2 # SPDX-License-Identifier: GPL-2.0 OR BSD-3-Clause
4 # This code is taken from the OpenSSL project but the author, Andy Polyakov,
5 # has relicensed it under the licenses specified in the SPDX header above.
6 # The original headers, including the original license headers, are
7 # included below for completeness.
9 # ====================================================================
10 # Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
11 # project. The module is, however, dual licensed under OpenSSL and
12 # CRYPTOGAMS licenses depending on where you obtain it. For further
13 # details see http://www.openssl.org/~appro/cryptogams/.
14 # ====================================================================
16 # This module implements Poly1305 hash for ARMv8.
20 # Numbers are cycles per processed byte with poly1305_blocks alone.
24 # Apple A7 1.86/+5% 0.72
25 # Cortex-A53 2.69/+58% 1.47
26 # Cortex-A57 2.70/+7% 1.14
27 # Denver 1.64/+50% 1.18(*)
28 # X-Gene 2.13/+68% 2.27
29 # Mongoose 1.77/+75% 1.12
32 # (*) estimate based on resources availability is less than 1.0,
33 # i.e. measured result is worse than expected, presumably binary
34 # translator is not almighty;
37 if ($flavour=~/\w[\w\-]*\.\w+$/) { $output=$flavour; undef $flavour; }
38 else { while (($output=shift) && ($output!~/\w[\w\-]*\.\w+$/)) {} }
40 if ($flavour && $flavour ne "void") {
41 $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
42 ( $xlate="${dir}arm-xlate.pl" and -f $xlate ) or
43 ( $xlate="${dir}../../perlasm/arm-xlate.pl" and -f $xlate) or
44 die "can't locate arm-xlate.pl";
46 open STDOUT,"| \"$^X\" $xlate $flavour $output";
48 open STDOUT,">$output";
51 my ($ctx,$inp,$len,$padbit) = map("x$_",(0..3));
52 my ($mac,$nonce)=($inp,$len);
54 my ($h0,$h1,$h2,$r0,$r1,$s1,$t0,$t1,$d0,$d1,$d2) = map("x$_",(4..14));
58 # include "arm_arch.h"
59 .extern OPENSSL_armcap_P
61 # define poly1305_init poly1305_init_arm
62 # define poly1305_blocks poly1305_blocks_arm
63 # define poly1305_emit poly1305_emit_arm
68 // forward "declarations" are required for Apple
69 .globl poly1305_blocks
72 .type poly1305_init,%function
76 stp xzr,xzr,[$ctx] // zero hash value
77 stp xzr,xzr,[$ctx,#16] // [along with is_base2_26]
84 ldrsw $t1,.LOPENSSL_armcap_P
86 ldr $t1,.LOPENSSL_armcap_P
88 adr $t0,.LOPENSSL_armcap_P
92 ldp $r0,$r1,[$inp] // load key
93 mov $s1,#0xfffffffc0fffffff
94 movk $s1,#0x0fff,lsl#48
96 rev $r0,$r0 // flip bytes
99 and $r0,$r0,$s1 // &=0ffffffc0fffffff
101 and $r1,$r1,$s1 // &=0ffffffc0ffffffc
102 stp $r0,$r1,[$ctx,#32] // save key value
107 adr $d0,poly1305_blocks
108 adr $r0,poly1305_blocks_neon
109 adr $d1,poly1305_emit
110 adr $r1,poly1305_emit_neon
127 .size poly1305_init,.-poly1305_init
129 .type poly1305_blocks,%function
135 ldp $h0,$h1,[$ctx] // load hash value
136 ldp $r0,$r1,[$ctx,#32] // load key value
