2 # Copyright 2015-2016 The OpenSSL Project Authors. All Rights Reserved.
4 # Licensed under the OpenSSL license (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 # ECP_NISTZ256 module for ARMv4.
21 # Original ECP_NISTZ256 submission targeting x86_64 is detailed in
22 # http://eprint.iacr.org/2013/816. In the process of adaptation
23 # original .c module was made 32-bit savvy in order to make this
24 # implementation possible.
26 # with/without -DECP_NISTZ256_ASM
29 # Cortex-A15 +100-316%
30 # Snapdragon S4 +66-187%
32 # Ranges denote minimum and maximum improvement coefficients depending
33 # on benchmark. Lower coefficients are for ECDSA sign, server-side
34 # operation. Keep in mind that +200% means 3x improvement.
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";
55 #if defined(__thumb2__)
62 ########################################################################
63 # Convert ecp_nistz256_table.c to layout expected by ecp_nistz_gather_w7
65 $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
66 open TABLE,"<ecp_nistz256_table.c" or
67 open TABLE,"<${dir}../ecp_nistz256_table.c" or
68 die "failed to open ecp_nistz256_table.c:",$!;
73 s/TOBN\(\s*(0x[0-9a-f]+),\s*(0x[0-9a-f]+)\s*\)/push @arr,hex($2),hex($1)/geo;
77 # See ecp_nistz256_table.c for explanation for why it's 64*16*37.
78 # 64*16*37-1 is because $#arr returns last valid index or @arr, not
80 die "insane number of elements" if ($#arr != 64*16*37-1);
83 .globl ecp_nistz256_precomputed
84 .type ecp_nistz256_precomputed,%object
86 ecp_nistz256_precomputed:
88 ########################################################################
89 # this conversion smashes P256_POINT_AFFINE by individual bytes with
90 # 64 byte interval, similar to
94 @tbl = splice(@arr,0,64*16);
95 for($i=0;$i<64;$i++) {
97 for($j=0;$j<64;$j++) {
98 push @line,(@tbl[$j*16+$i/4]>>(($i%4)*8))&0xff;
101 $code.=join(',',map { sprintf "0x%02x",$_} @line);
106 .size ecp_nistz256_precomputed,.-ecp_nistz256_precomputed
108 .LRR: @ 2^512 mod P precomputed for NIST P256 polynomial
109 .long 0x00000003, 0x00000000, 0xffffffff, 0xfffffffb
110 .long 0xfffffffe, 0xffffffff, 0xfffffffd, 0x00000004
112 .long 1,0,0,0,0,0,0,0
113 .asciz "ECP_NISTZ256 for ARMv4, CRYPTOGAMS by <appro\@openssl.org>"
117 ########################################################################
118 # common register layout, note that $t2 is link register, so that if
119 # internal subroutine uses $t2, then it has to offload lr...
121 ($r_ptr,$a_ptr,$b_ptr,$ff,$a0,$a1,$a2,$a3,$a4,$a5,$a6,$a7,$t1,$t2)=
122 map("r$_",(0..12,14));
123 ($t0,$t3)=($ff,$a_ptr);
126 @ void ecp_nistz256_to_mont(BN_ULONG r0[8],const BN_ULONG r1[8]);
127 .globl ecp_nistz256_to_mont
128 .type ecp_nistz256_to_mont,%function
129 ecp_nistz256_to_mont:
131 b .Lecp_nistz256_mul_mont
132 .size ecp_nistz256_to_mont,.-ecp_nistz256_to_mont
134 @ void ecp_nistz256_from_mont(BN_ULONG r0[8],const BN_ULONG r1[8]);
135 .globl ecp_nistz256_from_mont
136 .type ecp_nistz256_from_mont,%function
137 ecp_nistz256_from_mont:
139 b .Lecp_nistz256_mul_mont
140 .size ecp_nistz256_from_mont,.-ecp_nistz256_from_mont
142 @ void ecp_nistz256_mul_by_2(BN_ULONG r0[8],const BN_ULONG r1[8]);
143 .globl ecp_nistz256_mul_by_2
144 .type ecp_nistz256_mul_by_2,%function
146 ecp_nistz256_mul_by_2:
147 stmdb sp!,{r4-r12,lr}
148 bl __ecp_nistz256_mul_by_2
149 #if __ARM_ARCH__>=5 || !defined(__thumb__)
150 ldmia sp!,{r4-r12,pc}
152 ldmia sp!,{r4-r12,lr}
153 bx lr @ interoperable with Thumb ISA:-)
155 .size ecp_nistz256_mul_by_2,.-ecp_nistz256_mul_by_2
157 .type __ecp_nistz256_mul_by_2,%function
159 __ecp_nistz256_mul_by_2:
163 adds $a0,$a0,$a0 @ a[0:7]+=a[0:7], i.e. add with itself
180 movcs $ff,#-1 @ $ff = carry ? -1 : 0
183 .size __ecp_nistz256_mul_by_2,.-__ecp_nistz256_mul_by_2
185 @ void ecp_nistz256_add(BN_ULONG r0[8],const BN_ULONG r1[8],
186 @ const BN_ULONG r2[8]);
187 .globl ecp_nistz256_add
188 .type ecp_nistz256_add,%function
191 stmdb sp!,{r4-r12,lr}
192 bl __ecp_nistz256_add
193 #if __ARM_ARCH__>=5 || !defined(__thumb__)
194 ldmia sp!,{r4-r12,pc}
196 ldmia sp!,{r4-r12,lr}
197 bx lr @ interoperable with Thumb ISA:-)
199 .size ecp_nistz256_add,.-ecp_nistz256_add
201 .type __ecp_nistz256_add,%function
204 str lr,[sp,#-4]! @ push lr
234 movcs $ff,#-1 @ $ff = carry ? -1 : 0, "broadcast" carry
235 ldr lr,[sp],#4 @ pop lr
239 @ if a+b carries, subtract modulus.
241 @ Note that because mod has special form, i.e. consists of
242 @ 0xffffffff, 1 and 0s, we can conditionally synthesize it by
243 @ using value of broadcasted carry as a whole or extracting
244 @ single bit. Follow $ff register...
