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;
41 $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
42 ( $xlate="${dir}arm-xlate.pl" and -f $xlate ) or
43 ( $xlate="${dir}../../perlasm/arm-xlate.pl" and -f $xlate) or
44 die "can't locate arm-xlate.pl";
46 open OUT,"| \"$^X\" $xlate $flavour $output";
49 my ($ctx,$inp,$len,$padbit) = map("x$_",(0..3));
50 my ($mac,$nonce)=($inp,$len);
52 my ($h0,$h1,$h2,$r0,$r1,$s1,$t0,$t1,$d0,$d1,$d2) = map("x$_",(4..14));
59 // forward "declarations" are required for Apple
60 .extern OPENSSL_armcap_P
61 .globl poly1305_blocks
65 .type poly1305_init,%function
69 stp xzr,xzr,[$ctx] // zero hash value
70 stp xzr,xzr,[$ctx,#16] // [along with is_base2_26]
75 adrp x17,OPENSSL_armcap_P
76 ldr w17,[x17,#:lo12:OPENSSL_armcap_P]
78 ldp $r0,$r1,[$inp] // load key
79 mov $s1,#0xfffffffc0fffffff
80 movk $s1,#0x0fff,lsl#48
82 rev $r0,$r0 // flip bytes
85 and $r0,$r0,$s1 // &=0ffffffc0fffffff
87 and $r1,$r1,$s1 // &=0ffffffc0ffffffc
88 stp $r0,$r1,[$ctx,#32] // save key value
92 adr $d0,.Lpoly1305_blocks
93 adr $r0,.Lpoly1305_blocks_neon
94 adr $d1,.Lpoly1305_emit
95 adr $r1,.Lpoly1305_emit_neon
109 .size poly1305_init,.-poly1305_init
111 .type poly1305_blocks,%function
118 ldp $h0,$h1,[$ctx] // load hash value
119 ldp $r0,$r1,[$ctx,#32] // load key value
121 add $s1,$r1,$r1,lsr#2 // s1 = r1 + (r1 >> 2)
126 ldp $t0,$t1,[$inp],#16 // load input
132 adds $h0,$h0,$t0 // accumulate input
135 mul $d0,$h0,$r0 // h0*r0
139 mul $t0,$h1,$s1 // h1*5*r1
143 mul $t0,$h0,$r1 // h0*r1
148 mul $t0,$h1,$r0 // h1*r0
153 mul $t0,$h2,$s1 // h2*5*r1
155 mul $t1,$h2,$r0 // h2*r0
160 and $t0,$d2,#-4 // final reduction
162 add $t0,$t0,$d2,lsr#2
169 stp $h0,$h1,[$ctx] // store hash value
174 .size poly1305_blocks,.-poly1305_blocks
176 .type poly1305_emit,%function
180 ldp $h0,$h1,[$ctx] // load hash base 2^64
182 ldp $t0,$t1,[$nonce] // load nonce
184 adds $d0,$h0,#5 // compare to modulus
188 tst $d2,#-4 // see if it's carried/borrowed
194 ror $t0,$t0,#32 // flip nonce words
197 adds $h0,$h0,$t0 // accumulate nonce
200 rev $h0,$h0 // flip output bytes
203 stp $h0,$h1,[$mac] // write result
206 .size poly1305_emit,.-poly1305_emit
208 my ($R0,$R1,$S1,$R2,$S2,$R3,$S3,$R4,$S4) = map("v$_.4s",(0..8));
209 my ($IN01_0,$IN01_1,$IN01_2,$IN01_3,$IN01_4) = map("v$_.2s",(9..13));
210 my ($IN23_0,$IN23_1,$IN23_2,$IN23_3,$IN23_4) = map("v$_.2s",(14..18));
211 my ($ACC0,$ACC1,$ACC2,$ACC3,$ACC4) = map("v$_.2d",(19..23));
212 my ($H0,$H1,$H2,$H3,$H4) = map("v$_.2s",(24..28));
213 my ($T0,$T1,$MASK) = map("v$_",(29..31));
215 my ($in2,$zeros)=("x16","x17");
216 my $is_base2_26 = $zeros; # borrow
219 .type poly1305_mult,%function
222 mul $d0,$h0,$r0 // h0*r0
225 mul $t0,$h1,$s1 // h1*5*r1
229 mul $t0,$h0,$r1 // h0*r1
234 mul $t0,$h1,$r0 // h1*r0
239 mul $t0,$h2,$s1 // h2*5*r1
241 mul $t1,$h2,$r0 // h2*r0
246 and $t0,$d2,#-4 // final reduction
248 add $t0,$t0,$d2,lsr#2
254 .size poly1305_mult,.-poly1305_mult
