3 # ====================================================================
4 # Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
5 # project. The module is, however, dual licensed under OpenSSL and
6 # CRYPTOGAMS licenses depending on where you obtain it. For further
7 # details see http://www.openssl.org/~appro/cryptogams/.
8 # ====================================================================
10 # ECP_NISTZ256 module for ARMv4.
14 # Original ECP_NISTZ256 submission targeting x86_64 is detailed in
15 # http://eprint.iacr.org/2013/816. In the process of adaptation
16 # original .c module was made 32-bit savvy in order to make this
17 # implementation possible.
19 # with/without -DECP_NISTZ256_ASM
22 # Cortex-A15 +100-316%
23 # Snapdragon S4 +66-187%
25 # Ranges denote minimum and maximum improvement coefficients depending
26 # on benchmark. Lower coefficients are for ECDSA sign, server-side
27 # operation. Keep in mind that +200% means 3x improvement.
30 if ($flavour=~/^\w[\w\-]*\.\w+$/) { $output=$flavour; undef $flavour; }
31 else { while (($output=shift) && ($output!~/^\w[\w\-]*\.\w+$/)) {} }
33 if ($flavour && $flavour ne "void") {
34 $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
35 ( $xlate="${dir}arm-xlate.pl" and -f $xlate ) or
36 ( $xlate="${dir}../../perlasm/arm-xlate.pl" and -f $xlate) or
37 die "can't locate arm-xlate.pl";
39 open STDOUT,"| \"$^X\" $xlate $flavour $output";
41 open STDOUT,">$output";
48 #if defined(__thumb2__)
55 ########################################################################
56 # Convert ecp_nistz256_table.c to layout expected by ecp_nistz_gather_w7
58 $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
59 open TABLE,"<ecp_nistz256_table.c" or
60 open TABLE,"<${dir}../ecp_nistz256_table.c" or
61 die "failed to open ecp_nistz256_table.c:",$!;
66 s/TOBN\(\s*(0x[0-9a-f]+),\s*(0x[0-9a-f]+)\s*\)/push @arr,hex($2),hex($1)/geo;
70 # See ecp_nistz256_table.c for explanation for why it's 64*16*37.
71 # 64*16*37-1 is because $#arr returns last valid index or @arr, not
73 die "insane number of elements" if ($#arr != 64*16*37-1);
76 .globl ecp_nistz256_precomputed
77 .type ecp_nistz256_precomputed,%object
79 ecp_nistz256_precomputed:
81 ########################################################################
82 # this conversion smashes P256_POINT_AFFINE by individual bytes with
83 # 64 byte interval, similar to
87 @tbl = splice(@arr,0,64*16);
88 for($i=0;$i<64;$i++) {
90 for($j=0;$j<64;$j++) {
91 push @line,(@tbl[$j*16+$i/4]>>(($i%4)*8))&0xff;
94 $code.=join(',',map { sprintf "0x%02x",$_} @line);
99 .size ecp_nistz256_precomputed,.-ecp_nistz256_precomputed
101 .LRR: @ 2^512 mod P precomputed for NIST P256 polynomial
102 .long 0x00000003, 0x00000000, 0xffffffff, 0xfffffffb
103 .long 0xfffffffe, 0xffffffff, 0xfffffffd, 0x00000004
105 .long 1,0,0,0,0,0,0,0
106 .asciz "ECP_NISTZ256 for ARMv4, CRYPTOGAMS by <appro\@openssl.org>"
110 ########################################################################
111 # common register layout, note that $t2 is link register, so that if
112 # internal subroutine uses $t2, then it has to offload lr...
114 ($r_ptr,$a_ptr,$b_ptr,$ff,$a0,$a1,$a2,$a3,$a4,$a5,$a6,$a7,$t1,$t2)=
115 map("r$_",(0..12,14));
116 ($t0,$t3)=($ff,$a_ptr);
119 @ void ecp_nistz256_to_mont(BN_ULONG r0[8],const BN_ULONG r1[8]);
120 .globl ecp_nistz256_to_mont
121 .type ecp_nistz256_to_mont,%function
122 ecp_nistz256_to_mont:
124 b .Lecp_nistz256_mul_mont
125 .size ecp_nistz256_to_mont,.-ecp_nistz256_to_mont
127 @ void ecp_nistz256_from_mont(BN_ULONG r0[8],const BN_ULONG r1[8]);
128 .globl ecp_nistz256_from_mont
129 .type ecp_nistz256_from_mont,%function
130 ecp_nistz256_from_mont:
132 b .Lecp_nistz256_mul_mont
133 .size ecp_nistz256_from_mont,.-ecp_nistz256_from_mont
135 @ void ecp_nistz256_mul_by_2(BN_ULONG r0[8],const BN_ULONG r1[8]);
136 .globl ecp_nistz256_mul_by_2
137 .type ecp_nistz256_mul_by_2,%function
139 ecp_nistz256_mul_by_2:
140 stmdb sp!,{r4-r12,lr}
141 bl __ecp_nistz256_mul_by_2
142 #if __ARM_ARCH__>=5 || !defined(__thumb__)
143 ldmia sp!,{r4-r12,pc}
145 ldmia sp!,{r4-r12,lr}
146 bx lr @ interoperable with Thumb ISA:-)
148 .size ecp_nistz256_mul_by_2,.-ecp_nistz256_mul_by_2
150 .type __ecp_nistz256_mul_by_2,%function
152 __ecp_nistz256_mul_by_2:
156 adds $a0,$a0,$a0 @ a[0:7]+=a[0:7], i.e. add with itself
173 movcs $ff,#-1 @ $ff = carry ? -1 : 0
176 .size __ecp_nistz256_mul_by_2,.-__ecp_nistz256_mul_by_2
178 @ void ecp_nistz256_add(BN_ULONG r0[8],const BN_ULONG r1[8],
179 @ const BN_ULONG r2[8]);
180 .globl ecp_nistz256_add
181 .type ecp_nistz256_add,%function
184 stmdb sp!,{r4-r12,lr}
185 bl __ecp_nistz256_add
186 #if __ARM_ARCH__>=5 || !defined(__thumb__)
187 ldmia sp!,{r4-r12,pc}
189 ldmia sp!,{r4-r12,lr}
190 bx lr @ interoperable with Thumb ISA:-)
192 .size ecp_nistz256_add,.-ecp_nistz256_add
194 .type __ecp_nistz256_add,%function
197 str lr,[sp,#-4]! @ push lr
227 movcs $ff,#-1 @ $ff = carry ? -1 : 0, "broadcast" carry
228 ldr lr,[sp],#4 @ pop lr
232 @ if a+b carries, subtract modulus.