138 add $s1,$r1,$r1,lsr#2 // s1 = r1 + (r1 >> 2)
143 ldp $t0,$t1,[$inp],#16 // load input
149 adds $h0,$h0,$t0 // accumulate input
152 mul $d0,$h0,$r0 // h0*r0
156 mul $t0,$h1,$s1 // h1*5*r1
160 mul $t0,$h0,$r1 // h0*r1
165 mul $t0,$h1,$r0 // h1*r0
170 mul $t0,$h2,$s1 // h2*5*r1
172 mul $t1,$h2,$r0 // h2*r0
177 and $t0,$d2,#-4 // final reduction
179 add $t0,$t0,$d2,lsr#2
186 stp $h0,$h1,[$ctx] // store hash value
191 .size poly1305_blocks,.-poly1305_blocks
193 .type poly1305_emit,%function
196 ldp $h0,$h1,[$ctx] // load hash base 2^64
198 ldp $t0,$t1,[$nonce] // load nonce
200 adds $d0,$h0,#5 // compare to modulus
204 tst $d2,#-4 // see if it's carried/borrowed
210 ror $t0,$t0,#32 // flip nonce words
213 adds $h0,$h0,$t0 // accumulate nonce
216 rev $h0,$h0 // flip output bytes
219 stp $h0,$h1,[$mac] // write result
222 .size poly1305_emit,.-poly1305_emit
224 my ($R0,$R1,$S1,$R2,$S2,$R3,$S3,$R4,$S4) = map("v$_.4s",(0..8));
225 my ($IN01_0,$IN01_1,$IN01_2,$IN01_3,$IN01_4) = map("v$_.2s",(9..13));
226 my ($IN23_0,$IN23_1,$IN23_2,$IN23_3,$IN23_4) = map("v$_.2s",(14..18));
227 my ($ACC0,$ACC1,$ACC2,$ACC3,$ACC4) = map("v$_.2d",(19..23));
228 my ($H0,$H1,$H2,$H3,$H4) = map("v$_.2s",(24..28));
229 my ($T0,$T1,$MASK) = map("v$_",(29..31));
231 my ($in2,$zeros)=("x16","x17");
232 my $is_base2_26 = $zeros; # borrow
235 .type __poly1305_mult,%function
238 mul $d0,$h0,$r0 // h0*r0
241 mul $t0,$h1,$s1 // h1*5*r1
245 mul $t0,$h0,$r1 // h0*r1
250 mul $t0,$h1,$r0 // h1*r0
255 mul $t0,$h2,$s1 // h2*5*r1
257 mul $t1,$h2,$r0 // h2*r0
262 and $t0,$d2,#-4 // final reduction
264 add $t0,$t0,$d2,lsr#2
270 .size __poly1305_mult,.-__poly1305_mult
272 .type __poly1305_splat,%function
275 and x12,$h0,#0x03ffffff // base 2^64 -> base 2^26
278 and x14,x14,#0x03ffffff
282 str w12,[$ctx,#16*0] // r0
283 add w12,w13,w13,lsl#2 // r1*5
284 str w13,[$ctx,#16*1] // r1
285 add w13,w14,w14,lsl#2 // r2*5
286 str w12,[$ctx,#16*2] // s1
287 str w14,[$ctx,#16*3] // r2
288 add w14,w15,w15,lsl#2 // r3*5
289 str w13,[$ctx,#16*4] // s2
290 str w15,[$ctx,#16*5] // r3
291 add w15,w16,w16,lsl#2 // r4*5
292 str w14,[$ctx,#16*6] // s3
293 str w16,[$ctx,#16*7] // r4
294 str w15,[$ctx,#16*8] // s4
297 .size __poly1305_splat,.-__poly1305_splat
299 #if !defined(__KERNEL__) || defined(CONFIG_KERNEL_MODE_NEON)
301 .globl poly1305_blocks_neon
302 .globl poly1305_emit_neon
305 .type poly1305_blocks_neon,%function
307 poly1305_blocks_neon:
308 ldr $is_base2_26,[$ctx,#24]
311 cbz $is_base2_26,poly1305_blocks
314 stp x29,x30,[sp,#-80]!
320 cbz $is_base2_26,.Lbase2_64_neon
322 ldp w10,w11,[$ctx] // load hash value base 2^26
323 ldp w12,w13,[$ctx,#8]
329 ldp $r0,$r1,[$ctx,#32] // load key value
331 add $h0,x10,x11,lsl#26 // base 2^26 -> base 2^64
333 adds $h0,$h0,x12,lsl#52
334 add $h1,$h1,x13,lsl#14
337 adds $h1,$h1,x14,lsl#40
338 adc $d2,$h2,xzr // can be partially reduced...
