2 # Copyright 2016-2020 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 .hidden OPENSSL_armcap_P
67 .globl poly1305_blocks
68 .hidden poly1305_blocks
72 .type poly1305_init,%function
76 stp xzr,xzr,[$ctx] // zero hash value
77 stp xzr,xzr,[$ctx,#16] // [along with is_base2_26]
82 adrp x17,OPENSSL_armcap_P
83 ldr w17,[x17,#:lo12:OPENSSL_armcap_P]
85 ldp $r0,$r1,[$inp] // load key
86 mov $s1,#0xfffffffc0fffffff
87 movk $s1,#0x0fff,lsl#48
89 rev $r0,$r0 // flip bytes
92 and $r0,$r0,$s1 // &=0ffffffc0fffffff
94 and $r1,$r1,$s1 // &=0ffffffc0ffffffc
95 stp $r0,$r1,[$ctx,#32] // save key value
99 adr $d0,.Lpoly1305_blocks
100 adr $r0,.Lpoly1305_blocks_neon
101 adr $d1,.Lpoly1305_emit
102 adr $r1,.Lpoly1305_emit_neon
116 .size poly1305_init,.-poly1305_init
118 .type poly1305_blocks,%function
125 ldp $h0,$h1,[$ctx] // load hash value
126 ldp $r0,$r1,[$ctx,#32] // load key value
128 add $s1,$r1,$r1,lsr#2 // s1 = r1 + (r1 >> 2)
133 ldp $t0,$t1,[$inp],#16 // load input
139 adds $h0,$h0,$t0 // accumulate input
142 mul $d0,$h0,$r0 // h0*r0
146 mul $t0,$h1,$s1 // h1*5*r1
150 mul $t0,$h0,$r1 // h0*r1
155 mul $t0,$h1,$r0 // h1*r0
160 mul $t0,$h2,$s1 // h2*5*r1
162 mul $t1,$h2,$r0 // h2*r0
167 and $t0,$d2,#-4 // final reduction
169 add $t0,$t0,$d2,lsr#2
176 stp $h0,$h1,[$ctx] // store hash value
181 .size poly1305_blocks,.-poly1305_blocks
183 .type poly1305_emit,%function
187 ldp $h0,$h1,[$ctx] // load hash base 2^64
189 ldp $t0,$t1,[$nonce] // load nonce
191 adds $d0,$h0,#5 // compare to modulus
195 tst $d2,#-4 // see if it's carried/borrowed
201 ror $t0,$t0,#32 // flip nonce words
204 adds $h0,$h0,$t0 // accumulate nonce
207 rev $h0,$h0 // flip output bytes
210 stp $h0,$h1,[$mac] // write result
213 .size poly1305_emit,.-poly1305_emit
215 my ($R0,$R1,$S1,$R2,$S2,$R3,$S3,$R4,$S4) = map("v$_.4s",(0..8));
216 my ($IN01_0,$IN01_1,$IN01_2,$IN01_3,$IN01_4) = map("v$_.2s",(9..13));
217 my ($IN23_0,$IN23_1,$IN23_2,$IN23_3,$IN23_4) = map("v$_.2s",(14..18));
218 my ($ACC0,$ACC1,$ACC2,$ACC3,$ACC4) = map("v$_.2d",(19..23));
219 my ($H0,$H1,$H2,$H3,$H4) = map("v$_.2s",(24..28));
220 my ($T0,$T1,$MASK) = map("v$_",(29..31));
222 my ($in2,$zeros)=("x16","x17");
223 my $is_base2_26 = $zeros; # borrow
226 .type poly1305_mult,%function
229 mul $d0,$h0,$r0 // h0*r0
232 mul $t0,$h1,$s1 // h1*5*r1
236 mul $t0,$h0,$r1 // h0*r1
241 mul $t0,$h1,$r0 // h1*r0
246 mul $t0,$h2,$s1 // h2*5*r1
248 mul $t1,$h2,$r0 // h2*r0
253 and $t0,$d2,#-4 // final reduction
255 add $t0,$t0,$d2,lsr#2
261 .size poly1305_mult,.-poly1305_mult
263 .type poly1305_splat,%function
266 and x12,$h0,#0x03ffffff // base 2^64 -> base 2^26
269 and x14,x14,#0x03ffffff
273 str w12,[$ctx,#16*0] // r0
274 add w12,w13,w13,lsl#2 // r1*5
275 str w13,[$ctx,#16*1] // r1
276 add w13,w14,w14,lsl#2 // r2*5
277 str w12,[$ctx,#16*2] // s1
278 str w14,[$ctx,#16*3] // r2
279 add w14,w15,w15,lsl#2 // r3*5
280 str w13,[$ctx,#16*4] // s2
281 str w15,[$ctx,#16*5] // r3
282 add w15,w16,w16,lsl#2 // r4*5
283 str w14,[$ctx,#16*6] // s3
284 str w16,[$ctx,#16*7] // r4
285 str w15,[$ctx,#16*8] // s4
288 .size poly1305_splat,.-poly1305_splat
290 .type poly1305_blocks_neon,%function
292 poly1305_blocks_neon:
293 .Lpoly1305_blocks_neon:
294 ldr $is_base2_26,[$ctx,#24]
297 cbz $is_base2_26,.Lpoly1305_blocks
300 .inst 0xd503233f // paciasp
301 stp x29,x30,[sp,#-80]!