246 subs $a0,$a0,$ff @ subtract synthesized modulus
257 sbcs $a6,$a6,$ff,lsr#31
264 .size __ecp_nistz256_add,.-__ecp_nistz256_add
266 @ void ecp_nistz256_mul_by_3(BN_ULONG r0[8],const BN_ULONG r1[8]);
267 .globl ecp_nistz256_mul_by_3
268 .type ecp_nistz256_mul_by_3,%function
270 ecp_nistz256_mul_by_3:
271 stmdb sp!,{r4-r12,lr}
272 bl __ecp_nistz256_mul_by_3
273 #if __ARM_ARCH__>=5 || !defined(__thumb__)
274 ldmia sp!,{r4-r12,pc}
276 ldmia sp!,{r4-r12,lr}
277 bx lr @ interoperable with Thumb ISA:-)
279 .size ecp_nistz256_mul_by_3,.-ecp_nistz256_mul_by_3
281 .type __ecp_nistz256_mul_by_3,%function
283 __ecp_nistz256_mul_by_3:
284 str lr,[sp,#-4]! @ push lr
286 @ As multiplication by 3 is performed as 2*n+n, below are inline
287 @ copies of __ecp_nistz256_mul_by_2 and __ecp_nistz256_add, see
288 @ corresponding subroutines for details.
293 adds $a0,$a0,$a0 @ a[0:7]+=a[0:7]
310 movcs $ff,#-1 @ $ff = carry ? -1 : 0, "broadcast" carry
312 subs $a0,$a0,$ff @ subtract synthesized modulus, see
313 @ .Lreduce_by_sub for details, except
314 @ that we don't write anything to
315 @ memory, but keep intermediate
316 @ results in registers...
321 ldr $b_ptr,[$a_ptr,#0]
324 sbcs $a6,$a6,$ff,lsr#31
329 adds $a0,$a0,$b_ptr @ 2*a[0:7]+=a[0:7]
330 ldr $b_ptr,[$a_ptr,#16]
345 movcs $ff,#-1 @ $ff = carry ? -1 : 0, "broadcast" carry
346 ldr lr,[sp],#4 @ pop lr
349 .size ecp_nistz256_mul_by_3,.-ecp_nistz256_mul_by_3
351 @ void ecp_nistz256_div_by_2(BN_ULONG r0[8],const BN_ULONG r1[8]);
352 .globl ecp_nistz256_div_by_2
353 .type ecp_nistz256_div_by_2,%function
355 ecp_nistz256_div_by_2:
356 stmdb sp!,{r4-r12,lr}
357 bl __ecp_nistz256_div_by_2
358 #if __ARM_ARCH__>=5 || !defined(__thumb__)
359 ldmia sp!,{r4-r12,pc}
361 ldmia sp!,{r4-r12,lr}
362 bx lr @ interoperable with Thumb ISA:-)
364 .size ecp_nistz256_div_by_2,.-ecp_nistz256_div_by_2
366 .type __ecp_nistz256_div_by_2,%function
368 __ecp_nistz256_div_by_2:
369 @ ret = (a is odd ? a+mod : a) >> 1
374 mov $ff,$a0,lsl#31 @ place least significant bit to most
375 @ significant position, now arithmetic
376 @ right shift by 31 will produce -1 or
377 @ 0, while logical right shift 1 or 0,
378 @ this is how modulus is conditionally
379 @ synthesized in this case...
381 adds $a0,$a0,$ff,asr#31
383 adcs $a1,$a1,$ff,asr#31
385 adcs $a2,$a2,$ff,asr#31
390 mov $a0,$a0,lsr#1 @ a[0:7]>>=1, we can start early
391 @ because it doesn't affect flags
393 orr $a0,$a0,$a1,lsl#31
394 adcs $a6,$a6,$ff,lsr#31
396 adcs $a7,$a7,$ff,asr#31
398 adc $b_ptr,$b_ptr,#0 @ top-most carry bit from addition
400 orr $a1,$a1,$a2,lsl#31
403 orr $a2,$a2,$a3,lsl#31
406 orr $a3,$a3,$a4,lsl#31
409 orr $a4,$a4,$a5,lsl#31
412 orr $a5,$a5,$a6,lsl#31
415 orr $a6,$a6,$a7,lsl#31
418 orr $a7,$a7,$b_ptr,lsl#31 @ don't forget the top-most carry bit
423 .size __ecp_nistz256_div_by_2,.-__ecp_nistz256_div_by_2
425 @ void ecp_nistz256_sub(BN_ULONG r0[8],const BN_ULONG r1[8],
426 @ const BN_ULONG r2[8]);
427 .globl ecp_nistz256_sub
428 .type ecp_nistz256_sub,%function
431 stmdb sp!,{r4-r12,lr}
432 bl __ecp_nistz256_sub
433 #if __ARM_ARCH__>=5 || !defined(__thumb__)
434 ldmia sp!,{r4-r12,pc}
436 ldmia sp!,{r4-r12,lr}
437 bx lr @ interoperable with Thumb ISA:-)
439 .size ecp_nistz256_sub,.-ecp_nistz256_sub
441 .type __ecp_nistz256_sub,%function
444 str lr,[sp,#-4]! @ push lr
470 sbc $ff,$ff,$ff @ broadcast borrow bit
471 ldr lr,[sp],#4 @ pop lr
475 @ if a-b borrows, add modulus.
477 @ Note that because mod has special form, i.e. consists of
478 @ 0xffffffff, 1 and 0s, we can conditionally synthesize it by
479 @ broadcasting borrow bit to a register, $ff, and using it as
480 @ a whole or extracting single bit.