340 ldp $d0,$d1,[$inp],#16 // load input
342 add $s1,$r1,$r1,lsr#2 // s1 = r1 + (r1 >> 2)
344 and $t0,$d2,#-4 // ... so reduce
346 add $t0,$t0,$d2,lsr#2
355 adds $h0,$h0,$d0 // accumulate input
362 cbz $padbit,.Lstore_base2_64_neon
364 and x10,$h0,#0x03ffffff // base 2^64 -> base 2^26
367 and x12,x12,#0x03ffffff
371 cbnz $len,.Leven_neon
373 stp w10,w11,[$ctx] // store hash value base 2^26
374 stp w12,w13,[$ctx,#8]
379 .Lstore_base2_64_neon:
380 stp $h0,$h1,[$ctx] // store hash value base 2^64
381 stp $h2,xzr,[$ctx,#16] // note that is_base2_26 is zeroed
386 ldp $r0,$r1,[$ctx,#32] // load key value
388 ldp $h0,$h1,[$ctx] // load hash value base 2^64
394 ldp $d0,$d1,[$inp],#16 // load input
396 add $s1,$r1,$r1,lsr#2 // s1 = r1 + (r1 >> 2)
401 adds $h0,$h0,$d0 // accumulate input
408 and x10,$h0,#0x03ffffff // base 2^64 -> base 2^26
411 and x12,x12,#0x03ffffff
415 stp d8,d9,[sp,#16] // meet ABI requirements
426 ////////////////////////////////// initialize r^n table
428 add $s1,$r1,$r1,lsr#2 // s1 = r1 + (r1 >> 2)
434 bl __poly1305_mult // r^2
438 bl __poly1305_mult // r^3
442 bl __poly1305_mult // r^4
450 csel $in2,$zeros,$in2,lo
453 str x4,[$ctx,#-24] // set is_base2_26
454 sub $ctx,$ctx,#48 // restore original $ctx
462 csel $in2,$zeros,$in2,lo
464 stp d8,d9,[sp,#16] // meet ABI requirements
476 ldp x8,x12,[$in2],#16 // inp[2:3] (or zero)
477 ldp x9,x13,[$in2],#48
479 lsl $padbit,$padbit,#24
488 and x4,x8,#0x03ffffff // base 2^64 -> base 2^26
489 and x5,x9,#0x03ffffff
492 add x4,x4,x5,lsl#32 // bfi x4,x5,#32,#32
495 add x6,x6,x7,lsl#32 // bfi x6,x7,#32,#32
497 and x8,x8,#0x03ffffff
498 and x9,x9,#0x03ffffff
501 add x12,$padbit,x12,lsr#40
502 add x13,$padbit,x13,lsr#40
503 add x8,x8,x9,lsl#32 // bfi x8,x9,#32,#32
505 add x10,x10,x11,lsl#32 // bfi x10,x11,#32,#32
506 add x12,x12,x13,lsl#32 // bfi x12,x13,#32,#32
511 ldp x8,x12,[$inp],#16 // inp[0:1]
512 ldp x9,x13,[$inp],#48
514 ld1 {$R0,$R1,$S1,$R2},[x15],#64
515 ld1 {$S2,$R3,$S3,$R4},[x15],#64
524 and x4,x8,#0x03ffffff // base 2^64 -> base 2^26
525 and x5,x9,#0x03ffffff
528 add x4,x4,x5,lsl#32 // bfi x4,x5,#32,#32
531 add x6,x6,x7,lsl#32 // bfi x6,x7,#32,#32
533 and x8,x8,#0x03ffffff
534 and x9,x9,#0x03ffffff
537 add x12,$padbit,x12,lsr#40
538 add x13,$padbit,x13,lsr#40
539 add x8,x8,x9,lsl#32 // bfi x8,x9,#32,#32
541 add x10,x10,x11,lsl#32 // bfi x10,x11,#32,#32
542 add x12,x12,x13,lsl#32 // bfi x12,x13,#32,#32
547 ushr $MASK.2d,$MASK.2d,#38
553 ////////////////////////////////////////////////////////////////
554 // ((inp[0]*r^4+inp[2]*r^2+inp[4])*r^4+inp[6]*r^2
555 // ((inp[1]*r^4+inp[3]*r^2+inp[5])*r^3+inp[7]*r
556 // \___________________/
557 // ((inp[0]*r^4+inp[2]*r^2+inp[4])*r^4+inp[6]*r^2+inp[8])*r^2
558 // ((inp[1]*r^4+inp[3]*r^2+inp[5])*r^4+inp[7]*r^2+inp[9])*r
559 // \___________________/ \____________________/
561 // Note that we start with inp[2:3]*r^2. This is because it
562 // doesn't depend on reduction in previous iteration.