307 cbz $is_base2_26,.Lbase2_64_neon
309 ldp w10,w11,[$ctx] // load hash value base 2^26
310 ldp w12,w13,[$ctx,#8]
316 ldp $r0,$r1,[$ctx,#32] // load key value
318 add $h0,x10,x11,lsl#26 // base 2^26 -> base 2^64
320 adds $h0,$h0,x12,lsl#52
321 add $h1,$h1,x13,lsl#14
324 adds $h1,$h1,x14,lsl#40
325 adc $d2,$h2,xzr // can be partially reduced...
327 ldp $d0,$d1,[$inp],#16 // load input
329 add $s1,$r1,$r1,lsr#2 // s1 = r1 + (r1 >> 2)
331 and $t0,$d2,#-4 // ... so reduce
333 add $t0,$t0,$d2,lsr#2
342 adds $h0,$h0,$d0 // accumulate input
349 cbz $padbit,.Lstore_base2_64_neon
351 and x10,$h0,#0x03ffffff // base 2^64 -> base 2^26
354 and x12,x12,#0x03ffffff
358 cbnz $len,.Leven_neon
360 stp w10,w11,[$ctx] // store hash value base 2^26
361 stp w12,w13,[$ctx,#8]
366 .Lstore_base2_64_neon:
367 stp $h0,$h1,[$ctx] // store hash value base 2^64
368 stp $h2,xzr,[$ctx,#16] // note that is_base2_26 is zeroed
373 ldp $r0,$r1,[$ctx,#32] // load key value
375 ldp $h0,$h1,[$ctx] // load hash value base 2^64
381 ldp $d0,$d1,[$inp],#16 // load input
383 add $s1,$r1,$r1,lsr#2 // s1 = r1 + (r1 >> 2)
388 adds $h0,$h0,$d0 // accumulate input
395 and x10,$h0,#0x03ffffff // base 2^64 -> base 2^26
398 and x12,x12,#0x03ffffff
402 stp d8,d9,[sp,#16] // meet ABI requirements
413 ////////////////////////////////// initialize r^n table
415 add $s1,$r1,$r1,lsr#2 // s1 = r1 + (r1 >> 2)
421 bl poly1305_mult // r^2
425 bl poly1305_mult // r^3
429 bl poly1305_mult // r^4
437 csel $in2,$zeros,$in2,lo
440 stur x4,[$ctx,#-24] // set is_base2_26
441 sub $ctx,$ctx,#48 // restore original $ctx
449 csel $in2,$zeros,$in2,lo
451 stp d8,d9,[sp,#16] // meet ABI requirements
463 ldp x8,x12,[$in2],#16 // inp[2:3] (or zero)
464 ldp x9,x13,[$in2],#48
466 lsl $padbit,$padbit,#24
475 and x4,x8,#0x03ffffff // base 2^64 -> base 2^26
476 and x5,x9,#0x03ffffff
479 add x4,x4,x5,lsl#32 // bfi x4,x5,#32,#32
482 add x6,x6,x7,lsl#32 // bfi x6,x7,#32,#32
484 and x8,x8,#0x03ffffff
485 and x9,x9,#0x03ffffff
488 add x12,$padbit,x12,lsr#40
489 add x13,$padbit,x13,lsr#40
490 add x8,x8,x9,lsl#32 // bfi x8,x9,#32,#32
492 add x10,x10,x11,lsl#32 // bfi x10,x11,#32,#32
493 add x12,x12,x13,lsl#32 // bfi x12,x13,#32,#32
498 ldp x8,x12,[$inp],#16 // inp[0:1]
499 ldp x9,x13,[$inp],#48
501 ld1 {$R0,$R1,$S1,$R2},[x15],#64
502 ld1 {$S2,$R3,$S3,$R4},[x15],#64
511 and x4,x8,#0x03ffffff // base 2^64 -> base 2^26
512 and x5,x9,#0x03ffffff
515 add x4,x4,x5,lsl#32 // bfi x4,x5,#32,#32