482 adds $a0,$a0,$ff @ add synthesized modulus
493 adcs $a6,$a6,$ff,lsr#31
500 .size __ecp_nistz256_sub,.-__ecp_nistz256_sub
502 @ void ecp_nistz256_neg(BN_ULONG r0[8],const BN_ULONG r1[8]);
503 .globl ecp_nistz256_neg
504 .type ecp_nistz256_neg,%function
507 stmdb sp!,{r4-r12,lr}
508 bl __ecp_nistz256_neg
509 #if __ARM_ARCH__>=5 || !defined(__thumb__)
510 ldmia sp!,{r4-r12,pc}
512 ldmia sp!,{r4-r12,lr}
513 bx lr @ interoperable with Thumb ISA:-)
515 .size ecp_nistz256_neg,.-ecp_nistz256_neg
517 .type __ecp_nistz256_neg,%function
540 .size __ecp_nistz256_neg,.-__ecp_nistz256_neg
543 my @acc=map("r$_",(3..11));
544 my ($t0,$t1,$bj,$t2,$t3)=map("r$_",(0,1,2,12,14));
547 @ void ecp_nistz256_sqr_mont(BN_ULONG r0[8],const BN_ULONG r1[8]);
548 .globl ecp_nistz256_sqr_mont
549 .type ecp_nistz256_sqr_mont,%function
551 ecp_nistz256_sqr_mont:
553 b .Lecp_nistz256_mul_mont
554 .size ecp_nistz256_sqr_mont,.-ecp_nistz256_sqr_mont
556 @ void ecp_nistz256_mul_mont(BN_ULONG r0[8],const BN_ULONG r1[8],
557 @ const BN_ULONG r2[8]);
558 .globl ecp_nistz256_mul_mont
559 .type ecp_nistz256_mul_mont,%function
561 ecp_nistz256_mul_mont:
562 .Lecp_nistz256_mul_mont:
563 stmdb sp!,{r4-r12,lr}
564 bl __ecp_nistz256_mul_mont
565 #if __ARM_ARCH__>=5 || !defined(__thumb__)
566 ldmia sp!,{r4-r12,pc}
568 ldmia sp!,{r4-r12,lr}
569 bx lr @ interoperable with Thumb ISA:-)
571 .size ecp_nistz256_mul_mont,.-ecp_nistz256_mul_mont
573 .type __ecp_nistz256_mul_mont,%function
575 __ecp_nistz256_mul_mont:
576 stmdb sp!,{r0-r2,lr} @ make a copy of arguments too
578 ldr $bj,[$b_ptr,#0] @ b[0]
579 ldmia $a_ptr,{@acc[1]-@acc[8]}
581 umull @acc[0],$t3,@acc[1],$bj @ r[0]=a[0]*b[0]
582 stmdb sp!,{$acc[1]-@acc[8]} @ copy a[0-7] to stack, so
583 @ that it can be addressed
584 @ without spending register
586 umull @acc[1],$t0,@acc[2],$bj @ r[1]=a[1]*b[0]
587 umull @acc[2],$t1,@acc[3],$bj
588 adds @acc[1],@acc[1],$t3 @ accumulate high part of mult
589 umull @acc[3],$t2,@acc[4],$bj
590 adcs @acc[2],@acc[2],$t0
591 umull @acc[4],$t3,@acc[5],$bj
592 adcs @acc[3],@acc[3],$t1
593 umull @acc[5],$t0,@acc[6],$bj
594 adcs @acc[4],@acc[4],$t2
595 umull @acc[6],$t1,@acc[7],$bj
596 adcs @acc[5],@acc[5],$t3
597 umull @acc[7],$t2,@acc[8],$bj
598 adcs @acc[6],@acc[6],$t0
599 adcs @acc[7],@acc[7],$t1
600 eor $t3,$t3,$t3 @ first overflow bit is zero
603 for(my $i=1;$i<8;$i++) {
606 # Reduction iteration is normally performed by accumulating
607 # result of multiplication of modulus by "magic" digit [and
608 # omitting least significant word, which is guaranteed to
609 # be 0], but thanks to special form of modulus and "magic"
610 # digit being equal to least significant word, it can be
611 # performed with additions and subtractions alone. Indeed:
613 # ffff.0001.0000.0000.0000.ffff.ffff.ffff
615 # + xxxx.xxxx.xxxx.xxxx.xxxx.xxxx.xxxx.xxxx.abcd
617 # Now observing that ff..ff*x = (2^n-1)*x = 2^n*x-x, we
620 # xxxx.xxxx.xxxx.xxxx.xxxx.xxxx.xxxx.xxxx.abcd
621 # + abcd.0000.abcd.0000.0000.abcd.0000.0000.0000
622 # - abcd.0000.0000.0000.0000.0000.0000.abcd
624 # or marking redundant operations:
626 # xxxx.xxxx.xxxx.xxxx.xxxx.xxxx.xxxx.xxxx.----
627 # + abcd.0000.abcd.0000.0000.abcd.----.----.----
628 # - abcd.----.----.----.----.----.----.----
631 @ multiplication-less reduction $i
632 adds @acc[3],@acc[3],@acc[0] @ r[3]+=r[0]
633 ldr $bj,[sp,#40] @ restore b_ptr
634 adcs @acc[4],@acc[4],#0 @ r[4]+=0
635 adcs @acc[5],@acc[5],#0 @ r[5]+=0
636 adcs @acc[6],@acc[6],@acc[0] @ r[6]+=r[0]
637 ldr $t1,[sp,#0] @ load a[0]
638 adcs @acc[7],@acc[7],#0 @ r[7]+=0
639 ldr $bj,[$bj,#4*$i] @ load b[i]
640 adcs @acc[8],@acc[8],@acc[0] @ r[8]+=r[0]
642 adc $t3,$t3,#0 @ overflow bit
643 subs @acc[7],@acc[7],@acc[0] @ r[7]-=r[0]
644 ldr $t2,[sp,#4] @ a[1]
645 sbcs @acc[8],@acc[8],#0 @ r[8]-=0
646 umlal @acc[1],$t0,$t1,$bj @ "r[0]"+=a[0]*b[i]
648 sbc @acc[0],$t3,#0 @ overflow bit, keep in mind
649 @ that netto result is
650 @ addition of a value which
651 @ makes underflow impossible
653 ldr $t3,[sp,#8] @ a[2]
654 umlal @acc[2],$t1,$t2,$bj @ "r[1]"+=a[1]*b[i]
655 str @acc[0],[sp,#36] @ temporarily offload overflow
657 ldr $t4,[sp,#12] @ a[3], $t4 is alias @acc[0]
658 umlal @acc[3],$t2,$t3,$bj @ "r[2]"+=a[2]*b[i]
660 adds @acc[2],@acc[2],$t0 @ accumulate high part of mult
661 ldr $t0,[sp,#16] @ a[4]
662 umlal @acc[4],$t3,$t4,$bj @ "r[3]"+=a[3]*b[i]
664 adcs @acc[3],@acc[3],$t1
665 ldr $t1,[sp,#20] @ a[5]
666 umlal @acc[5],$t4,$t0,$bj @ "r[4]"+=a[4]*b[i]
668 adcs @acc[4],@acc[4],$t2
669 ldr $t2,[sp,#24] @ a[6]
670 umlal @acc[6],$t0,$t1,$bj @ "r[5]"+=a[5]*b[i]
672 adcs @acc[5],@acc[5],$t3
673 ldr $t3,[sp,#28] @ a[7]
674 umlal @acc[7],$t1,$t2,$bj @ "r[6]"+=a[6]*b[i]
676 adcs @acc[6],@acc[6],$t4
677 ldr @acc[0],[sp,#36] @ restore overflow bit
678 umlal @acc[8],$t2,$t3,$bj @ "r[7]"+=a[7]*b[i]
680 adcs @acc[7],@acc[7],$t0
681 adcs @acc[8],@acc[8],$t1
682 adcs @acc[0],$acc[0],$t2
683 adc $t3,$t3,#0 @ new overflow bit
685 push(@acc,shift(@acc)); # rotate registers, so that
686 # "r[i]" becomes r[i]
689 @ last multiplication-less reduction
690 adds @acc[3],@acc[3],@acc[0]
691 ldr $r_ptr,[sp,#32] @ restore r_ptr
692 adcs @acc[4],@acc[4],#0
693 adcs @acc[5],@acc[5],#0
694 adcs @acc[6],@acc[6],@acc[0]
695 adcs @acc[7],@acc[7],#0
696 adcs @acc[8],@acc[8],@acc[0]
698 subs @acc[7],@acc[7],@acc[0]
699 sbcs @acc[8],@acc[8],#0
700 sbc @acc[0],$t3,#0 @ overflow bit
702 @ Final step is "if result > mod, subtract mod", but we do it
703 @ "other way around", namely subtract modulus from result
704 @ and if it borrowed, add modulus back.