563 ////////////////////////////////////////////////////////////////
564 // d4 = h0*r4 + h1*r3 + h2*r2 + h3*r1 + h4*r0
565 // d3 = h0*r3 + h1*r2 + h2*r1 + h3*r0 + h4*5*r4
566 // d2 = h0*r2 + h1*r1 + h2*r0 + h3*5*r4 + h4*5*r3
567 // d1 = h0*r1 + h1*r0 + h2*5*r4 + h3*5*r3 + h4*5*r2
568 // d0 = h0*r0 + h1*5*r4 + h2*5*r3 + h3*5*r2 + h4*5*r1
571 umull $ACC4,$IN23_0,${R4}[2]
572 csel $in2,$zeros,$in2,lo
573 umull $ACC3,$IN23_0,${R3}[2]
574 umull $ACC2,$IN23_0,${R2}[2]
575 ldp x8,x12,[$in2],#16 // inp[2:3] (or zero)
576 umull $ACC1,$IN23_0,${R1}[2]
577 ldp x9,x13,[$in2],#48
578 umull $ACC0,$IN23_0,${R0}[2]
586 umlal $ACC4,$IN23_1,${R3}[2]
587 and x4,x8,#0x03ffffff // base 2^64 -> base 2^26
588 umlal $ACC3,$IN23_1,${R2}[2]
589 and x5,x9,#0x03ffffff
590 umlal $ACC2,$IN23_1,${R1}[2]
592 umlal $ACC1,$IN23_1,${R0}[2]
594 umlal $ACC0,$IN23_1,${S4}[2]
595 add x4,x4,x5,lsl#32 // bfi x4,x5,#32,#32
597 umlal $ACC4,$IN23_2,${R2}[2]
599 umlal $ACC3,$IN23_2,${R1}[2]
601 umlal $ACC2,$IN23_2,${R0}[2]
602 add x6,x6,x7,lsl#32 // bfi x6,x7,#32,#32
603 umlal $ACC1,$IN23_2,${S4}[2]
605 umlal $ACC0,$IN23_2,${S3}[2]
606 and x8,x8,#0x03ffffff
608 umlal $ACC4,$IN23_3,${R1}[2]
609 and x9,x9,#0x03ffffff
610 umlal $ACC3,$IN23_3,${R0}[2]
612 umlal $ACC2,$IN23_3,${S4}[2]
614 umlal $ACC1,$IN23_3,${S3}[2]
615 add x8,x8,x9,lsl#32 // bfi x8,x9,#32,#32
616 umlal $ACC0,$IN23_3,${S2}[2]
619 add $IN01_2,$IN01_2,$H2
620 add x12,$padbit,x12,lsr#40
621 umlal $ACC4,$IN23_4,${R0}[2]
622 add x13,$padbit,x13,lsr#40
623 umlal $ACC3,$IN23_4,${S4}[2]
624 add x10,x10,x11,lsl#32 // bfi x10,x11,#32,#32
625 umlal $ACC2,$IN23_4,${S3}[2]
626 add x12,x12,x13,lsl#32 // bfi x12,x13,#32,#32
627 umlal $ACC1,$IN23_4,${S2}[2]
629 umlal $ACC0,$IN23_4,${S1}[2]
632 ////////////////////////////////////////////////////////////////
633 // (hash+inp[0:1])*r^4 and accumulate
635 add $IN01_0,$IN01_0,$H0
637 umlal $ACC3,$IN01_2,${R1}[0]
638 ldp x8,x12,[$inp],#16 // inp[0:1]
639 umlal $ACC0,$IN01_2,${S3}[0]
640 ldp x9,x13,[$inp],#48
641 umlal $ACC4,$IN01_2,${R2}[0]
642 umlal $ACC1,$IN01_2,${S4}[0]
643 umlal $ACC2,$IN01_2,${R0}[0]
651 add $IN01_1,$IN01_1,$H1
652 umlal $ACC3,$IN01_0,${R3}[0]
653 umlal $ACC4,$IN01_0,${R4}[0]