518 add x6,x6,x7,lsl#32 // bfi x6,x7,#32,#32
520 and x8,x8,#0x03ffffff
521 and x9,x9,#0x03ffffff
524 add x12,$padbit,x12,lsr#40
525 add x13,$padbit,x13,lsr#40
526 add x8,x8,x9,lsl#32 // bfi x8,x9,#32,#32
528 add x10,x10,x11,lsl#32 // bfi x10,x11,#32,#32
529 add x12,x12,x13,lsl#32 // bfi x12,x13,#32,#32
534 ushr $MASK.2d,$MASK.2d,#38
540 ////////////////////////////////////////////////////////////////
541 // ((inp[0]*r^4+inp[2]*r^2+inp[4])*r^4+inp[6]*r^2
542 // ((inp[1]*r^4+inp[3]*r^2+inp[5])*r^3+inp[7]*r
543 // \___________________/
544 // ((inp[0]*r^4+inp[2]*r^2+inp[4])*r^4+inp[6]*r^2+inp[8])*r^2
545 // ((inp[1]*r^4+inp[3]*r^2+inp[5])*r^4+inp[7]*r^2+inp[9])*r
546 // \___________________/ \____________________/
548 // Note that we start with inp[2:3]*r^2. This is because it
549 // doesn't depend on reduction in previous iteration.
550 ////////////////////////////////////////////////////////////////
551 // d4 = h0*r4 + h1*r3 + h2*r2 + h3*r1 + h4*r0
552 // d3 = h0*r3 + h1*r2 + h2*r1 + h3*r0 + h4*5*r4
553 // d2 = h0*r2 + h1*r1 + h2*r0 + h3*5*r4 + h4*5*r3
554 // d1 = h0*r1 + h1*r0 + h2*5*r4 + h3*5*r3 + h4*5*r2
555 // d0 = h0*r0 + h1*5*r4 + h2*5*r3 + h3*5*r2 + h4*5*r1
558 umull $ACC4,$IN23_0,${R4}[2]
559 csel $in2,$zeros,$in2,lo
560 umull $ACC3,$IN23_0,${R3}[2]
561 umull $ACC2,$IN23_0,${R2}[2]
562 ldp x8,x12,[$in2],#16 // inp[2:3] (or zero)
563 umull $ACC1,$IN23_0,${R1}[2]
564 ldp x9,x13,[$in2],#48
565 umull $ACC0,$IN23_0,${R0}[2]
573 umlal $ACC4,$IN23_1,${R3}[2]
574 and x4,x8,#0x03ffffff // base 2^64 -> base 2^26
575 umlal $ACC3,$IN23_1,${R2}[2]
576 and x5,x9,#0x03ffffff
577 umlal $ACC2,$IN23_1,${R1}[2]
579 umlal $ACC1,$IN23_1,${R0}[2]
581 umlal $ACC0,$IN23_1,${S4}[2]
582 add x4,x4,x5,lsl#32 // bfi x4,x5,#32,#32
584 umlal $ACC4,$IN23_2,${R2}[2]
586 umlal $ACC3,$IN23_2,${R1}[2]
588 umlal $ACC2,$IN23_2,${R0}[2]
589 add x6,x6,x7,lsl#32 // bfi x6,x7,#32,#32
590 umlal $ACC1,$IN23_2,${S4}[2]
592 umlal $ACC0,$IN23_2,${S3}[2]
593 and x8,x8,#0x03ffffff
595 umlal $ACC4,$IN23_3,${R1}[2]
596 and x9,x9,#0x03ffffff
597 umlal $ACC3,$IN23_3,${R0}[2]
599 umlal $ACC2,$IN23_3,${S4}[2]
601 umlal $ACC1,$IN23_3,${S3}[2]
602 add x8,x8,x9,lsl#32 // bfi x8,x9,#32,#32
603 umlal $ACC0,$IN23_3,${S2}[2]
606 add $IN01_2,$IN01_2,$H2
607 add x12,$padbit,x12,lsr#40
608 umlal $ACC4,$IN23_4,${R0}[2]
609 add x13,$padbit,x13,lsr#40
610 umlal $ACC3,$IN23_4,${S4}[2]
611 add x10,x10,x11,lsl#32 // bfi x10,x11,#32,#32