234 @ Note that because mod has special form, i.e. consists of
235 @ 0xffffffff, 1 and 0s, we can conditionally synthesize it by
236 @ using value of broadcasted carry as a whole or extracting
237 @ single bit. Follow $ff register...
239 subs $a0,$a0,$ff @ subtract synthesized modulus
250 sbcs $a6,$a6,$ff,lsr#31
257 .size __ecp_nistz256_add,.-__ecp_nistz256_add
259 @ void ecp_nistz256_mul_by_3(BN_ULONG r0[8],const BN_ULONG r1[8]);
260 .globl ecp_nistz256_mul_by_3
261 .type ecp_nistz256_mul_by_3,%function
263 ecp_nistz256_mul_by_3:
264 stmdb sp!,{r4-r12,lr}
265 bl __ecp_nistz256_mul_by_3
266 #if __ARM_ARCH__>=5 || !defined(__thumb__)
267 ldmia sp!,{r4-r12,pc}
269 ldmia sp!,{r4-r12,lr}
270 bx lr @ interoperable with Thumb ISA:-)
272 .size ecp_nistz256_mul_by_3,.-ecp_nistz256_mul_by_3
274 .type __ecp_nistz256_mul_by_3,%function
276 __ecp_nistz256_mul_by_3:
277 str lr,[sp,#-4]! @ push lr
279 @ As multiplication by 3 is performed as 2*n+n, below are inline
280 @ copies of __ecp_nistz256_mul_by_2 and __ecp_nistz256_add, see
281 @ corresponding subroutines for details.
286 adds $a0,$a0,$a0 @ a[0:7]+=a[0:7]
303 movcs $ff,#-1 @ $ff = carry ? -1 : 0, "broadcast" carry
305 subs $a0,$a0,$ff @ subtract synthesized modulus, see
306 @ .Lreduce_by_sub for details, except
307 @ that we don't write anything to
308 @ memory, but keep intermediate
309 @ results in registers...
314 ldr $b_ptr,[$a_ptr,#0]
317 sbcs $a6,$a6,$ff,lsr#31
322 adds $a0,$a0,$b_ptr @ 2*a[0:7]+=a[0:7]
323 ldr $b_ptr,[$a_ptr,#16]
338 movcs $ff,#-1 @ $ff = carry ? -1 : 0, "broadcast" carry
339 ldr lr,[sp],#4 @ pop lr
342 .size ecp_nistz256_mul_by_3,.-ecp_nistz256_mul_by_3
344 @ void ecp_nistz256_div_by_2(BN_ULONG r0[8],const BN_ULONG r1[8]);
345 .globl ecp_nistz256_div_by_2
346 .type ecp_nistz256_div_by_2,%function
348 ecp_nistz256_div_by_2:
349 stmdb sp!,{r4-r12,lr}
350 bl __ecp_nistz256_div_by_2
351 #if __ARM_ARCH__>=5 || !defined(__thumb__)
352 ldmia sp!,{r4-r12,pc}
354 ldmia sp!,{r4-r12,lr}
355 bx lr @ interoperable with Thumb ISA:-)
357 .size ecp_nistz256_div_by_2,.-ecp_nistz256_div_by_2
359 .type __ecp_nistz256_div_by_2,%function
361 __ecp_nistz256_div_by_2:
362 @ ret = (a is odd ? a+mod : a) >> 1
367 mov $ff,$a0,lsl#31 @ place least significant bit to most
368 @ significant position, now arithmetic
369 @ right shift by 31 will produce -1 or
370 @ 0, while logical rigth shift 1 or 0,
371 @ this is how modulus is conditionally
372 @ synthesized in this case...