256 .type poly1305_splat,%function
259 and x12,$h0,#0x03ffffff // base 2^64 -> base 2^26
262 and x14,x14,#0x03ffffff
266 str w12,[$ctx,#16*0] // r0
267 add w12,w13,w13,lsl#2 // r1*5
268 str w13,[$ctx,#16*1] // r1
269 add w13,w14,w14,lsl#2 // r2*5
270 str w12,[$ctx,#16*2] // s1
271 str w14,[$ctx,#16*3] // r2
272 add w14,w15,w15,lsl#2 // r3*5
273 str w13,[$ctx,#16*4] // s2
274 str w15,[$ctx,#16*5] // r3
275 add w15,w16,w16,lsl#2 // r4*5
276 str w14,[$ctx,#16*6] // s3
277 str w16,[$ctx,#16*7] // r4
278 str w15,[$ctx,#16*8] // s4
281 .size poly1305_splat,.-poly1305_splat
283 .type poly1305_blocks_neon,%function
285 poly1305_blocks_neon:
286 .Lpoly1305_blocks_neon:
287 ldr $is_base2_26,[$ctx,#24]
290 cbz $is_base2_26,.Lpoly1305_blocks
293 .inst 0xd503233f // paciasp
294 stp x29,x30,[sp,#-80]!
300 cbz $is_base2_26,.Lbase2_64_neon
302 ldp w10,w11,[$ctx] // load hash value base 2^26
303 ldp w12,w13,[$ctx,#8]
309 ldp $r0,$r1,[$ctx,#32] // load key value
311 add $h0,x10,x11,lsl#26 // base 2^26 -> base 2^64
313 adds $h0,$h0,x12,lsl#52
314 add $h1,$h1,x13,lsl#14
317 adds $h1,$h1,x14,lsl#40
318 adc $d2,$h2,xzr // can be partially reduced...
320 ldp $d0,$d1,[$inp],#16 // load input
322 add $s1,$r1,$r1,lsr#2 // s1 = r1 + (r1 >> 2)
324 and $t0,$d2,#-4 // ... so reduce
326 add $t0,$t0,$d2,lsr#2
335 adds $h0,$h0,$d0 // accumulate input
342 cbz $padbit,.Lstore_base2_64_neon
344 and x10,$h0,#0x03ffffff // base 2^64 -> base 2^26
347 and x12,x12,#0x03ffffff
351 cbnz $len,.Leven_neon
353 stp w10,w11,[$ctx] // store hash value base 2^26
354 stp w12,w13,[$ctx,#8]
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
366 ldp $r0,$r1,[$ctx,#32] // load key value
368 ldp $h0,$h1,[$ctx] // load hash value base 2^64
374 ldp $d0,$d1,[$inp],#16 // load input
376 add $s1,$r1,$r1,lsr#2 // s1 = r1 + (r1 >> 2)
381 adds $h0,$h0,$d0 // accumulate input
388 and x10,$h0,#0x03ffffff // base 2^64 -> base 2^26
391 and x12,x12,#0x03ffffff
395 stp d8,d9,[sp,#16] // meet ABI requirements
406 ////////////////////////////////// initialize r^n table
408 add $s1,$r1,$r1,lsr#2 // s1 = r1 + (r1 >> 2)
414 bl poly1305_mult // r^2
418 bl poly1305_mult // r^3
422 bl poly1305_mult // r^4
430 csel $in2,$zeros,$in2,lo
433 stur x4,[$ctx,#-24] // set is_base2_26
434 sub $ctx,$ctx,#48 // restore original $ctx
442 csel $in2,$zeros,$in2,lo
444 stp d8,d9,[sp,#16] // meet ABI requirements
456 ldp x8,x12,[$in2],#16 // inp[2:3] (or zero)
457 ldp x9,x13,[$in2],#48
459 lsl $padbit,$padbit,#24
468 and x4,x8,#0x03ffffff // base 2^64 -> base 2^26
469 and x5,x9,#0x03ffffff
472 add x4,x4,x5,lsl#32 // bfi x4,x5,#32,#32
475 add x6,x6,x7,lsl#32 // bfi x6,x7,#32,#32
477 and x8,x8,#0x03ffffff
478 and x9,x9,#0x03ffffff
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
485 add x10,x10,x11,lsl#32 // bfi x10,x11,#32,#32
486 add x12,x12,x13,lsl#32 // bfi x12,x13,#32,#32
491 ldp x8,x12,[$inp],#16 // inp[0:1]
492 ldp x9,x13,[$inp],#48
494 ld1 {$R0,$R1,$S1,$R2},[x15],#64