612 umlal $ACC2,$IN23_4,${S3}[2]
613 add x12,x12,x13,lsl#32 // bfi x12,x13,#32,#32
614 umlal $ACC1,$IN23_4,${S2}[2]
616 umlal $ACC0,$IN23_4,${S1}[2]
619 ////////////////////////////////////////////////////////////////
620 // (hash+inp[0:1])*r^4 and accumulate
622 add $IN01_0,$IN01_0,$H0
624 umlal $ACC3,$IN01_2,${R1}[0]
625 ldp x8,x12,[$inp],#16 // inp[0:1]
626 umlal $ACC0,$IN01_2,${S3}[0]
627 ldp x9,x13,[$inp],#48
628 umlal $ACC4,$IN01_2,${R2}[0]
629 umlal $ACC1,$IN01_2,${S4}[0]
630 umlal $ACC2,$IN01_2,${R0}[0]
638 add $IN01_1,$IN01_1,$H1
639 umlal $ACC3,$IN01_0,${R3}[0]
640 umlal $ACC4,$IN01_0,${R4}[0]
641 and x4,x8,#0x03ffffff // base 2^64 -> base 2^26
642 umlal $ACC2,$IN01_0,${R2}[0]
643 and x5,x9,#0x03ffffff
644 umlal $ACC0,$IN01_0,${R0}[0]
646 umlal $ACC1,$IN01_0,${R1}[0]
649 add $IN01_3,$IN01_3,$H3
650 add x4,x4,x5,lsl#32 // bfi x4,x5,#32,#32
651 umlal $ACC3,$IN01_1,${R2}[0]
653 umlal $ACC4,$IN01_1,${R3}[0]
655 umlal $ACC0,$IN01_1,${S4}[0]
656 add x6,x6,x7,lsl#32 // bfi x6,x7,#32,#32
657 umlal $ACC2,$IN01_1,${R1}[0]
659 umlal $ACC1,$IN01_1,${R0}[0]
660 and x8,x8,#0x03ffffff
662 add $IN01_4,$IN01_4,$H4
663 and x9,x9,#0x03ffffff
664 umlal $ACC3,$IN01_3,${R0}[0]
666 umlal $ACC0,$IN01_3,${S2}[0]
668 umlal $ACC4,$IN01_3,${R1}[0]
669 add x8,x8,x9,lsl#32 // bfi x8,x9,#32,#32
670 umlal $ACC1,$IN01_3,${S3}[0]
672 umlal $ACC2,$IN01_3,${S4}[0]
673 add x12,$padbit,x12,lsr#40
675 umlal $ACC3,$IN01_4,${S4}[0]
676 add x13,$padbit,x13,lsr#40
677 umlal $ACC0,$IN01_4,${S1}[0]
678 add x10,x10,x11,lsl#32 // bfi x10,x11,#32,#32
679 umlal $ACC4,$IN01_4,${R0}[0]
680 add x12,x12,x13,lsl#32 // bfi x12,x13,#32,#32
681 umlal $ACC1,$IN01_4,${S2}[0]
683 umlal $ACC2,$IN01_4,${S3}[0]
687 /////////////////////////////////////////////////////////////////
688 // lazy reduction as discussed in "NEON crypto" by D.J. Bernstein
691 // [see discussion in poly1305-armv4 module]
693 ushr $T0.2d,$ACC3,#26
695 ushr $T1.2d,$ACC0,#26
696 and $ACC0,$ACC0,$MASK.2d
697 add $ACC4,$ACC4,$T0.2d // h3 -> h4
698 bic $H3,#0xfc,lsl#24 // &=0x03ffffff
699 add $ACC1,$ACC1,$T1.2d // h0 -> h1
701 ushr $T0.2d,$ACC4,#26
703 ushr $T1.2d,$ACC1,#26
706 add $ACC2,$ACC2,$T1.2d // h1 -> h2
708 add $ACC0,$ACC0,$T0.2d
710 shrn $T1.2s,$ACC2,#26
712 add $ACC0,$ACC0,$T0.2d // h4 -> h0
714 add $H3,$H3,$T1.2s // h2 -> h3
717 shrn $T0.2s,$ACC0,#26
722 add $H1,$H1,$T0.2s // h0 -> h1
723 add $H4,$H4,$T1.2s // h3 -> h4
728 dup $IN23_2,${IN23_2}[0]
