1,$r1,$r1,lsr#2 // s1 = r1 + (r1 >> 2)
327 and $t0,$d2,#-4 // ... so reduce
329 add $t0,$t0,$d2,lsr#2
338 adds $h0,$h0,$d0 // accumulate input
345 cbz $padbit,.Lstore_base2_64_neon
347 and x10,$h0,#0x03ffffff // base 2^64 -> base 2^26
350 and x12,x12,#0x03ffffff
354 cbnz $len,.Leven_neon
356 stp w10,w11,[$ctx] // store hash value base 2^26
357 stp w12,w13,[$ctx,#8]
362 .Lstore_base2_64_neon:
363 stp $h0,$h1,[$ctx] // store hash value base 2^64
364 stp $h2,xzr,[$ctx,#16] // note that is_base2_26 is zeroed
369 ldp $r0,$r1,[$ctx,#32] // load key value
371 ldp $h0,$h1,[$ctx] // load hash value base 2^64
377 ldp $d0,$d1,[$inp],#16 // load input
379 add $s1,$r1,$r1,lsr#2 // s1 = r1 + (r1 >> 2)
384 adds $h0,$h0,$d0 // accumulate input
391 and x10,$h0,#0x03ffffff // base 2^64 -> base 2^26
394 and x12,x12,#0x03ffffff
398 stp d8,d9,[sp,#16] // meet ABI requirements
409 ////////////////////////////////// initialize r^n table
411 add $s1,$r1,$r1,lsr#2 // s1 = r1 + (r1 >> 2)
417 bl poly1305_mult // r^2
421 bl poly1305_mult // r^3
425 bl poly1305_mult // r^4
433 csel $in2,$zeros,$in2,lo
436 stur x4,[$ctx,#-24] // set is_base2_26
437 sub $ctx,$ctx,#48 // restore original $ctx
445 csel $in2,$zeros,$in2,lo
447 stp d8,d9,[sp,#16] // meet ABI requirements
459 ldp x8,x12,[$in2],#16 // inp[2:3] (or zero)
460 ldp x9,x13,[$in2],#48
462 lsl $padbit,$padbit,#24
471 and x4,x8,#0x03ffffff // base 2^64 -> base 2^26
472 and x5,x9,#0x03ffffff
475 add x4,x4,x5,lsl#32 // bfi x4,x5,#32,#32
478 add x6,x6,x7,lsl#32 // bfi x6,x7,#32,#32
480 and x8,x8,#0x03ffffff
481 and x9,x9,#0x03ffffff
484 add x12,$padbit,x12,lsr#40
485 add x13,$padbit,x13,lsr#40
486 add x8,x8,x9,lsl#32 // bfi x8,x9,#32,#32
488 add x10,x10,x11,lsl#32 // bfi x10,x11,#32,#32
489 add x12,x12,x13,lsl#32 // bfi x12,x13,#32,#32
494 ldp x8,x12,[$inp],#16 // inp[0:1]
495 ldp x9,x13,[$inp],#48
497 ld1 {$R0,$R1,$S1,$R2},[x15],#64
498 ld1 {$S2,$R3,$S3,$R4},[x15],#64
507 and x4,x8,#0x03ffffff // base 2^64 -> base 2^26
508 and x5,x9,#0x03ffffff
511 add x4,x4,x5,lsl#32 // bfi x4,x5,#32,#32
514 add x6,x6,x7,lsl#32 // bfi x6,x7,#32,#32
516 and x8,x8,#0x03ffffff
517 and x9,x9,#0x03ffffff
520 add x12,$padbit,x12,lsr#40
521 add x13,$padbit,x13,lsr#40
522 add x8,x8,x9,lsl#32 // bfi x8,x9,#32,#32
524 add x10,x10,x11,lsl#32 // bfi x10,x11,#32,#32
525 add x12,x12,x13,lsl#32 // bfi x12,x13,#32,#32
530 ushr $MASK.2d,$MASK.2d,#38
536 ////////////////////////////////////////////////////////////////
537 // ((inp[0]*r^4+inp[2]*r^2+inp[4])*r^4+inp[6]*r^2
538 // ((inp[1]*r^4+inp[3]*r^2+inp[5])*r^3+inp[7]*r
539 // \___________________/
540 // ((inp[0]*r^4+inp[2]*r^2+inp[4])*r^4+inp[6]*r^2+inp[8])*r^2
541 // ((inp[1]*r^4+inp[3]*r^2+inp[5])*r^4+inp[7]*r^2+inp[9])*r
542 // \___________________/ \____________________/
544 // Note that we start with inp[2:3]*r^2. This is because it
545 // doesn't depend on reduction in previous iteration.