495 ld1 {$S2,$R3,$S3,$R4},[x15],#64
504 and x4,x8,#0x03ffffff // base 2^64 -> base 2^26
505 and x5,x9,#0x03ffffff
508 add x4,x4,x5,lsl#32 // bfi x4,x5,#32,#32
511 add x6,x6,x7,lsl#32 // bfi x6,x7,#32,#32
513 and x8,x8,#0x03ffffff
514 and x9,x9,#0x03ffffff
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
521 add x10,x10,x11,lsl#32 // bfi x10,x11,#32,#32
522 add x12,x12,x13,lsl#32 // bfi x12,x13,#32,#32
527 ushr $MASK.2d,$MASK.2d,#38
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 // \___________________/ \____________________/
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
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]
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]
572 umlal $ACC1,$IN23_1,${R0}[2]
574 umlal $ACC0,$IN23_1,${S4}[2]
575 add x4,x4,x5,lsl#32 // bfi x4,x5,#32,#32
577 umlal $ACC4,$IN23_2,${R2}[2]
579 umlal $ACC3,$IN23_2,${R1}[2]
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]
585 umlal $ACC0,$IN23_2,${S3}[2]
586 and x8,x8,#0x03ffffff
588 umlal $ACC4,$IN23_3,${R1}[2]
589 and x9,x9,#0x03ffffff
590 umlal $ACC3,$IN23_3,${R0}[2]
592 umlal $ACC2,$IN23_3,${S4}[2]
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]
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]
609 umlal $ACC0,$IN23_4,${S1}[2]
612 ////////////////////////////////////////////////////////////////
613 // (hash+inp[0:1])*r^4 and accumulate
615 add $IN01_0,$IN01_0,$H0
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]
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]
639 umlal $ACC1,$IN01_0,${R1}[0]
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]
646 umlal $ACC4,$IN01_1,${R3}[0]
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]
652 umlal $ACC1,$IN01_1,${R0}[0]
653 and x8,x8,#0x03ffffff
655 add $IN01_4,$IN01_4,$H4
656 and x9,x9,#0x03ffffff
657 umlal $ACC3,$IN01_3,${R0}[0]
659 umlal $ACC0,$IN01_3,${S2}[0]
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]
665 umlal $ACC2,$IN01_3,${S4}[0]
666 add x12,$padbit,x12,lsr#40
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]
676 umlal $ACC2,$IN01_4,${S3}[0]
680 /////////////////////////////////////////////////////////////////
681 // lazy reduction as discussed in "NEON crypto" by D.J. Bernstein
684 // [see discussion in poly1305-armv4 module]
686 ushr $T0.2d,$ACC3,#26
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
694 ushr $T0.2d,$ACC4,#26
696 ushr $T1.2d,$ACC1,#26
699 add $ACC2,$ACC2,$T1.2d // h1 -> h2
701 add $ACC0,$ACC0,$T0.2d
703 shrn $T1.2s,$ACC2,#26
705 add $ACC0,$ACC0,$T0.2d // h4 -> h0
707 add $H3,$H3,$T1.2s // h2 -> h3
710 shrn $T0.2s,$ACC0,#26
715 add $H1,$H1,$T0.2s // h0 -> h1
716 add $H4,$H4,$T1.2s // h3 -> h4
721 dup $IN23_2,${IN23_2}[0]
722 add $IN01_2,$IN01_2,$H2
724 ////////////////////////////////////////////////////////////////
725 // multiply (inp[0:1]+hash) or inp[2:3] by r^2:r^1
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
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}