729 add $IN01_2,$IN01_2,$H2
731 ////////////////////////////////////////////////////////////////
732 // multiply (inp[0:1]+hash) or inp[2:3] by r^2:r^1
737 dup $IN23_2,${IN01_2}[0]
738 add $IN23_0,$IN01_0,$H0
739 add $IN23_3,$IN01_3,$H3
740 add $IN23_1,$IN01_1,$H1
741 add $IN23_4,$IN01_4,$H4
744 dup $IN23_0,${IN23_0}[0]
745 umull2 $ACC0,$IN23_2,${S3}
746 umull2 $ACC3,$IN23_2,${R1}
747 umull2 $ACC4,$IN23_2,${R2}
748 umull2 $ACC2,$IN23_2,${R0}
749 umull2 $ACC1,$IN23_2,${S4}
751 dup $IN23_1,${IN23_1}[0]
752 umlal2 $ACC0,$IN23_0,${R0}
753 umlal2 $ACC2,$IN23_0,${R2}
754 umlal2 $ACC3,$IN23_0,${R3}
755 umlal2 $ACC4,$IN23_0,${R4}
756 umlal2 $ACC1,$IN23_0,${R1}
758 dup $IN23_3,${IN23_3}[0]
759 umlal2 $ACC0,$IN23_1,${S4}
760 umlal2 $ACC3,$IN23_1,${R2}
761 umlal2 $ACC2,$IN23_1,${R1}
762 umlal2 $ACC4,$IN23_1,${R3}
763 umlal2 $ACC1,$IN23_1,${R0}
765 dup $IN23_4,${IN23_4}[0]
766 umlal2 $ACC3,$IN23_3,${R0}
767 umlal2 $ACC4,$IN23_3,${R1}
768 umlal2 $ACC0,$IN23_3,${S2}
769 umlal2 $ACC1,$IN23_3,${S3}
770 umlal2 $ACC2,$IN23_3,${S4}
772 umlal2 $ACC3,$IN23_4,${S4}
773 umlal2 $ACC0,$IN23_4,${S1}
774 umlal2 $ACC4,$IN23_4,${R0}
775 umlal2 $ACC1,$IN23_4,${S2}
776 umlal2 $ACC2,$IN23_4,${S3}
780 ////////////////////////////////////////////////////////////////
781 // (hash+inp[0:1])*r^4:r^3 and accumulate
783 add $IN01_0,$IN01_0,$H0
784 umlal $ACC3,$IN01_2,${R1}
785 umlal $ACC0,$IN01_2,${S3}
786 umlal $ACC4,$IN01_2,${R2}
787 umlal $ACC1,$IN01_2,${S4}
788 umlal $ACC2,$IN01_2,${R0}
790 add $IN01_1,$IN01_1,$H1
791 umlal $ACC3,$IN01_0,${R3}
792 umlal $ACC0,$IN01_0,${R0}
793 umlal $ACC4,$IN01_0,${R4}
794 umlal $ACC1,$IN01_0,${R1}
795 umlal $ACC2,$IN01_0,${R2}
797 add $IN01_3,$IN01_3,$H3
798 umlal $ACC3,$IN01_1,${R2}
799 umlal $ACC0,$IN01_1,${S4}
800 umlal $ACC4,$IN01_1,${R3}
801 umlal $ACC1,$IN01_1,${R0}
802 umlal $ACC2,$IN01_1,${R1}
804 add $IN01_4,$IN01_4,$H4
805 umlal $ACC3,$IN01_3,${R0}
806 umlal $ACC0,$IN01_3,${S2}
807 umlal $ACC4,$IN01_3,${R1}
808 umlal $ACC1,$IN01_3,${S3}
809 umlal $ACC2,$IN01_3,${S4}
811 umlal $ACC3,$IN01_4,${S4}
812 umlal $ACC0,$IN01_4,${S1}
813 umlal $ACC4,$IN01_4,${R0}
814 umlal $ACC1,$IN01_4,${S2}
815 umlal $ACC2,$IN01_4,${S3}
818 ////////////////////////////////////////////////////////////////
821 addp $ACC3,$ACC3,$ACC3
822 ldp d8,d9,[sp,#16] // meet ABI requirements
823 addp $ACC0,$ACC0,$ACC0
825 addp $ACC4,$ACC4,$ACC4
827 addp $ACC1,$ACC1,$ACC1
829 addp $ACC2,$ACC2,$ACC2