374 adds $a0,$a0,$ff,asr#31
376 adcs $a1,$a1,$ff,asr#31
378 adcs $a2,$a2,$ff,asr#31
383 mov $a0,$a0,lsr#1 @ a[0:7]>>=1, we can start early
384 @ because it doesn't affect flags
386 orr $a0,$a0,$a1,lsl#31
387 adcs $a6,$a6,$ff,lsr#31
389 adcs $a7,$a7,$ff,asr#31
391 adc $b_ptr,$b_ptr,#0 @ top-most carry bit from addition
393 orr $a1,$a1,$a2,lsl#31
396 orr $a2,$a2,$a3,lsl#31
399 orr $a3,$a3,$a4,lsl#31
402 orr $a4,$a4,$a5,lsl#31
405 orr $a5,$a5,$a6,lsl#31
408 orr $a6,$a6,$a7,lsl#31
411 orr $a7,$a7,$b_ptr,lsl#31 @ don't forget the top-most carry bit
416 .size __ecp_nistz256_div_by_2,.-__ecp_nistz256_div_by_2
418 @ void ecp_nistz256_sub(BN_ULONG r0[8],const BN_ULONG r1[8],
419 @ const BN_ULONG r2[8]);
420 .globl ecp_nistz256_sub
421 .type ecp_nistz256_sub,%function
424 stmdb sp!,{r4-r12,lr}
425 bl __ecp_nistz256_sub
426 #if __ARM_ARCH__>=5 || !defined(__thumb__)
427 ldmia sp!,{r4-r12,pc}
429 ldmia sp!,{r4-r12,lr}
430 bx lr @ interoperable with Thumb ISA:-)
432 .size ecp_nistz256_sub,.-ecp_nistz256_sub
434 .type __ecp_nistz256_sub,%function
437 str lr,[sp,#-4]! @ push lr
463 sbc $ff,$ff,$ff @ broadcast borrow bit
464 ldr lr,[sp],#4 @ pop lr
468 @ if a-b borrows, add modulus.
470 @ Note that because mod has special form, i.e. consists of
471 @ 0xffffffff, 1 and 0s, we can conditionally synthesize it by
472 @ broadcasting borrow bit to a register, $ff, and using it as
473 @ a whole or extracting single bit.
475 adds $a0,$a0,$ff @ add synthesized modulus
486 adcs $a6,$a6,$ff,lsr#31
493 .size __ecp_nistz256_sub,.-__ecp_nistz256_sub
495 @ void ecp_nistz256_neg(BN_ULONG r0[8],const BN_ULONG r1[8]);
496 .globl ecp_nistz256_neg
497 .type ecp_nistz256_neg,%function
500 stmdb sp!,{r4-r12,lr}
501 bl __ecp_nistz256_neg
502 #if __ARM_ARCH__>=5 || !defined(__thumb__)
503 ldmia sp!,{r4-r12,pc}
505 ldmia sp!,{r4-r12,lr}
506 bx lr @ interoperable with Thumb ISA:-)
508 .size ecp_nistz256_neg,.-ecp_nistz256_neg
510 .type __ecp_nistz256_neg,%function
533 .size __ecp_nistz256_neg,.-__ecp_nistz256_neg
536 my @acc=map("r$_",(3..11));
537 my ($t0,$t1,$bj,$t2,$t3)=map("r$_",(0,1,2,12,14));
540 @ void ecp_nistz256_sqr_mont(BN_ULONG r0[8],const BN_ULONG r1[8]);
541 .globl ecp_nistz256_sqr_mont
542 .type ecp_nistz256_sqr_mont,%function
544 ecp_nistz256_sqr_mont:
546 b .Lecp_nistz256_mul_mont
547 .size ecp_nistz256_sqr_mont,.-ecp_nistz256_sqr_mont
549 @ void ecp_nistz256_mul_mont(BN_ULONG r0[8],const BN_ULONG r1[8],
550 @ const BN_ULONG r2[8]);
551 .globl ecp_nistz256_mul_mont
552 .type ecp_nistz256_mul_mont,%function
554 ecp_nistz256_mul_mont:
555 .Lecp_nistz256_mul_mont:
556 stmdb sp!,{r4-r12,lr}
557 bl __ecp_nistz256_mul_mont
558 #if __ARM_ARCH__>=5 || !defined(__thumb__)
559 ldmia sp!,{r4-r12,pc}
561 ldmia sp!,{r4-r12,lr}
562 bx lr @ interoperable with Thumb ISA:-)
564 .size ecp_nistz256_mul_mont,.-ecp_nistz256_mul_mont
566 .type __ecp_nistz256_mul_mont,%function
568 __ecp_nistz256_mul_mont:
569 stmdb sp!,{r0-r2,lr} @ make a copy of arguments too
571 ldr $bj,[$b_ptr,#0] @ b[0]
572 ldmia $a_ptr,{@acc[1]-@acc[8]}
574 umull @acc[0],$t3,@acc[1],$bj @ r[0]=a[0]*b[0]
575 stmdb sp!,{$acc[1]-@acc[8]} @ copy a[0-7] to stack, so
576 @ that it can be addressed
577 @ without spending register
579 umull @acc[1],$t0,@acc[2],$bj @ r[1]=a[1]*b[0]
580 umull @acc[2],$t1,@acc[3],$bj
581 adds @acc[1],@acc[1],$t3 @ accumulate high part of mult
582 umull @acc[3],$t2,@acc[4],$bj
583 adcs @acc[2],@acc[2],$t0
584 umull @acc[4],$t3,@acc[5],$bj
585 adcs @acc[3],@acc[3],$t1