546 ////////////////////////////////////////////////////////////////
547 // d4 = h0*r4 + h1*r3 + h2*r2 + h3*r1 + h4*r0
548 // d3 = h0*r3 + h1*r2 + h2*r1 + h3*r0 + h4*5*r4
549 // d2 = h0*r2 + h1*r1 + h2*r0 + h3*5*r4 + h4*5*r3
550 // d1 = h0*r1 + h1*r0 + h2*5*r4 + h3*5*r3 + h4*5*r2
551 // d0 = h0*r0 + h1*5*r4 + h2*5*r3 + h3*5*r2 + h4*5*r1
554 umull $ACC4,$IN23_0,${R4}[2]
555 csel $in2,$zeros,$in2,lo
556 umull $ACC3,$IN23_0,${R3}[2]
557 umull $ACC2,$IN23_0,${R2}[2]
558 ldp x8,x12,[$in2],#16 // inp[2:3] (or zero)
559 umull $ACC1,$IN23_0,${R1}[2]
560 ldp x9,x13,[$in2],#48
561 umull $ACC0,$IN23_0,${R0}[2]
569 umlal $ACC4,$IN23_1,${R3}[2]
570 and x4,x8,#0x03ffffff // base 2^64 -> base 2^26
571 umlal $ACC3,$IN23_1,${R2}[2]
572 and x5,x9,#0x03ffffff
573 umlal $ACC2,$IN23_1,${R1}[2]
575 umlal $ACC1,$IN23_1,${R0}[2]
577 umlal $ACC0,$IN23_1,${S4}[2]
578 add x4,x4,x5,lsl#32 // bfi x4,x5,#32,#32
580 umlal $ACC4,$IN23_2,${R2}[2]
582 umlal $ACC3,$IN23_2,${R1}[2]
584 umlal $ACC2,$IN23_2,${R0}[2]
585 add x6,x6,x7,lsl#32 // bfi x6,x7,#32,#32
586 umlal $ACC1,$IN23_2,${S4}[2]
588 umlal $ACC0,$IN23_2,${S3}[2]
589 and x8,x8,#0x03ffffff
591 umlal $ACC4,$IN23_3,${R1}[2]
592 and x9,x9,#0x03ffffff
593 umlal $ACC3,$IN23_3,${R0}[2]
595 umlal $ACC2,$IN23_3,${S4}[2]
597 umlal $ACC1,$IN23_3,${S3}[2]
598 add x8,x8,x9,lsl#32 // bfi x8,x9,#32,#32
599 umlal $ACC0,$IN23_3,${S2}[2]
602 add $IN01_2,$IN01_2,$H2
603 add x12,$padbit,x12,lsr#40
604 umlal $ACC4,$IN23_4,${R0}[2]
605 add x13,$padbit,x13,lsr#40
606 umlal $ACC3,$IN23_4,${S4}[2]
607 add x10,x10,x11,lsl#32 // bfi x10,x11,#32,#32
608 umlal $ACC2,$IN23_4,${S3}[2]
609 add x12,x12,x13,lsl#32 // bfi x12,x13,#32,#32
610 umlal $ACC1,$IN23_4,${S2}[2]
612 umlal $ACC0,$IN23_4,${S1}[2]
615 ////////////////////////////////////////////////////////////////
616 // (hash+inp[0:1])*r^4 and accumulate
618 add $IN01_0,$IN01_0,$H0
620 umlal $ACC3,$IN01_2,${R1}[0]
621 ldp x8,x12,[$inp],#16 // inp[0:1]
622 umlal $ACC0,$IN01_2,${S3}[0]
623 ldp x9,x13,[$inp],#48
624 umlal $ACC4,$IN01_2,${R2}[0]
625 umlal $ACC1,$IN01_2,${S4}[0]
626 umlal $ACC2,$IN01_2,${R0}[0]
634 add $IN01_1,$IN01_1,$H1
635 umlal $ACC3,$IN01_0,${R3}[0]
636 umlal $ACC4,$IN01_0,${R4}[0]
637 and x4,x8,#0x03ffffff // base 2^64 -> base 2^26