654 and x4,x8,#0x03ffffff // base 2^64 -> base 2^26
655 umlal $ACC2,$IN01_0,${R2}[0]
656 and x5,x9,#0x03ffffff
657 umlal $ACC0,$IN01_0,${R0}[0]
659 umlal $ACC1,$IN01_0,${R1}[0]
662 add $IN01_3,$IN01_3,$H3
663 add x4,x4,x5,lsl#32 // bfi x4,x5,#32,#32
664 umlal $ACC3,$IN01_1,${R2}[0]
666 umlal $ACC4,$IN01_1,${R3}[0]
668 umlal $ACC0,$IN01_1,${S4}[0]
669 add x6,x6,x7,lsl#32 // bfi x6,x7,#32,#32
670 umlal $ACC2,$IN01_1,${R1}[0]
672 umlal $ACC1,$IN01_1,${R0}[0]
673 and x8,x8,#0x03ffffff
675 add $IN01_4,$IN01_4,$H4
676 and x9,x9,#0x03ffffff
677 umlal $ACC3,$IN01_3,${R0}[0]
679 umlal $ACC0,$IN01_3,${S2}[0]
681 umlal $ACC4,$IN01_3,${R1}[0]
682 add x8,x8,x9,lsl#32 // bfi x8,x9,#32,#32
683 umlal $ACC1,$IN01_3,${S3}[0]
685 umlal $ACC2,$IN01_3,${S4}[0]
686 add x12,$padbit,x12,lsr#40
688 umlal $ACC3,$IN01_4,${S4}[0]
689 add x13,$padbit,x13,lsr#40
690 umlal $ACC0,$IN01_4,${S1}[0]
691 add x10,x10,x11,lsl#32 // bfi x10,x11,#32,#32
692 umlal $ACC4,$IN01_4,${R0}[0]
693 add x12,x12,x13,lsl#32 // bfi x12,x13,#32,#32
694 umlal $ACC1,$IN01_4,${S2}[0]
696 umlal $ACC2,$IN01_4,${S3}[0]
700 /////////////////////////////////////////////////////////////////
701 // lazy reduction as discussed in "NEON crypto" by D.J. Bernstein
704 // [see discussion in poly1305-armv4 module]
706 ushr $T0.2d,$ACC3,#26
708 ushr $T1.2d,$ACC0,#26
709 and $ACC0,$ACC0,$MASK.2d
710 add $ACC4,$ACC4,$T0.2d // h3 -> h4
711 bic $H3,#0xfc,lsl#24 // &=0x03ffffff
712 add $ACC1,$ACC1,$T1.2d // h0 -> h1
714 ushr $T0.2d,$ACC4,#26
716 ushr $T1.2d,$ACC1,#26
719 add $ACC2,$ACC2,$T1.2d // h1 -> h2
721 add $ACC0,$ACC0,$T0.2d
723 shrn $T1.2s,$ACC2,#26
725 add $ACC0,$ACC0,$T0.2d // h4 -> h0
727 add $H3,$H3,$T1.2s // h2 -> h3
730 shrn $T0.2s,$ACC0,#26
735 add $H1,$H1,$T0.2s // h0 -> h1
736 add $H4,$H4,$T1.2s // h3 -> h4
741 dup $IN23_2,${IN23_2}[0]
742 add $IN01_2,$IN01_2,$H2
744 ////////////////////////////////////////////////////////////////
745 // multiply (inp[0:1]+hash) or inp[2:3] by r^2:r^1
750 dup $IN23_2,${IN01_2}[0]
751 add $IN23_0,$IN01_0,$H0
752 add $IN23_3,$IN01_3,$H3
753 add $IN23_1,$IN01_1,$H1
754 add $IN23_4,$IN01_4,$H4
757 dup $IN23_0,${IN23_0}[0]
758 umull2 $ACC0,$IN23_2,${S3}
759 umull2 $ACC3,$IN23_2,${R1}