586 umull @acc[5],$t0,@acc[6],$bj
587 adcs @acc[4],@acc[4],$t2
588 umull @acc[6],$t1,@acc[7],$bj
589 adcs @acc[5],@acc[5],$t3
590 umull @acc[7],$t2,@acc[8],$bj
591 adcs @acc[6],@acc[6],$t0
592 adcs @acc[7],@acc[7],$t1
593 eor $t3,$t3,$t3 @ first overflow bit is zero
596 for(my $i=1;$i<8;$i++) {
599 # Reduction iteration is normally performed by accumulating
600 # result of multiplication of modulus by "magic" digit [and
601 # omitting least significant word, which is guaranteed to
602 # be 0], but thanks to special form of modulus and "magic"
603 # digit being equal to least significant word, it can be
604 # performed with additions and subtractions alone. Indeed:
606 # ffff.0001.0000.0000.0000.ffff.ffff.ffff
608 # + xxxx.xxxx.xxxx.xxxx.xxxx.xxxx.xxxx.xxxx.abcd
610 # Now observing that ff..ff*x = (2^n-1)*x = 2^n*x-x, we
613 # xxxx.xxxx.xxxx.xxxx.xxxx.xxxx.xxxx.xxxx.abcd
614 # + abcd.0000.abcd.0000.0000.abcd.0000.0000.0000
615 # - abcd.0000.0000.0000.0000.0000.0000.abcd
617 # or marking redundant operations:
619 # xxxx.xxxx.xxxx.xxxx.xxxx.xxxx.xxxx.xxxx.----
620 # + abcd.0000.abcd.0000.0000.abcd.----.----.----
621 # - abcd.----.----.----.----.----.----.----
624 @ multiplication-less reduction $i
625 adds @acc[3],@acc[3],@acc[0] @ r[3]+=r[0]
626 ldr $bj,[sp,#40] @ restore b_ptr
627 adcs @acc[4],@acc[4],#0 @ r[4]+=0
628 adcs @acc[5],@acc[5],#0 @ r[5]+=0
629 adcs @acc[6],@acc[6],@acc[0] @ r[6]+=r[0]
630 ldr $t1,[sp,#0] @ load a[0]
631 adcs @acc[7],@acc[7],#0 @ r[7]+=0
632 ldr $bj,[$bj,#4*$i] @ load b[i]
633 adcs @acc[8],@acc[8],@acc[0] @ r[8]+=r[0]
635 adc $t3,$t3,#0 @ overflow bit
636 subs @acc[7],@acc[7],@acc[0] @ r[7]-=r[0]
637 ldr $t2,[sp,#4] @ a[1]
638 sbcs @acc[8],@acc[8],#0 @ r[8]-=0
639 umlal @acc[1],$t0,$t1,$bj @ "r[0]"+=a[0]*b[i]
641 sbc @acc[0],$t3,#0 @ overflow bit, keep in mind
642 @ that netto result is
643 @ addition of a value which
644 @ makes underflow impossible
646 ldr $t3,[sp,#8] @ a[2]
647 umlal @acc[2],$t1,$t2,$bj @ "r[1]"+=a[1]*b[i]
648 str @acc[0],[sp,#36] @ temporarily offload overflow
650 ldr $t4,[sp,#12] @ a[3], $t4 is alias @acc[0]
651 umlal @acc[3],$t2,$t3,$bj @ "r[2]"+=a[2]*b[i]
653 adds @acc[2],@acc[2],$t0 @ accumulate high part of mult
654 ldr $t0,[sp,#16] @ a[4]
655 umlal @acc[4],$t3,$t4,$bj @ "r[3]"+=a[3]*b[i]
657 adcs @acc[3],@acc[3],$t1
658 ldr $t1,[sp,#20] @ a[5]
659 umlal @acc[5],$t4,$t0,$bj @ "r[4]"+=a[4]*b[i]
661 adcs @acc[4],@acc[4],$t2
662 ldr $t2,[sp,#24] @ a[6]
663 umlal @acc[6],$t0,$t1,$bj @ "r[5]"+=a[5]*b[i]
665 adcs @acc[5],@acc[5],$t3
666 ldr $t3,[sp,#28] @ a[7]
667 umlal @acc[7],$t1,$t2,$bj @ "r[6]"+=a[6]*b[i]
669 adcs @acc[6],@acc[6],$t4
670 ldr @acc[0],[sp,#36] @ restore overflow bit
671 umlal @acc[8],$t2,$t3,$bj @ "r[7]"+=a[7]*b[i]
673 adcs @acc[7],@acc[7],$t0
674 adcs @acc[8],@acc[8],$t1
675 adcs @acc[0],$acc[0],$t2
676 adc $t3,$t3,#0 @ new overflow bit
678 push(@acc,shift(@acc)); # rotate registers, so that
679 # "r[i]" becomes r[i]
682 @ last multiplication-less reduction
683 adds @acc[3],@acc[3],@acc[0]
684 ldr $r_ptr,[sp,#32] @ restore r_ptr
685 adcs @acc[4],@acc[4],#0
686 adcs @acc[5],@acc[5],#0
687 adcs @acc[6],@acc[6],@acc[0]
688 adcs @acc[7],@acc[7],#0
689 adcs @acc[8],@acc[8],@acc[0]
691 subs @acc[7],@acc[7],@acc[0]
692 sbcs @acc[8],@acc[8],#0
693 sbc @acc[0],$t3,#0 @ overflow bit
695 @ Final step is "if result > mod, subtract mod", but we do it
696 @ "other way around", namely subtract modulus from result
697 @ and if it borrowed, add modulus back.