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}
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}
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}
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}
773 ////////////////////////////////////////////////////////////////
774 // (hash+inp[0:1])*r^4:r^3 and accumulate
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}
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}
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}
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}
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}
811 ////////////////////////////////////////////////////////////////
814 addp $ACC3,$ACC3,$ACC3
815 ldp d8,d9,[sp,#16] // meet ABI requirements
816 addp $ACC0,$ACC0,$ACC0
818 addp $ACC4,$ACC4,$ACC4
820 addp $ACC1,$ACC1,$ACC1
822 addp $ACC2,$ACC2,$ACC2
824 ////////////////////////////////////////////////////////////////
825 // lazy reduction, but without narrowing
827 ushr $T0.2d,$ACC3,#26
828 and $ACC3,$ACC3,$MASK.2d
829 ushr $T1.2d,$ACC0,#26
830 and $ACC0,$ACC0,$MASK.2d
832 add $ACC4,$ACC4,$T0.2d // h3 -> h4
833 add $ACC1,$ACC1,$T1.2d // h0 -> h1
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
841 add $ACC0,$ACC0,$T0.2d
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
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,$T1.2d // h3 -> h4
855 ////////////////////////////////////////////////////////////////
856 // write the result, can be partially reduced
858 st4 {$ACC0,$ACC1,$ACC2,$ACC3}[0],[$ctx],#16
859 st1 {$ACC4}[0],[$ctx]
862 .inst 0xd50323bf // autiasp
865 .size poly1305_blocks_neon,.-poly1305_blocks_neon
867 .type poly1305_emit_neon,%function
870 .Lpoly1305_emit_neon:
871 ldr $is_base2_26,[$ctx,#24]
872 cbz $is_base2_26,poly1305_emit
874 ldp w10,w11,[$ctx] // load hash value base 2^26
875 ldp w12,w13,[$ctx,#8]
878 add $h0,x10,x11,lsl#26 // base 2^26 -> base 2^64
880 adds $h0,$h0,x12,lsl#52
881 add $h1,$h1,x13,lsl#14
884 adds $h1,$h1,x14,lsl#40
885 adc $h2,$h2,xzr // can be partially reduced...
887 ldp $t0,$t1,[$nonce] // load nonce
889 and $d0,$h2,#-4 // ... so reduce
890 add $d0,$d0,$h2,lsr#2
896 adds $d0,$h0,#5 // compare to modulus
900 tst $d2,#-4 // see if it's carried/borrowed
906 ror $t0,$t0,#32 // flip nonce words
909 adds $h0,$h0,$t0 // accumulate nonce
912 rev $h0,$h0 // flip output bytes
915 stp $h0,$h1,[$mac] // write result
918 .size poly1305_emit_neon,.-poly1305_emit_neon
922 .long 0,0,0,0,0,0,0,0
923 .asciz "Poly1305 for ARMv8, CRYPTOGAMS by <appro\@openssl.org>"
927 foreach (split("\n",$code)) {
928 s/\b(shrn\s+v[0-9]+)\.[24]d/$1.2s/ or
929 s/\b(fmov\s+)v([0-9]+)[^,]*,\s*x([0-9]+)/$1d$2,x$3/ or
930 (m/\bdup\b/ and (s/\.[24]s/.2d/g or 1)) or
931 (m/\b(eor|and)/ and (s/\.[248][sdh]/.16b/g or 1)) or
932 (m/\bum(ul|la)l\b/ and (s/\.4s/.2s/g or 1)) or
933 (m/\bum(ul|la)l2\b/ and (s/\.2s/.4s/g or 1)) or
934 (m/\bst[1-4]\s+{[^}]+}\[/ and (s/\.[24]d/.s/g or 1));
936 s/\.[124]([sd])\[/.$1\[/;