706 adds @acc[1],@acc[1],#1 @ subs @acc[1],@acc[1],#-1
707 adcs @acc[2],@acc[2],#0 @ sbcs @acc[2],@acc[2],#-1
708 adcs @acc[3],@acc[3],#0 @ sbcs @acc[3],@acc[3],#-1
709 sbcs @acc[4],@acc[4],#0
710 sbcs @acc[5],@acc[5],#0
711 sbcs @acc[6],@acc[6],#0
712 sbcs @acc[7],@acc[7],#1
713 adcs @acc[8],@acc[8],#0 @ sbcs @acc[8],@acc[8],#-1
714 ldr lr,[sp,#44] @ restore lr
715 sbc @acc[0],@acc[0],#0 @ broadcast borrow bit
718 @ Note that because mod has special form, i.e. consists of
719 @ 0xffffffff, 1 and 0s, we can conditionally synthesize it by
720 @ broadcasting borrow bit to a register, @acc[0], and using it as
721 @ a whole or extracting single bit.
723 adds @acc[1],@acc[1],@acc[0] @ add modulus or zero
724 adcs @acc[2],@acc[2],@acc[0]
725 str @acc[1],[$r_ptr,#0]
726 adcs @acc[3],@acc[3],@acc[0]
727 str @acc[2],[$r_ptr,#4]
728 adcs @acc[4],@acc[4],#0
729 str @acc[3],[$r_ptr,#8]
730 adcs @acc[5],@acc[5],#0
731 str @acc[4],[$r_ptr,#12]
732 adcs @acc[6],@acc[6],#0
733 str @acc[5],[$r_ptr,#16]
734 adcs @acc[7],@acc[7],@acc[0],lsr#31
735 str @acc[6],[$r_ptr,#20]
736 adc @acc[8],@acc[8],@acc[0]
737 str @acc[7],[$r_ptr,#24]
738 str @acc[8],[$r_ptr,#28]
741 .size __ecp_nistz256_mul_mont,.-__ecp_nistz256_mul_mont
746 my ($out,$inp,$index,$mask)=map("r$_",(0..3));
748 @ void ecp_nistz256_scatter_w5(void *r0,const P256_POINT *r1,
750 .globl ecp_nistz256_scatter_w5
751 .type ecp_nistz256_scatter_w5,%function
753 ecp_nistz256_scatter_w5:
756 add $out,$out,$index,lsl#2
758 ldmia $inp!,{r4-r11} @ X
759 str r4,[$out,#64*0-4]
760 str r5,[$out,#64*1-4]
761 str r6,[$out,#64*2-4]
762 str r7,[$out,#64*3-4]
763 str r8,[$out,#64*4-4]
764 str r9,[$out,#64*5-4]
765 str r10,[$out,#64*6-4]
766 str r11,[$out,#64*7-4]
769 ldmia $inp!,{r4-r11} @ Y
770 str r4,[$out,#64*0-4]
771 str r5,[$out,#64*1-4]
772 str r6,[$out,#64*2-4]
773 str r7,[$out,#64*3-4]
774 str r8,[$out,#64*4-4]
775 str r9,[$out,#64*5-4]
776 str r10,[$out,#64*6-4]
777 str r11,[$out,#64*7-4]
780 ldmia $inp,{r4-r11} @ Z
781 str r4,[$out,#64*0-4]
782 str r5,[$out,#64*1-4]
783 str r6,[$out,#64*2-4]
784 str r7,[$out,#64*3-4]
785 str r8,[$out,#64*4-4]
786 str r9,[$out,#64*5-4]
787 str r10,[$out,#64*6-4]
788 str r11,[$out,#64*7-4]
791 #if __ARM_ARCH__>=5 || defined(__thumb__)
796 .size ecp_nistz256_scatter_w5,.-ecp_nistz256_scatter_w5
798 @ void ecp_nistz256_gather_w5(P256_POINT *r0,const void *r1,
800 .globl ecp_nistz256_gather_w5
801 .type ecp_nistz256_gather_w5,%function
803 ecp_nistz256_gather_w5:
811 subne $index,$index,#1
813 add $inp,$inp,$index,lsl#2
832 stmia $out!,{r4-r11} @ X
851 stmia $out!,{r4-r11} @ Y
869 stmia $out,{r4-r11} @ Z
872 #if __ARM_ARCH__>=5 || defined(__thumb__)
877 .size ecp_nistz256_gather_w5,.-ecp_nistz256_gather_w5
879 @ void ecp_nistz256_scatter_w7(void *r0,const P256_POINT_AFFINE *r1,
881 .globl ecp_nistz256_scatter_w7
882 .type ecp_nistz256_scatter_w7,%function
884 ecp_nistz256_scatter_w7:
889 subs $index,$index,#1
890 strb $mask,[$out,#64*0-1]
891 mov $mask,$mask,lsr#8
892 strb $mask,[$out,#64*1-1]
893 mov $mask,$mask,lsr#8
894 strb $mask,[$out,#64*2-1]
895 mov $mask,$mask,lsr#8
896 strb $mask,[$out,#64*3-1]
900 #if __ARM_ARCH__>=5 || defined(__thumb__)
905 .size ecp_nistz256_scatter_w7,.-ecp_nistz256_scatter_w7
907 @ void ecp_nistz256_gather_w7(P256_POINT_AFFINE *r0,const void *r1,
909 .globl ecp_nistz256_gather_w7
910 .type ecp_nistz256_gather_w7,%function
912 ecp_nistz256_gather_w7:
920 subne $index,$index,#1
927 subs $index,$index,#1
940 #if __ARM_ARCH__>=5 || defined(__thumb__)
945 .size ecp_nistz256_gather_w7,.-ecp_nistz256_gather_w7
949 # In comparison to integer-only equivalent of below subroutine:
955 # As not all time is spent in multiplication, overall impact is deemed
956 # too low to care about.
958 my ($A0,$A1,$A2,$A3,$Bi,$zero,$temp)=map("d$_",(0..7));
961 my @AxB=map("q$_",(8..15));
963 my ($rptr,$aptr,$bptr,$toutptr)=map("r$_",(0..3));
969 .globl ecp_nistz256_mul_mont_neon
970 .type ecp_nistz256_mul_mont_neon,%function
972 ecp_nistz256_mul_mont_neon:
975 vstmdb sp!,{q4-q5} @ ABI specification says so
978 vld1.32 {${Bi}[0]},[$bptr,:32]!