760 umull2 $ACC4,$IN23_2,${R2}
761 umull2 $ACC2,$IN23_2,${R0}
762 umull2 $ACC1,$IN23_2,${S4}
764 dup $IN23_1,${IN23_1}[0]
765 umlal2 $ACC0,$IN23_0,${R0}
766 umlal2 $ACC2,$IN23_0,${R2}
767 umlal2 $ACC3,$IN23_0,${R3}
768 umlal2 $ACC4,$IN23_0,${R4}
769 umlal2 $ACC1,$IN23_0,${R1}
771 dup $IN23_3,${IN23_3}[0]
772 umlal2 $ACC0,$IN23_1,${S4}
773 umlal2 $ACC3,$IN23_1,${R2}
774 umlal2 $ACC2,$IN23_1,${R1}
775 umlal2 $ACC4,$IN23_1,${R3}
776 umlal2 $ACC1,$IN23_1,${R0}
778 dup $IN23_4,${IN23_4}[0]
779 umlal2 $ACC3,$IN23_3,${R0}
780 umlal2 $ACC4,$IN23_3,${R1}
781 umlal2 $ACC0,$IN23_3,${S2}
782 umlal2 $ACC1,$IN23_3,${S3}
783 umlal2 $ACC2,$IN23_3,${S4}
785 umlal2 $ACC3,$IN23_4,${S4}
786 umlal2 $ACC0,$IN23_4,${S1}
787 umlal2 $ACC4,$IN23_4,${R0}
788 umlal2 $ACC1,$IN23_4,${S2}
789 umlal2 $ACC2,$IN23_4,${S3}
793 ////////////////////////////////////////////////////////////////
794 // (hash+inp[0:1])*r^4:r^3 and accumulate
796 add $IN01_0,$IN01_0,$H0
797 umlal $ACC3,$IN01_2,${R1}
798 umlal $ACC0,$IN01_2,${S3}
799 umlal $ACC4,$IN01_2,${R2}
800 umlal $ACC1,$IN01_2,${S4}
801 umlal $ACC2,$IN01_2,${R0}
803 add $IN01_1,$IN01_1,$H1
804 umlal $ACC3,$IN01_0,${R3}
805 umlal $ACC0,$IN01_0,${R0}
806 umlal $ACC4,$IN01_0,${R4}
807 umlal $ACC1,$IN01_0,${R1}
808 umlal $ACC2,$IN01_0,${R2}
810 add $IN01_3,$IN01_3,$H3
811 umlal $ACC3,$IN01_1,${R2}
812 umlal $ACC0,$IN01_1,${S4}
813 umlal $ACC4,$IN01_1,${R3}
814 umlal $ACC1,$IN01_1,${R0}
815 umlal $ACC2,$IN01_1,${R1}
817 add $IN01_4,$IN01_4,$H4
818 umlal $ACC3,$IN01_3,${R0}
819 umlal $ACC0,$IN01_3,${S2}
820 umlal $ACC4,$IN01_3,${R1}
821 umlal $ACC1,$IN01_3,${S3}
822 umlal $ACC2,$IN01_3,${S4}
824 umlal $ACC3,$IN01_4,${S4}
825 umlal $ACC0,$IN01_4,${S1}
826 umlal $ACC4,$IN01_4,${R0}
827 umlal $ACC1,$IN01_4,${S2}
828 umlal $ACC2,$IN01_4,${S3}
831 ////////////////////////////////////////////////////////////////
834 addp $ACC3,$ACC3,$ACC3
835 ldp d8,d9,[sp,#16] // meet ABI requirements
836 addp $ACC0,$ACC0,$ACC0
838 addp $ACC4,$ACC4,$ACC4
840 addp $ACC1,$ACC1,$ACC1
842 addp $ACC2,$ACC2,$ACC2
844 ////////////////////////////////////////////////////////////////
845 // lazy reduction, but without narrowing
847 ushr $T0.2d,$ACC3,#26
848 and $ACC3,$ACC3,$MASK.2d