831 ////////////////////////////////////////////////////////////////
832 // lazy reduction, but without narrowing
834 ushr $T0.2d,$ACC3,#26
835 and $ACC3,$ACC3,$MASK.2d
836 ushr $T1.2d,$ACC0,#26
837 and $ACC0,$ACC0,$MASK.2d
839 add $ACC4,$ACC4,$T0.2d // h3 -> h4
840 add $ACC1,$ACC1,$T1.2d // h0 -> h1
842 ushr $T0.2d,$ACC4,#26
843 and $ACC4,$ACC4,$MASK.2d
844 ushr $T1.2d,$ACC1,#26
845 and $ACC1,$ACC1,$MASK.2d
846 add $ACC2,$ACC2,$T1.2d // h1 -> h2
848 add $ACC0,$ACC0,$T0.2d
850 ushr $T1.2d,$ACC2,#26
851 and $ACC2,$ACC2,$MASK.2d
852 add $ACC0,$ACC0,$T0.2d // h4 -> h0
853 add $ACC3,$ACC3,$T1.2d // h2 -> h3
855 ushr $T0.2d,$ACC0,#26
856 and $ACC0,$ACC0,$MASK.2d
857 ushr $T1.2d,$ACC3,#26
858 and $ACC3,$ACC3,$MASK.2d
859 add $ACC1,$ACC1,$T0.2d // h0 -> h1
860 add $ACC4,$ACC4,$T1.2d // h3 -> h4
862 ////////////////////////////////////////////////////////////////
863 // write the result, can be partially reduced
865 st4 {$ACC0,$ACC1,$ACC2,$ACC3}[0],[$ctx],#16
866 st1 {$ACC4}[0],[$ctx]
869 .inst 0xd50323bf // autiasp
872 .size poly1305_blocks_neon,.-poly1305_blocks_neon
874 .type poly1305_emit_neon,%function
877 .Lpoly1305_emit_neon:
878 ldr $is_base2_26,[$ctx,#24]
879 cbz $is_base2_26,poly1305_emit
881 ldp w10,w11,[$ctx] // load hash value base 2^26
882 ldp w12,w13,[$ctx,#8]
885 add $h0,x10,x11,lsl#26 // base 2^26 -> base 2^64
887 adds $h0,$h0,x12,lsl#52
888 add $h1,$h1,x13,lsl#14
891 adds $h1,$h1,x14,lsl#40
892 adc $h2,$h2,xzr // can be partially reduced...
894 ldp $t0,$t1,[$nonce] // load nonce
896 and $d0,$h2,#-4 // ... so reduce
897 add $d0,$d0,$h2,lsr#2
903 adds $d0,$h0,#5 // compare to modulus
907 tst $d2,#-4 // see if it's carried/borrowed
913 ror $t0,$t0,#32 // flip nonce words
916 adds $h0,$h0,$t0 // accumulate nonce
919 rev $h0,$h0 // flip output bytes
922 stp $h0,$h1,[$mac] // write result
925 .size poly1305_emit_neon,.-poly1305_emit_neon
929 .long 0,0,0,0,0,0,0,0
930 .asciz "Poly1305 for ARMv8, CRYPTOGAMS by <appro\@openssl.org>"
934 foreach (split("\n",$code)) {
935 s/\b(shrn\s+v[0-9]+)\.[24]d/$1.2s/ or
936 s/\b(fmov\s+)v([0-9]+)[^,]*,\s*x([0-9]+)/$1d$2,x$3/ or
937 (m/\bdup\b/ and (s/\.[24]s/.2d/g or 1)) or
938 (m/\b(eor|and)/ and (s/\.[248][sdh]/.16b/g or 1)) or
939 (m/\bum(ul|la)l\b/ and (s/\.4s/.2s/g or 1)) or
940 (m/\bum(ul|la)l2\b/ and (s/\.2s/.4s/g or 1)) or
941 (m/\bst[1-4]\s+{[^}]+}\[/ and (s/\.[24]d/.s/g or 1));
943 s/\.[124]([sd])\[/.$1\[/;
947 close STDOUT or die "error closing STDOUT: $!";