979 veor $zero,$zero,$zero
980 vld1.32 {$A0-$A3}, [$aptr] @ can't specify :32 :-(
982 mov sp,$toutptr @ alloca
983 vmov.i64 $mask,#0xffff
985 vmull.u32 @AxB[0],$Bi,${A0}[0]
986 vmull.u32 @AxB[1],$Bi,${A0}[1]
987 vmull.u32 @AxB[2],$Bi,${A1}[0]
988 vmull.u32 @AxB[3],$Bi,${A1}[1]
989 vshr.u64 $temp,@AxB[0]#lo,#16
990 vmull.u32 @AxB[4],$Bi,${A2}[0]
991 vadd.u64 @AxB[0]#hi,@AxB[0]#hi,$temp
992 vmull.u32 @AxB[5],$Bi,${A2}[1]
993 vshr.u64 $temp,@AxB[0]#hi,#16 @ upper 32 bits of a[0]*b[0]
994 vmull.u32 @AxB[6],$Bi,${A3}[0]
995 vand.u64 @AxB[0],@AxB[0],$mask @ lower 32 bits of a[0]*b[0]
996 vmull.u32 @AxB[7],$Bi,${A3}[1]
998 for($i=1;$i<8;$i++) {
1000 vld1.32 {${Bi}[0]},[$bptr,:32]!
1001 veor $zero,$zero,$zero
1002 vadd.u64 @AxB[1]#lo,@AxB[1]#lo,$temp @ reduction
1003 vshl.u64 $mult,@AxB[0],#32
1004 vadd.u64 @AxB[3],@AxB[3],@AxB[0]
1005 vsub.u64 $mult,$mult,@AxB[0]
1007 vadd.u64 @AxB[6],@AxB[6],@AxB[0]
1008 vadd.u64 @AxB[7],@AxB[7],$mult
1010 push(@AxB,shift(@AxB));
1012 vmlal.u32 @AxB[0],$Bi,${A0}[0]
1013 vmlal.u32 @AxB[1],$Bi,${A0}[1]
1014 vmlal.u32 @AxB[2],$Bi,${A1}[0]
1015 vmlal.u32 @AxB[3],$Bi,${A1}[1]
1016 vshr.u64 $temp,@AxB[0]#lo,#16
1017 vmlal.u32 @AxB[4],$Bi,${A2}[0]
1018 vadd.u64 @AxB[0]#hi,@AxB[0]#hi,$temp
1019 vmlal.u32 @AxB[5],$Bi,${A2}[1]
1020 vshr.u64 $temp,@AxB[0]#hi,#16 @ upper 33 bits of a[0]*b[i]+t[0]
1021 vmlal.u32 @AxB[6],$Bi,${A3}[0]
1022 vand.u64 @AxB[0],@AxB[0],$mask @ lower 32 bits of a[0]*b[0]
1023 vmull.u32 @AxB[7],$Bi,${A3}[1]
1027 vadd.u64 @AxB[1]#lo,@AxB[1]#lo,$temp @ last reduction
1028 vshl.u64 $mult,@AxB[0],#32
1029 vadd.u64 @AxB[3],@AxB[3],@AxB[0]
1030 vsub.u64 $mult,$mult,@AxB[0]
1031 vadd.u64 @AxB[6],@AxB[6],@AxB[0]
1032 vadd.u64 @AxB[7],@AxB[7],$mult
1034 vshr.u64 $temp,@AxB[1]#lo,#16 @ convert
1035 vadd.u64 @AxB[1]#hi,@AxB[1]#hi,$temp
1036 vshr.u64 $temp,@AxB[1]#hi,#16
1037 vzip.16 @AxB[1]#lo,@AxB[1]#hi
1041 vadd.u64 @AxB[$_]#lo,@AxB[$_]#lo,$temp
1042 vst1.32 {@AxB[$_-1]#lo[0]},[$toutptr,:32]!
1043 vshr.u64 $temp,@AxB[$_]#lo,#16
1044 vadd.u64 @AxB[$_]#hi,@AxB[$_]#hi,$temp
1045 vshr.u64 $temp,@AxB[$_]#hi,#16
1046 vzip.16 @AxB[$_]#lo,@AxB[$_]#hi
1050 vst1.32 {@AxB[7]#lo[0]},[$toutptr,:32]!
1051 vst1.32 {$temp},[$toutptr] @ upper 33 bits
1067 ldr r9,[sp,#32] @ top-most bit
1085 adcs r7,r7,r9,lsr#31
1093 .size ecp_nistz256_mul_mont_neon,.-ecp_nistz256_mul_mont_neon
1099 ########################################################################
1100 # Below $aN assignment matches order in which 256-bit result appears in
1101 # register bank at return from __ecp_nistz256_mul_mont, so that we can
1102 # skip over reloading it from memory. This means that below functions
1103 # use custom calling sequence accepting 256-bit input in registers,
1104 # output pointer in r0, $r_ptr, and optional pointer in r2, $b_ptr.
1106 # See their "normal" counterparts for insights on calculations.