699 adds @acc[1],@acc[1],#1 @ subs @acc[1],@acc[1],#-1
700 adcs @acc[2],@acc[2],#0 @ sbcs @acc[2],@acc[2],#-1
701 adcs @acc[3],@acc[3],#0 @ sbcs @acc[3],@acc[3],#-1
702 sbcs @acc[4],@acc[4],#0
703 sbcs @acc[5],@acc[5],#0
704 sbcs @acc[6],@acc[6],#0
705 sbcs @acc[7],@acc[7],#1
706 adcs @acc[8],@acc[8],#0 @ sbcs @acc[8],@acc[8],#-1
707 ldr lr,[sp,#44] @ restore lr
708 sbc @acc[0],@acc[0],#0 @ broadcast borrow bit
711 @ Note that because mod has special form, i.e. consists of
712 @ 0xffffffff, 1 and 0s, we can conditionally synthesize it by
713 @ broadcasting borrow bit to a register, @acc[0], and using it as
714 @ a whole or extracting single bit.
716 adds @acc[1],@acc[1],@acc[0] @ add modulus or zero
717 adcs @acc[2],@acc[2],@acc[0]
718 str @acc[1],[$r_ptr,#0]
719 adcs @acc[3],@acc[3],@acc[0]
720 str @acc[2],[$r_ptr,#4]
721 adcs @acc[4],@acc[4],#0
722 str @acc[3],[$r_ptr,#8]
723 adcs @acc[5],@acc[5],#0
724 str @acc[4],[$r_ptr,#12]
725 adcs @acc[6],@acc[6],#0
726 str @acc[5],[$r_ptr,#16]
727 adcs @acc[7],@acc[7],@acc[0],lsr#31
728 str @acc[6],[$r_ptr,#20]
729 adc @acc[8],@acc[8],@acc[0]
730 str @acc[7],[$r_ptr,#24]
731 str @acc[8],[$r_ptr,#28]
734 .size __ecp_nistz256_mul_mont,.-__ecp_nistz256_mul_mont
739 my ($out,$inp,$index,$mask)=map("r$_",(0..3));
741 @ void ecp_nistz256_scatter_w5(void *r0,const P256_POINT *r1,
743 .globl ecp_nistz256_scatter_w5
744 .type ecp_nistz256_scatter_w5,%function
746 ecp_nistz256_scatter_w5:
749 add $out,$out,$index,lsl#2
751 ldmia $inp!,{r4-r11} @ X
752 str r4,[$out,#64*0-4]
753 str r5,[$out,#64*1-4]
754 str r6,[$out,#64*2-4]
755 str r7,[$out,#64*3-4]
756 str r8,[$out,#64*4-4]
757 str r9,[$out,#64*5-4]
758 str r10,[$out,#64*6-4]
759 str r11,[$out,#64*7-4]
762 ldmia $inp!,{r4-r11} @ Y
763 str r4,[$out,#64*0-4]
764 str r5,[$out,#64*1-4]
765 str r6,[$out,#64*2-4]
766 str r7,[$out,#64*3-4]
767 str r8,[$out,#64*4-4]
768 str r9,[$out,#64*5-4]
769 str r10,[$out,#64*6-4]
770 str r11,[$out,#64*7-4]
773 ldmia $inp,{r4-r11} @ Z
774 str r4,[$out,#64*0-4]
775 str r5,[$out,#64*1-4]
776 str r6,[$out,#64*2-4]
777 str r7,[$out,#64*3-4]
778 str r8,[$out,#64*4-4]
779 str r9,[$out,#64*5-4]
780 str r10,[$out,#64*6-4]
781 str r11,[$out,#64*7-4]
784 #if __ARM_ARCH__>=5 || defined(__thumb__)
789 .size ecp_nistz256_scatter_w5,.-ecp_nistz256_scatter_w5
791 @ void ecp_nistz256_gather_w5(P256_POINT *r0,const void *r1,
793 .globl ecp_nistz256_gather_w5
794 .type ecp_nistz256_gather_w5,%function
796 ecp_nistz256_gather_w5:
804 subne $index,$index,#1
806 add $inp,$inp,$index,lsl#2
825 stmia $out!,{r4-r11} @ X
844 stmia $out!,{r4-r11} @ Y
862 stmia $out,{r4-r11} @ Z
865 #if __ARM_ARCH__>=5 || defined(__thumb__)
870 .size ecp_nistz256_gather_w5,.-ecp_nistz256_gather_w5
872 @ void ecp_nistz256_scatter_w7(void *r0,const P256_POINT_AFFINE *r1,
874 .globl ecp_nistz256_scatter_w7
875 .type ecp_nistz256_scatter_w7,%function
877 ecp_nistz256_scatter_w7:
882 subs $index,$index,#1
883 strb $mask,[$out,#64*0-1]
884 mov $mask,$mask,lsr#8
885 strb $mask,[$out,#64*1-1]
886 mov $mask,$mask,lsr#8
887 strb $mask,[$out,#64*2-1]
888 mov $mask,$mask,lsr#8
889 strb $mask,[$out,#64*3-1]
893 #if __ARM_ARCH__>=5 || defined(__thumb__)
898 .size ecp_nistz256_scatter_w7,.-ecp_nistz256_scatter_w7
900 @ void ecp_nistz256_gather_w7(P256_POINT_AFFINE *r0,const void *r1,
902 .globl ecp_nistz256_gather_w7
903 .type ecp_nistz256_gather_w7,%function
905 ecp_nistz256_gather_w7:
913 subne $index,$index,#1
920 subs $index,$index,#1
933 #if __ARM_ARCH__>=5 || defined(__thumb__)
938 .size ecp_nistz256_gather_w7,.-ecp_nistz256_gather_w7
942 # In comparison to integer-only equivalent of below subroutine:
948 # As not all time is spent in multiplication, overall impact is deemed
949 # too low to care about.