638 umlal $ACC2,$IN01_0,${R2}[0]
639 and x5,x9,#0x03ffffff
640 umlal $ACC0,$IN01_0,${R0}[0]
642 umlal $ACC1,$IN01_0,${R1}[0]
645 add $IN01_3,$IN01_3,$H3
646 add x4,x4,x5,lsl#32 // bfi x4,x5,#32,#32
647 umlal $ACC3,$IN01_1,${R2}[0]
649 umlal $ACC4,$IN01_1,${R3}[0]
651 umlal $ACC0,$IN01_1,${S4}[0]
652 add x6,x6,x7,lsl#32 // bfi x6,x7,#32,#32
653 umlal $ACC2,$IN01_1,${R1}[0]
655 umlal $ACC1,$IN01_1,${R0}[0]
656 and x8,x8,#0x03ffffff
658 add $IN01_4,$IN01_4,$H4
659 and x9,x9,#0x03ffffff
660 umlal $ACC3,$IN01_3,${R0}[0]
662 umlal $ACC0,$IN01_3,${S2}[0]
664 umlal $ACC4,$IN01_3,${R1}[0]
665 add x8,x8,x9,lsl#32 // bfi x8,x9,#32,#32
666 umlal $ACC1,$IN01_3,${S3}[0]
668 umlal $ACC2,$IN01_3,${S4}[0]
669 add x12,$padbit,x12,lsr#40
671 umlal $ACC3,$IN01_4,${S4}[0]
672 add x13,$padbit,x13,lsr#40
673 umlal $ACC0,$IN01_4,${S1}[0]
674 add x10,x10,x11,lsl#32 // bfi x10,x11,#32,#32
675 umlal $ACC4,$IN01_4,${R0}[0]
676 add x12,x12,x13,lsl#32 // bfi x12,x13,#32,#32
677 umlal $ACC1,$IN01_4,${S2}[0]
679 umlal $ACC2,$IN01_4,${S3}[0]
683 /////////////////////////////////////////////////////////////////
684 // lazy reduction as discussed in "NEON crypto" by D.J. Bernstein
687 // [see discussion in poly1305-armv4 module]
689 ushr $T0.2d,$ACC3,#26
691 ushr $T1.2d,$ACC0,#26
692 and $ACC0,$ACC0,$MASK.2d
693 add $ACC4,$ACC4,$T0.2d // h3 -> h4
694 bic $H3,#0xfc,lsl#24 // &=0x03ffffff
695 add $ACC1,$ACC1,$T1.2d // h0 -> h1
697 ushr $T0.2d,$ACC4,#26
699 ushr $T1.2d,$ACC1,#26
702 add $ACC2,$ACC2,$T1.2d // h1 -> h2
704 add $ACC0,$ACC0,$T0.2d
706 shrn $T1.2s,$ACC2,#26
708 add $ACC0,$ACC0,$T0.2d // h4 -> h0
710 add $H3,$H3,$T1.2s // h2 -> h3
713 shrn $T0.2s,$ACC0,#26
718 add $H1,$H1,$T0.2s // h0 -> h1
719 add $H4,$H4,$T1.2s // h3 -> h4
724 dup $IN23_2,${IN23_2}[0]
725 add $IN01_2,$IN01_2,$H2
727 ////////////////////////////////////////////////////////////////
728 // multiply (inp[0:1]+hash) or inp[2:3] by r^2:r^1
733 dup $IN23_2,${IN01_2}[0]
734 add $IN23_0,$IN01_0,$H0
735 add $IN23_3,$IN01_3,$H3
736 add $IN23_1,$IN01_1,$H1
737 add $IN23_4,$IN01_4,$H4
740 dup $IN23_0,${IN23_0}[0]
741 umull2 $ACC0,$IN23_2,${S3}
742 umull2 $ACC3,$IN23_2,${R1}
743 umull2 $ACC4,$IN23_2,${R2}
744 umull2 $ACC2,$IN23_2,${R0}