1108 my ($a0,$a1,$a2,$a3,$a4,$a5,$a6,$a7,
1109 $t0,$t1,$t2,$t3)=map("r$_",(11,3..10,12,14,1));
1113 .type __ecp_nistz256_sub_from,%function
1115 __ecp_nistz256_sub_from:
1116 str lr,[sp,#-4]! @ push lr
1121 ldr $t3,[$b_ptr,#12]
1123 ldr $t0,[$b_ptr,#16]
1125 ldr $t1,[$b_ptr,#20]
1127 ldr $t2,[$b_ptr,#24]
1129 ldr $t3,[$b_ptr,#28]
1134 sbc $ff,$ff,$ff @ broadcast borrow bit
1135 ldr lr,[sp],#4 @ pop lr
1137 adds $a0,$a0,$ff @ add synthesized modulus
1145 str $a3,[$r_ptr,#12]
1147 str $a4,[$r_ptr,#16]
1148 adcs $a6,$a6,$ff,lsr#31
1149 str $a5,[$r_ptr,#20]
1151 str $a6,[$r_ptr,#24]
1152 str $a7,[$r_ptr,#28]
1155 .size __ecp_nistz256_sub_from,.-__ecp_nistz256_sub_from
1157 .type __ecp_nistz256_sub_morf,%function
1159 __ecp_nistz256_sub_morf:
1160 str lr,[sp,#-4]! @ push lr
1165 ldr $t3,[$b_ptr,#12]
1167 ldr $t0,[$b_ptr,#16]
1169 ldr $t1,[$b_ptr,#20]
1171 ldr $t2,[$b_ptr,#24]
1173 ldr $t3,[$b_ptr,#28]
1178 sbc $ff,$ff,$ff @ broadcast borrow bit
1179 ldr lr,[sp],#4 @ pop lr
1181 adds $a0,$a0,$ff @ add synthesized modulus
1189 str $a3,[$r_ptr,#12]
1191 str $a4,[$r_ptr,#16]
1192 adcs $a6,$a6,$ff,lsr#31
1193 str $a5,[$r_ptr,#20]
1195 str $a6,[$r_ptr,#24]
1196 str $a7,[$r_ptr,#28]
1199 .size __ecp_nistz256_sub_morf,.-__ecp_nistz256_sub_morf
1201 .type __ecp_nistz256_add_self,%function
1203 __ecp_nistz256_add_self:
1204 adds $a0,$a0,$a0 @ a[0:7]+=a[0:7]
1216 movcs $ff,#-1 @ $ff = carry ? -1 : 0
1218 subs $a0,$a0,$ff @ subtract synthesized modulus
1226 str $a3,[$r_ptr,#12]
1228 str $a4,[$r_ptr,#16]
1229 sbcs $a6,$a6,$ff,lsr#31
1230 str $a5,[$r_ptr,#20]
1232 str $a6,[$r_ptr,#24]
1233 str $a7,[$r_ptr,#28]
1236 .size __ecp_nistz256_add_self,.-__ecp_nistz256_add_self
1240 ########################################################################
1241 # following subroutines are "literal" implementation of those found in
1244 ########################################################################
1245 # void ecp_nistz256_point_double(P256_POINT *out,const P256_POINT *inp);
1248 my ($S,$M,$Zsqr,$in_x,$tmp0)=map(32*$_,(0..4));
1249 # above map() describes stack layout with 5 temporary
1250 # 256-bit vectors on top. Then note that we push
1251 # starting from r0, which means that we have copy of
1252 # input arguments just below these temporary vectors.
1255 .globl ecp_nistz256_point_double
1256 .type ecp_nistz256_point_double,%function
1258 ecp_nistz256_point_double:
1259 stmdb sp!,{r0-r12,lr} @ push from r0, unusual, but intentional
1262 .Lpoint_double_shortcut:
1264 ldmia $a_ptr!,{r4-r11} @ copy in_x
1268 bl __ecp_nistz256_mul_by_2 @ p256_mul_by_2(S, in_y);
1270 add $b_ptr,$a_ptr,#32
1271 add $a_ptr,$a_ptr,#32
1272 add $r_ptr,sp,#$Zsqr
1273 bl __ecp_nistz256_mul_mont @ p256_sqr_mont(Zsqr, in_z);
1278 bl __ecp_nistz256_mul_mont @ p256_sqr_mont(S, S);
1280 ldr $b_ptr,[sp,#32*5+4]
1281 add $a_ptr,$b_ptr,#32
1282 add $b_ptr,$b_ptr,#64
1283 add $r_ptr,sp,#$tmp0
1284 bl __ecp_nistz256_mul_mont @ p256_mul_mont(tmp0, in_z, in_y);
1286 ldr $r_ptr,[sp,#32*5]
1287 add $r_ptr,$r_ptr,#64
1288 bl __ecp_nistz256_add_self @ p256_mul_by_2(res_z, tmp0);
1290 add $a_ptr,sp,#$in_x
1291 add $b_ptr,sp,#$Zsqr
1293 bl __ecp_nistz256_add @ p256_add(M, in_x, Zsqr);
1295 add $a_ptr,sp,#$in_x
1296 add $b_ptr,sp,#$Zsqr
1297 add $r_ptr,sp,#$Zsqr
1298 bl __ecp_nistz256_sub @ p256_sub(Zsqr, in_x, Zsqr);
1302 add $r_ptr,sp,#$tmp0
1303 bl __ecp_nistz256_mul_mont @ p256_sqr_mont(tmp0, S);
1305 add $a_ptr,sp,#$Zsqr
1308 bl __ecp_nistz256_mul_mont @ p256_mul_mont(M, M, Zsqr);
1310 ldr $r_ptr,[sp,#32*5]
1311 add $a_ptr,sp,#$tmp0
1312 add $r_ptr,$r_ptr,#32
1313 bl __ecp_nistz256_div_by_2 @ p256_div_by_2(res_y, tmp0);
1317 bl __ecp_nistz256_mul_by_3 @ p256_mul_by_3(M, M);
1319 add $a_ptr,sp,#$in_x
1322 bl __ecp_nistz256_mul_mont @ p256_mul_mont(S, S, in_x);
1324 add $r_ptr,sp,#$tmp0
1325 bl __ecp_nistz256_add_self @ p256_mul_by_2(tmp0, S);
1327 ldr $r_ptr,[sp,#32*5]
1330 bl __ecp_nistz256_mul_mont @ p256_sqr_mont(res_x, M);
1332 add $b_ptr,sp,#$tmp0
1333 bl __ecp_nistz256_sub_from @ p256_sub(res_x, res_x, tmp0);
1337 bl __ecp_nistz256_sub_morf @ p256_sub(S, S, res_x);
1341 bl __ecp_nistz256_mul_mont @ p256_mul_mont(S, S, M);
1343 ldr $r_ptr,[sp,#32*5]
1344 add $b_ptr,$r_ptr,#32
1345 add $r_ptr,$r_ptr,#32
1346 bl __ecp_nistz256_sub_from @ p256_sub(res_y, S, res_y);
1348 add sp,sp,#32*5+16 @ +16 means "skip even over saved r0-r3"
1349 #if __ARM_ARCH__>=5 || !defined(__thumb__)
1350 ldmia sp!,{r4-r12,pc}
1352 ldmia sp!,{r4-r12,lr}
1353 bx lr @ interoperable with Thumb ISA:-)
1355 .size ecp_nistz256_point_double,.-ecp_nistz256_point_double
1359 ########################################################################
1360 # void ecp_nistz256_point_add(P256_POINT *out,const P256_POINT *in1,
1361 # const P256_POINT *in2);
1363 my ($res_x,$res_y,$res_z,
1364 $in1_x,$in1_y,$in1_z,
1365 $in2_x,$in2_y,$in2_z,
1366 $H,$Hsqr,$R,$Rsqr,$Hcub,
1367 $U1,$U2,$S1,$S2)=map(32*$_,(0..17));
1368 my ($Z1sqr, $Z2sqr) = ($Hsqr, $Rsqr);
1369 # above map() describes stack layout with 18 temporary
1370 # 256-bit vectors on top. Then note that we push
1371 # starting from r0, which means that we have copy of
1372 # input arguments just below these temporary vectors.