951 my ($A0,$A1,$A2,$A3,$Bi,$zero,$temp)=map("d$_",(0..7));
954 my @AxB=map("q$_",(8..15));
956 my ($rptr,$aptr,$bptr,$toutptr)=map("r$_",(0..3));
962 .globl ecp_nistz256_mul_mont_neon
963 .type ecp_nistz256_mul_mont_neon,%function
965 ecp_nistz256_mul_mont_neon:
968 vstmdb sp!,{q4-q5} @ ABI specification says so
971 vld1.32 {${Bi}[0]},[$bptr,:32]!
972 veor $zero,$zero,$zero
973 vld1.32 {$A0-$A3}, [$aptr] @ can't specify :32 :-(
975 mov sp,$toutptr @ alloca
976 vmov.i64 $mask,#0xffff
978 vmull.u32 @AxB[0],$Bi,${A0}[0]
979 vmull.u32 @AxB[1],$Bi,${A0}[1]
980 vmull.u32 @AxB[2],$Bi,${A1}[0]
981 vmull.u32 @AxB[3],$Bi,${A1}[1]
982 vshr.u64 $temp,@AxB[0]#lo,#16
983 vmull.u32 @AxB[4],$Bi,${A2}[0]
984 vadd.u64 @AxB[0]#hi,@AxB[0]#hi,$temp
985 vmull.u32 @AxB[5],$Bi,${A2}[1]
986 vshr.u64 $temp,@AxB[0]#hi,#16 @ upper 32 bits of a[0]*b[0]
987 vmull.u32 @AxB[6],$Bi,${A3}[0]
988 vand.u64 @AxB[0],@AxB[0],$mask @ lower 32 bits of a[0]*b[0]
989 vmull.u32 @AxB[7],$Bi,${A3}[1]
991 for($i=1;$i<8;$i++) {
993 vld1.32 {${Bi}[0]},[$bptr,:32]!
994 veor $zero,$zero,$zero
995 vadd.u64 @AxB[1]#lo,@AxB[1]#lo,$temp @ reduction
996 vshl.u64 $mult,@AxB[0],#32
997 vadd.u64 @AxB[3],@AxB[3],@AxB[0]
998 vsub.u64 $mult,$mult,@AxB[0]
1000 vadd.u64 @AxB[6],@AxB[6],@AxB[0]
1001 vadd.u64 @AxB[7],@AxB[7],$mult
1003 push(@AxB,shift(@AxB));
1005 vmlal.u32 @AxB[0],$Bi,${A0}[0]
1006 vmlal.u32 @AxB[1],$Bi,${A0}[1]
1007 vmlal.u32 @AxB[2],$Bi,${A1}[0]
1008 vmlal.u32 @AxB[3],$Bi,${A1}[1]
1009 vshr.u64 $temp,@AxB[0]#lo,#16
1010 vmlal.u32 @AxB[4],$Bi,${A2}[0]
1011 vadd.u64 @AxB[0]#hi,@AxB[0]#hi,$temp
1012 vmlal.u32 @AxB[5],$Bi,${A2}[1]
1013 vshr.u64 $temp,@AxB[0]#hi,#16 @ upper 33 bits of a[0]*b[i]+t[0]
1014 vmlal.u32 @AxB[6],$Bi,${A3}[0]
1015 vand.u64 @AxB[0],@AxB[0],$mask @ lower 32 bits of a[0]*b[0]
1016 vmull.u32 @AxB[7],$Bi,${A3}[1]
1020 vadd.u64 @AxB[1]#lo,@AxB[1]#lo,$temp @ last reduction
1021 vshl.u64 $mult,@AxB[0],#32
1022 vadd.u64 @AxB[3],@AxB[3],@AxB[0]
1023 vsub.u64 $mult,$mult,@AxB[0]
1024 vadd.u64 @AxB[6],@AxB[6],@AxB[0]
1025 vadd.u64 @AxB[7],@AxB[7],$mult
1027 vshr.u64 $temp,@AxB[1]#lo,#16 @ convert
1028 vadd.u64 @AxB[1]#hi,@AxB[1]#hi,$temp
1029 vshr.u64 $temp,@AxB[1]#hi,#16
1030 vzip.16 @AxB[1]#lo,@AxB[1]#hi
1034 vadd.u64 @AxB[$_]#lo,@AxB[$_]#lo,$temp
1035 vst1.32 {@AxB[$_-1]#lo[0]},[$toutptr,:32]!
1036 vshr.u64 $temp,@AxB[$_]#lo,#16
1037 vadd.u64 @AxB[$_]#hi,@AxB[$_]#hi,$temp
1038 vshr.u64 $temp,@AxB[$_]#hi,#16
1039 vzip.16 @AxB[$_]#lo,@AxB[$_]#hi
1043 vst1.32 {@AxB[7]#lo[0]},[$toutptr,:32]!
1044 vst1.32 {$temp},[$toutptr] @ upper 33 bits
1060 ldr r9,[sp,#32] @ top-most bit
1078 adcs r7,r7,r9,lsr#31
1086 .size ecp_nistz256_mul_mont_neon,.-ecp_nistz256_mul_mont_neon
1092 ########################################################################
1093 # Below $aN assignment matches order in which 256-bit result appears in
1094 # register bank at return from __ecp_nistz256_mul_mont, so that we can
1095 # skip over reloading it from memory. This means that below functions
1096 # use custom calling sequence accepting 256-bit input in registers,
1097 # output pointer in r0, $r_ptr, and optional pointer in r2, $b_ptr.