745 umull2 $ACC1,$IN23_2,${S4}
747 dup $IN23_1,${IN23_1}[0]
748 umlal2 $ACC0,$IN23_0,${R0}
749 umlal2 $ACC2,$IN23_0,${R2}
750 umlal2 $ACC3,$IN23_0,${R3}
751 umlal2 $ACC4,$IN23_0,${R4}
752 umlal2 $ACC1,$IN23_0,${R1}
754 dup $IN23_3,${IN23_3}[0]
755 umlal2 $ACC0,$IN23_1,${S4}
756 umlal2 $ACC3,$IN23_1,${R2}
757 umlal2 $ACC2,$IN23_1,${R1}
758 umlal2 $ACC4,$IN23_1,${R3}
759 umlal2 $ACC1,$IN23_1,${R0}
761 dup $IN23_4,${IN23_4}[0]
762 umlal2 $ACC3,$IN23_3,${R0}
763 umlal2 $ACC4,$IN23_3,${R1}
764 umlal2 $ACC0,$IN23_3,${S2}
765 umlal2 $ACC1,$IN23_3,${S3}
766 umlal2 $ACC2,$IN23_3,${S4}
768 umlal2 $ACC3,$IN23_4,${S4}
769 umlal2 $ACC0,$IN23_4,${S1}
770 umlal2 $ACC4,$IN23_4,${R0}
771 umlal2 $ACC1,$IN23_4,${S2}
772 umlal2 $ACC2,$IN23_4,${S3}
776 ////////////////////////////////////////////////////////////////
777 // (hash+inp[0:1])*r^4:r^3 and accumulate
779 add $IN01_0,$IN01_0,$H0
780 umlal $ACC3,$IN01_2,${R1}
781 umlal $ACC0,$IN01_2,${S3}
782 umlal $ACC4,$IN01_2,${R2}
783 umlal $ACC1,$IN01_2,${S4}
784 umlal $ACC2,$IN01_2,${R0}
786 add $IN01_1,$IN01_1,$H1
787 umlal $ACC3,$IN01_0,${R3}
788 umlal $ACC0,$IN01_0,${R0}
789 umlal $ACC4,$IN01_0,${R4}
790 umlal $ACC1,$IN01_0,${R1}
791 umlal $ACC2,$IN01_0,${R2}
793 add $IN01_3,$IN01_3,$H3
794 umlal $ACC3,$IN01_1,${R2}
795 umlal $ACC0,$IN01_1,${S4}
796 umlal $ACC4,$IN01_1,${R3}
797 umlal $ACC1,$IN01_1,${R0}
798 umlal $ACC2,$IN01_1,${R1}
800 add $IN01_4,$IN01_4,$H4
801 umlal $ACC3,$IN01_3,${R0}
802 umlal $ACC0,$IN01_3,${S2}
803 umlal $ACC4,$IN01_3,${R1}
804 umlal $ACC1,$IN01_3,${S3}
805 umlal $ACC2,$IN01_3,${S4}
807 umlal $ACC3,$IN01_4,${S4}
808 umlal $ACC0,$IN01_4,${S1}
809 umlal $ACC4,$IN01_4,${R0}
810 umlal $ACC1,$IN01_4,${S2}
811 umlal $ACC2,$IN01_4,${S3}
814 ////////////////////////////////////////////////////////////////
817 addp $ACC3,$ACC3,$ACC3
818 ldp d8,d9,[sp,#16] // meet ABI requirements
819 addp $ACC0,$ACC0,$ACC0
821 addp $ACC4,$ACC4,$ACC4
823 addp $ACC1,$ACC1,$ACC1
825 addp $ACC2,$ACC2,$ACC2
827 ////////////////////////////////////////////////////////////////
828 // lazy reduction, but without narrowing
830 ushr $T0.2d,$ACC3,#26
831 and $ACC3,$ACC3,$MASK.2d
832 ushr $T1.2d,$ACC0,#26
833 and $ACC0,$ACC0,$MASK.2d