1373 # We use three of them for !in1infty, !in2intfy and
1374 # result of check for zero.
1377 .globl ecp_nistz256_point_add
1378 .type ecp_nistz256_point_add,%function
1380 ecp_nistz256_point_add:
1381 stmdb sp!,{r0-r12,lr} @ push from r0, unusual, but intentional
1384 ldmia $b_ptr!,{r4-r11} @ copy in2
1394 ldmia $b_ptr!,{r4-r11}
1404 ldmia $b_ptr,{r4-r11}
1411 str r12,[sp,#32*18+8] @ !in2infty
1413 ldmia $a_ptr!,{r4-r11} @ copy in1
1423 ldmia $a_ptr!,{r4-r11}
1433 ldmia $a_ptr,{r4-r11}
1440 str r12,[sp,#32*18+4] @ !in1infty
1442 add $a_ptr,sp,#$in2_z
1443 add $b_ptr,sp,#$in2_z
1444 add $r_ptr,sp,#$Z2sqr
1445 bl __ecp_nistz256_mul_mont @ p256_sqr_mont(Z2sqr, in2_z);
1447 add $a_ptr,sp,#$in1_z
1448 add $b_ptr,sp,#$in1_z
1449 add $r_ptr,sp,#$Z1sqr
1450 bl __ecp_nistz256_mul_mont @ p256_sqr_mont(Z1sqr, in1_z);
1452 add $a_ptr,sp,#$in2_z
1453 add $b_ptr,sp,#$Z2sqr
1455 bl __ecp_nistz256_mul_mont @ p256_mul_mont(S1, Z2sqr, in2_z);
1457 add $a_ptr,sp,#$in1_z
1458 add $b_ptr,sp,#$Z1sqr
1460 bl __ecp_nistz256_mul_mont @ p256_mul_mont(S2, Z1sqr, in1_z);
1462 add $a_ptr,sp,#$in1_y
1465 bl __ecp_nistz256_mul_mont @ p256_mul_mont(S1, S1, in1_y);
1467 add $a_ptr,sp,#$in2_y
1470 bl __ecp_nistz256_mul_mont @ p256_mul_mont(S2, S2, in2_y);
1474 bl __ecp_nistz256_sub_from @ p256_sub(R, S2, S1);
1476 orr $a0,$a0,$a1 @ see if result is zero
1482 add $a_ptr,sp,#$in1_x
1484 add $b_ptr,sp,#$Z2sqr
1485 str $a0,[sp,#32*18+12]
1488 bl __ecp_nistz256_mul_mont @ p256_mul_mont(U1, in1_x, Z2sqr);
1490 add $a_ptr,sp,#$in2_x
1491 add $b_ptr,sp,#$Z1sqr
1493 bl __ecp_nistz256_mul_mont @ p256_mul_mont(U2, in2_x, Z1sqr);
1497 bl __ecp_nistz256_sub_from @ p256_sub(H, U2, U1);
1499 orr $a0,$a0,$a1 @ see if result is zero
1507 bne .Ladd_proceed @ is_equal(U1,U2)?
1509 ldr $t0,[sp,#32*18+4]
1510 ldr $t1,[sp,#32*18+8]
1511 ldr $t2,[sp,#32*18+12]
1513 beq .Ladd_proceed @ (in1infty || in2infty)?
1515 beq .Ladd_double @ is_equal(S1,S2)?
1517 ldr $r_ptr,[sp,#32*18+16]
1526 stmia $r_ptr!,{r4-r11}
1527 stmia $r_ptr!,{r4-r11}
1528 stmia $r_ptr!,{r4-r11}
1533 ldr $a_ptr,[sp,#32*18+20]
1534 add sp,sp,#32*(18-5)+16 @ difference in frame sizes
1535 b .Lpoint_double_shortcut
1541 add $r_ptr,sp,#$Rsqr
1542 bl __ecp_nistz256_mul_mont @ p256_sqr_mont(Rsqr, R);
1545 add $b_ptr,sp,#$in1_z
1546 add $r_ptr,sp,#$res_z
1547 bl __ecp_nistz256_mul_mont @ p256_mul_mont(res_z, H, in1_z);
1551 add $r_ptr,sp,#$Hsqr
1552 bl __ecp_nistz256_mul_mont @ p256_sqr_mont(Hsqr, H);
1554 add $a_ptr,sp,#$in2_z
1555 add $b_ptr,sp,#$res_z
1556 add $r_ptr,sp,#$res_z
1557 bl __ecp_nistz256_mul_mont @ p256_mul_mont(res_z, res_z, in2_z);
1560 add $b_ptr,sp,#$Hsqr
1561 add $r_ptr,sp,#$Hcub
1562 bl __ecp_nistz256_mul_mont @ p256_mul_mont(Hcub, Hsqr, H);
1564 add $a_ptr,sp,#$Hsqr
1567 bl __ecp_nistz256_mul_mont @ p256_mul_mont(U2, U1, Hsqr);
1569 add $r_ptr,sp,#$Hsqr
1570 bl __ecp_nistz256_add_self @ p256_mul_by_2(Hsqr, U2);
1572 add $b_ptr,sp,#$Rsqr
1573 add $r_ptr,sp,#$res_x
1574 bl __ecp_nistz256_sub_morf @ p256_sub(res_x, Rsqr, Hsqr);
1576 add $b_ptr,sp,#$Hcub
1577 bl __ecp_nistz256_sub_from @ p256_sub(res_x, res_x, Hcub);
1580 add $r_ptr,sp,#$res_y
1581 bl __ecp_nistz256_sub_morf @ p256_sub(res_y, U2, res_x);
1583 add $a_ptr,sp,#$Hcub
1586 bl __ecp_nistz256_mul_mont @ p256_mul_mont(S2, S1, Hcub);
1589 add $b_ptr,sp,#$res_y
1590 add $r_ptr,sp,#$res_y
1591 bl __ecp_nistz256_mul_mont @ p256_mul_mont(res_y, res_y, R);
1594 bl __ecp_nistz256_sub_from @ p256_sub(res_y, res_y, S2);
1596 ldr r11,[sp,#32*18+4] @ !in1intfy
1597 ldr r12,[sp,#32*18+8] @ !in2intfy
1605 ldr $r_ptr,[sp,#32*18+16]
1607 for($i=0;$i<96;$i+=8) { # conditional moves
1609 ldmia r1!,{r4-r5} @ res_x
1610 ldmia r2!,{r6-r7} @ in2_x
1611 ldmia r3!,{r8-r9} @ in1_x
1622 stmia $r_ptr!,{r4-r5}
1627 add sp,sp,#32*18+16+16 @ +16 means "skip even over saved r0-r3"
1628 #if __ARM_ARCH__>=5 || defined(__thumb__)
1629 ldmia sp!,{r4-r12,pc}
1631 ldmia sp!,{r4-r12,lr}
1632 bx lr @ interoperable with Thumb ISA:-)
1634 .size ecp_nistz256_point_add,.-ecp_nistz256_point_add
1638 ########################################################################
1639 # void ecp_nistz256_point_add_affine(P256_POINT *out,const P256_POINT *in1,
1640 # const P256_POINT_AFFINE *in2);
1642 my ($res_x,$res_y,$res_z,
1643 $in1_x,$in1_y,$in1_z,
1645 $U2,$S2,$H,$R,$Hsqr,$Hcub,$Rsqr)=map(32*$_,(0..14));
1647 # above map() describes stack layout with 18 temporary
1648 # 256-bit vectors on top. Then note that we push
1649 # starting from r0, which means that we have copy of
1650 # input arguments just below these temporary vectors.