849 ushr $T1.2d,$ACC0,#26
850 and $ACC0,$ACC0,$MASK.2d
852 add $ACC4,$ACC4,$T0.2d // h3 -> h4
853 add $ACC1,$ACC1,$T1.2d // h0 -> h1
855 ushr $T0.2d,$ACC4,#26
856 and $ACC4,$ACC4,$MASK.2d
857 ushr $T1.2d,$ACC1,#26
858 and $ACC1,$ACC1,$MASK.2d
859 add $ACC2,$ACC2,$T1.2d // h1 -> h2
861 add $ACC0,$ACC0,$T0.2d
863 ushr $T1.2d,$ACC2,#26
864 and $ACC2,$ACC2,$MASK.2d
865 add $ACC0,$ACC0,$T0.2d // h4 -> h0
866 add $ACC3,$ACC3,$T1.2d // h2 -> h3
868 ushr $T0.2d,$ACC0,#26
869 and $ACC0,$ACC0,$MASK.2d
870 ushr $T1.2d,$ACC3,#26
871 and $ACC3,$ACC3,$MASK.2d
872 add $ACC1,$ACC1,$T0.2d // h0 -> h1
873 add $ACC4,$ACC4,$T1.2d // h3 -> h4
875 ////////////////////////////////////////////////////////////////
876 // write the result, can be partially reduced
878 st4 {$ACC0,$ACC1,$ACC2,$ACC3}[0],[$ctx],#16
879 st1 {$ACC4}[0],[$ctx]
884 .size poly1305_blocks_neon,.-poly1305_blocks_neon
886 .type poly1305_emit_neon,%function
889 ldr $is_base2_26,[$ctx,#24]
890 cbz $is_base2_26,poly1305_emit
892 ldp w10,w11,[$ctx] // load hash value base 2^26
893 ldp w12,w13,[$ctx,#8]
896 add $h0,x10,x11,lsl#26 // base 2^26 -> base 2^64
898 adds $h0,$h0,x12,lsl#52
899 add $h1,$h1,x13,lsl#14
902 adds $h1,$h1,x14,lsl#40
903 adc $h2,$h2,xzr // can be partially reduced...
905 ldp $t0,$t1,[$nonce] // load nonce
907 and $d0,$h2,#-4 // ... so reduce
908 add $d0,$d0,$h2,lsr#2
914 adds $d0,$h0,#5 // compare to modulus
918 tst $d2,#-4 // see if it's carried/borrowed
924 ror $t0,$t0,#32 // flip nonce words
927 adds $h0,$h0,$t0 // accumulate nonce
930 rev $h0,$h0 // flip output bytes
933 stp $h0,$h1,[$mac] // write result
936 .size poly1305_emit_neon,.-poly1305_emit_neon
941 .long 0,0,0,0,0,0,0,0
945 .long OPENSSL_armcap_P-.
947 .quad OPENSSL_armcap_P-.
956 last if (!s/^#/\/\// and !/^$/);
961 foreach (split("\n",$code)) {
962 s/\b(shrn\s+v[0-9]+)\.[24]d/$1.2s/ or
963 s/\b(fmov\s+)v([0-9]+)[^,]*,\s*x([0-9]+)/$1d$2,x$3/ or
964 (m/\bdup\b/ and (s/\.[24]s/.2d/g or 1)) or
965 (m/\b(eor|and)/ and (s/\.[248][sdh]/.16b/g or 1)) or
966 (m/\bum(ul|la)l\b/ and (s/\.4s/.2s/g or 1)) or
967 (m/\bum(ul|la)l2\b/ and (s/\.2s/.4s/g or 1)) or
968 (m/\bst[1-4]\s+{[^}]+}\[/ and (s/\.[24]d/.s/g or 1));
970 s/\.[124]([sd])\[/.$1\[/;