835 add $ACC4,$ACC4,$T0.2d // h3 -> h4
836 add $ACC1,$ACC1,$T1.2d // h0 -> h1
838 ushr $T0.2d,$ACC4,#26
839 and $ACC4,$ACC4,$MASK.2d
840 ushr $T1.2d,$ACC1,#26
841 and $ACC1,$ACC1,$MASK.2d
842 add $ACC2,$ACC2,$T1.2d // h1 -> h2
844 add $ACC0,$ACC0,$T0.2d
846 ushr $T1.2d,$ACC2,#26
847 and $ACC2,$ACC2,$MASK.2d
848 add $ACC0,$ACC0,$T0.2d // h4 -> h0
849 add $ACC3,$ACC3,$T1.2d // h2 -> h3
851 ushr $T0.2d,$ACC0,#26
852 and $ACC0,$ACC0,$MASK.2d
853 ushr $T1.2d,$ACC3,#26
854 and $ACC3,$ACC3,$MASK.2d
855 add $ACC1,$ACC1,$T0.2d // h0 -> h1
856 add $ACC4,$ACC4,$T1.2d // h3 -> h4
858 ////////////////////////////////////////////////////////////////
859 // write the result, can be partially reduced
861 st4 {$ACC0,$ACC1,$ACC2,$ACC3}[0],[$ctx],#16
862 st1 {$ACC4}[0],[$ctx]
865 .inst 0xd50323bf // autiasp
868 .size poly1305_blocks_neon,.-poly1305_blocks_neon
870 .type poly1305_emit_neon,%function
873 .Lpoly1305_emit_neon:
874 ldr $is_base2_26,[$ctx,#24]
875 cbz $is_base2_26,poly1305_emit
877 ldp w10,w11,[$ctx] // load hash value base 2^26
878 ldp w12,w13,[$ctx,#8]
881 add $h0,x10,x11,lsl#26 // base 2^26 -> base 2^64
883 adds $h0,$h0,x12,lsl#52
884 add $h1,$h1,x13,lsl#14
887 adds $h1,$h1,x14,lsl#40
888 adc $h2,$h2,xzr // can be partially reduced...
890 ldp $t0,$t1,[$nonce] // load nonce
892 and $d0,$h2,#-4 // ... so reduce
893 add $d0,$d0,$h2,lsr#2
899 adds $d0,$h0,#5 // compare to modulus
903 tst $d2,#-4 // see if it's carried/borrowed
909 ror $t0,$t0,#32 // flip nonce words
912 adds $h0,$h0,$t0 // accumulate nonce
915 rev $h0,$h0 // flip output bytes
918 stp $h0,$h1,[$mac] // write result
921 .size poly1305_emit_neon,.-poly1305_emit_neon
925 .long 0,0,0,0,0,0,0,0
926 .asciz "Poly1305 for ARMv8, CRYPTOGAMS by <appro\@openssl.org>"
930 foreach (split("\n",$code)) {
931 s/\b(shrn\s+v[0-9]+)\.[24]d/$1.2s/ or
932 s/\b(fmov\s+)v([0-9]+)[^,]*,\s*x([0-9]+)/$1d$2,x$3/ or
933 (m/\bdup\b/ and (s/\.[24]s/.2d/g or 1)) or
934 (m/\b(eor|and)/ and (s/\.[248][sdh]/.16b/g or 1)) or
935 (m/\bum(ul|la)l\b/ and (s/\.4s/.2s/g or 1)) or
936 (m/\bum(ul|la)l2\b/ and (s/\.2s/.4s/g or 1)) or
937 (m/\bst[1-4]\s+{[^}]+}\[/ and (s/\.[24]d/.s/g or 1));
939 s/\.[124]([sd])\[/.$1\[/;
943 close STDOUT or die "error closing STDOUT";