1651 # We use two of them for !in1infty, !in2intfy.
1653 my @ONE_mont=(1,0,0,-1,-1,-1,-2,0);
1656 .globl ecp_nistz256_point_add_affine
1657 .type ecp_nistz256_point_add_affine,%function
1659 ecp_nistz256_point_add_affine:
1660 stmdb sp!,{r0-r12,lr} @ push from r0, unusual, but intentional
1663 ldmia $a_ptr!,{r4-r11} @ copy in1
1673 ldmia $a_ptr!,{r4-r11}
1683 ldmia $a_ptr,{r4-r11}
1690 str r12,[sp,#32*15+4] @ !in1infty
1692 ldmia $b_ptr!,{r4-r11} @ copy in2
1702 ldmia $b_ptr!,{r4-r11}
1717 str r12,[sp,#32*15+8] @ !in2infty
1719 add $a_ptr,sp,#$in1_z
1720 add $b_ptr,sp,#$in1_z
1721 add $r_ptr,sp,#$Z1sqr
1722 bl __ecp_nistz256_mul_mont @ p256_sqr_mont(Z1sqr, in1_z);
1724 add $a_ptr,sp,#$Z1sqr
1725 add $b_ptr,sp,#$in2_x
1727 bl __ecp_nistz256_mul_mont @ p256_mul_mont(U2, Z1sqr, in2_x);
1729 add $b_ptr,sp,#$in1_x
1731 bl __ecp_nistz256_sub_from @ p256_sub(H, U2, in1_x);
1733 add $a_ptr,sp,#$Z1sqr
1734 add $b_ptr,sp,#$in1_z
1736 bl __ecp_nistz256_mul_mont @ p256_mul_mont(S2, Z1sqr, in1_z);
1739 add $b_ptr,sp,#$in1_z
1740 add $r_ptr,sp,#$res_z
1741 bl __ecp_nistz256_mul_mont @ p256_mul_mont(res_z, H, in1_z);
1743 add $a_ptr,sp,#$in2_y
1746 bl __ecp_nistz256_mul_mont @ p256_mul_mont(S2, S2, in2_y);
1748 add $b_ptr,sp,#$in1_y
1750 bl __ecp_nistz256_sub_from @ p256_sub(R, S2, in1_y);
1754 add $r_ptr,sp,#$Hsqr
1755 bl __ecp_nistz256_mul_mont @ p256_sqr_mont(Hsqr, H);
1759 add $r_ptr,sp,#$Rsqr
1760 bl __ecp_nistz256_mul_mont @ p256_sqr_mont(Rsqr, R);
1763 add $b_ptr,sp,#$Hsqr
1764 add $r_ptr,sp,#$Hcub
1765 bl __ecp_nistz256_mul_mont @ p256_mul_mont(Hcub, Hsqr, H);
1767 add $a_ptr,sp,#$Hsqr
1768 add $b_ptr,sp,#$in1_x
1770 bl __ecp_nistz256_mul_mont @ p256_mul_mont(U2, in1_x, Hsqr);
1772 add $r_ptr,sp,#$Hsqr
1773 bl __ecp_nistz256_add_self @ p256_mul_by_2(Hsqr, U2);
1775 add $b_ptr,sp,#$Rsqr
1776 add $r_ptr,sp,#$res_x
1777 bl __ecp_nistz256_sub_morf @ p256_sub(res_x, Rsqr, Hsqr);
1779 add $b_ptr,sp,#$Hcub
1780 bl __ecp_nistz256_sub_from @ p256_sub(res_x, res_x, Hcub);
1783 add $r_ptr,sp,#$res_y
1784 bl __ecp_nistz256_sub_morf @ p256_sub(res_y, U2, res_x);
1786 add $a_ptr,sp,#$Hcub
1787 add $b_ptr,sp,#$in1_y
1789 bl __ecp_nistz256_mul_mont @ p256_mul_mont(S2, in1_y, Hcub);
1792 add $b_ptr,sp,#$res_y
1793 add $r_ptr,sp,#$res_y
1794 bl __ecp_nistz256_mul_mont @ p256_mul_mont(res_y, res_y, R);
1797 bl __ecp_nistz256_sub_from @ p256_sub(res_y, res_y, S2);
1799 ldr r11,[sp,#32*15+4] @ !in1intfy
1800 ldr r12,[sp,#32*15+8] @ !in2intfy
1808 ldr $r_ptr,[sp,#32*15]
1810 for($i=0;$i<64;$i+=8) { # conditional moves
1812 ldmia r1!,{r4-r5} @ res_x
1813 ldmia r2!,{r6-r7} @ in2_x
1814 ldmia r3!,{r8-r9} @ in1_x
1825 stmia $r_ptr!,{r4-r5}
1831 ldmia r1!,{r4-r5} @ res_z
1832 ldmia r3!,{r8-r9} @ in1_z
1835 and r6,r11,#@ONE_mont[$j]
1836 and r7,r11,#@ONE_mont[$j+1]
1843 stmia $r_ptr!,{r4-r5}
1847 add sp,sp,#32*15+16 @ +16 means "skip even over saved r0-r3"
1848 #if __ARM_ARCH__>=5 || !defined(__thumb__)
1849 ldmia sp!,{r4-r12,pc}
1851 ldmia sp!,{r4-r12,lr}
1852 bx lr @ interoperable with Thumb ISA:-)
1854 .size ecp_nistz256_point_add_affine,.-ecp_nistz256_point_add_affine
1858 foreach (split("\n",$code)) {
1859 s/\`([^\`]*)\`/eval $1/geo;
1861 s/\bq([0-9]+)#(lo|hi)/sprintf "d%d",2*$1+($2 eq "hi")/geo;
1865 close STDOUT; # enforce flush