1099 # See their "normal" counterparts for insights on calculations.
1101 my ($a0,$a1,$a2,$a3,$a4,$a5,$a6,$a7,
1102 $t0,$t1,$t2,$t3)=map("r$_",(11,3..10,12,14,1));
1106 .type __ecp_nistz256_sub_from,%function
1108 __ecp_nistz256_sub_from:
1109 str lr,[sp,#-4]! @ push lr
1114 ldr $t3,[$b_ptr,#12]
1116 ldr $t0,[$b_ptr,#16]
1118 ldr $t1,[$b_ptr,#20]
1120 ldr $t2,[$b_ptr,#24]
1122 ldr $t3,[$b_ptr,#28]
1127 sbc $ff,$ff,$ff @ broadcast borrow bit
1128 ldr lr,[sp],#4 @ pop lr
1130 adds $a0,$a0,$ff @ add synthesized modulus
1138 str $a3,[$r_ptr,#12]
1140 str $a4,[$r_ptr,#16]
1141 adcs $a6,$a6,$ff,lsr#31
1142 str $a5,[$r_ptr,#20]
1144 str $a6,[$r_ptr,#24]
1145 str $a7,[$r_ptr,#28]
1148 .size __ecp_nistz256_sub_from,.-__ecp_nistz256_sub_from
1150 .type __ecp_nistz256_sub_morf,%function
1152 __ecp_nistz256_sub_morf:
1153 str lr,[sp,#-4]! @ push lr
1158 ldr $t3,[$b_ptr,#12]
1160 ldr $t0,[$b_ptr,#16]
1162 ldr $t1,[$b_ptr,#20]
1164 ldr $t2,[$b_ptr,#24]
1166 ldr $t3,[$b_ptr,#28]
1171 sbc $ff,$ff,$ff @ broadcast borrow bit
1172 ldr lr,[sp],#4 @ pop lr
1174 adds $a0,$a0,$ff @ add synthesized modulus
1182 str $a3,[$r_ptr,#12]
1184 str $a4,[$r_ptr,#16]
1185 adcs $a6,$a6,$ff,lsr#31
1186 str $a5,[$r_ptr,#20]
1188 str $a6,[$r_ptr,#24]
1189 str $a7,[$r_ptr,#28]
1192 .size __ecp_nistz256_sub_morf,.-__ecp_nistz256_sub_morf
1194 .type __ecp_nistz256_add_self,%function
1196 __ecp_nistz256_add_self:
1197 adds $a0,$a0,$a0 @ a[0:7]+=a[0:7]
1209 movcs $ff,#-1 @ $ff = carry ? -1 : 0
1211 subs $a0,$a0,$ff @ subtract synthesized modulus
1219 str $a3,[$r_ptr,#12]
1221 str $a4,[$r_ptr,#16]
1222 sbcs $a6,$a6,$ff,lsr#31
1223 str $a5,[$r_ptr,#20]
1225 str $a6,[$r_ptr,#24]
1226 str $a7,[$r_ptr,#28]
1229 .size __ecp_nistz256_add_self,.-__ecp_nistz256_add_self
1233 ########################################################################
1234 # following subroutines are "literal" implemetation of those found in
1237 ########################################################################
1238 # void ecp_nistz256_point_double(P256_POINT *out,const P256_POINT *inp);
1241 my ($S,$M,$Zsqr,$in_x,$tmp0)=map(32*$_,(0..4));
1242 # above map() describes stack layout with 5 temporary
1243 # 256-bit vectors on top. Then note that we push
1244 # starting from r0, which means that we have copy of
1245 # input arguments just below these temporary vectors.
1248 .globl ecp_nistz256_point_double
1249 .type ecp_nistz256_point_double,%function
1251 ecp_nistz256_point_double:
1252 stmdb sp!,{r0-r12,lr} @ push from r0, unusual, but intentional
1255 .Lpoint_double_shortcut:
1257 ldmia $a_ptr!,{r4-r11} @ copy in_x
1261 bl __ecp_nistz256_mul_by_2 @ p256_mul_by_2(S, in_y);
1263 add $b_ptr,$a_ptr,#32
1264 add $a_ptr,$a_ptr,#32
1265 add $r_ptr,sp,#$Zsqr
1266 bl __ecp_nistz256_mul_mont @ p256_sqr_mont(Zsqr, in_z);
1271 bl __ecp_nistz256_mul_mont @ p256_sqr_mont(S, S);
1273 ldr $b_ptr,[sp,#32*5+4]
1274 add $a_ptr,$b_ptr,#32
1275 add $b_ptr,$b_ptr,#64
1276 add $r_ptr,sp,#$tmp0
1277 bl __ecp_nistz256_mul_mont @ p256_mul_mont(tmp0, in_z, in_y);
1279 ldr $r_ptr,[sp,#32*5]
1280 add $r_ptr,$r_ptr,#64
1281 bl __ecp_nistz256_add_self @ p256_mul_by_2(res_z, tmp0);
1283 add $a_ptr,sp,#$in_x
1284 add $b_ptr,sp,#$Zsqr
1286 bl __ecp_nistz256_add @ p256_add(M, in_x, Zsqr);
1288 add $a_ptr,sp,#$in_x
1289 add $b_ptr,sp,#$Zsqr
1290 add $r_ptr,sp,#$Zsqr
1291 bl __ecp_nistz256_sub @ p256_sub(Zsqr, in_x, Zsqr);
1295 add $r_ptr,sp,#$tmp0
1296 bl __ecp_nistz256_mul_mont @ p256_sqr_mont(tmp0, S);
1298 add $a_ptr,sp,#$Zsqr
1301 bl __ecp_nistz256_mul_mont @ p256_mul_mont(M, M, Zsqr);
1303 ldr $r_ptr,[sp,#32*5]
1304 add $a_ptr,sp,#$tmp0
1305 add $r_ptr,$r_ptr,#32
1306 bl __ecp_nistz256_div_by_2 @ p256_div_by_2(res_y, tmp0);
1310 bl __ecp_nistz256_mul_by_3 @ p256_mul_by_3(M, M);
1312 add $a_ptr,sp,#$in_x
1315 bl __ecp_nistz256_mul_mont @ p256_mul_mont(S, S, in_x);
1317 add $r_ptr,sp,#$tmp0
1318 bl __ecp_nistz256_add_self @ p256_mul_by_2(tmp0, S);
1320 ldr $r_ptr,[sp,#32*5]
1323 bl __ecp_nistz256_mul_mont @ p256_sqr_mont(res_x, M);
1325 add $b_ptr,sp,#$tmp0
1326 bl __ecp_nistz256_sub_from @ p256_sub(res_x, res_x, tmp0);
1330 bl __ecp_nistz256_sub_morf @ p256_sub(S, S, res_x);
1334 bl __ecp_nistz256_mul_mont @ p256_mul_mont(S, S, M);
1336 ldr $r_ptr,[sp,#32*5]
1337 add $b_ptr,$r_ptr,#32
1338 add $r_ptr,$r_ptr,#32
1339 bl __ecp_nistz256_sub_from @ p256_sub(res_y, S, res_y);
1341 add sp,sp,#32*5+16 @ +16 means "skip even over saved r0-r3"
1342 #if __ARM_ARCH__>=5 || !defined(__thumb__)
1343 ldmia sp!,{r4-r12,pc}
1345 ldmia sp!,{r4-r12,lr}
1346 bx lr @ interoperable with Thumb ISA:-)
1348 .size ecp_nistz256_point_double,.-ecp_nistz256_point_double
1352 ########################################################################
1353 # void ecp_nistz256_point_add(P256_POINT *out,const P256_POINT *in1,
1354 # const P256_POINT *in2);
1356 my ($res_x,$res_y,$res_z,
1357 $in1_x,$in1_y,$in1_z,
1358 $in2_x,$in2_y,$in2_z,
1359 $H,$Hsqr,$R,$Rsqr,$Hcub,
1360 $U1,$U2,$S1,$S2)=map(32*$_,(0..17));
1361 my ($Z1sqr, $Z2sqr) = ($Hsqr, $Rsqr);
1362 # above map() describes stack layout with 18 temporary
1363 # 256-bit vectors on top. Then note that we push
1364 # starting from r0, which means that we have copy of
1365 # input arguments just below these temporary vectors.
1366 # We use three of them for !in1infty, !in2intfy and
1367 # result of check for zero.
1370 .globl ecp_nistz256_point_add
1371 .type ecp_nistz256_point_add,%function
1373 ecp_nistz256_point_add:
1374 stmdb sp!,{r0-r12,lr} @ push from r0, unusual, but intentional
1377 ldmia $b_ptr!,{r4-r11} @ copy in2
1387 ldmia $b_ptr!,{r4-r11}
1397 ldmia $b_ptr,{r4-r11}
1404 str r12,[sp,#32*18+8] @ !in2infty
1406 ldmia $a_ptr!,{r4-r11} @ copy in1
1416 ldmia $a_ptr!,{r4-r11}
1426 ldmia $a_ptr,{r4-r11}
1433 str r12,[sp,#32*18+4] @ !in1infty
1435 add $a_ptr,sp,#$in2_z
1436 add $b_ptr,sp,#$in2_z
1437 add $r_ptr,sp,#$Z2sqr
1438 bl __ecp_nistz256_mul_mont @ p256_sqr_mont(Z2sqr, in2_z);
1440 add $a_ptr,sp,#$in1_z
1441 add $b_ptr,sp,#$in1_z
1442 add $r_ptr,sp,#$Z1sqr
1443 bl __ecp_nistz256_mul_mont @ p256_sqr_mont(Z1sqr, in1_z);
1445 add $a_ptr,sp,#$in2_z
1446 add $b_ptr,sp,#$Z2sqr
1448 bl __ecp_nistz256_mul_mont @ p256_mul_mont(S1, Z2sqr, in2_z);
1450 add $a_ptr,sp,#$in1_z
1451 add $b_ptr,sp,#$Z1sqr
1453 bl __ecp_nistz256_mul_mont @ p256_mul_mont(S2, Z1sqr, in1_z);
1455 add $a_ptr,sp,#$in1_y
1458 bl __ecp_nistz256_mul_mont @ p256_mul_mont(S1, S1, in1_y);
1460 add $a_ptr,sp,#$in2_y
1463 bl __ecp_nistz256_mul_mont @ p256_mul_mont(S2, S2, in2_y);
1467 bl __ecp_nistz256_sub_from @ p256_sub(R, S2, S1);
1469 orr $a0,$a0,$a1 @ see if result is zero
1475 add $a_ptr,sp,#$in1_x
1477 add $b_ptr,sp,#$Z2sqr
1478 str $a0,[sp,#32*18+12]
1481 bl __ecp_nistz256_mul_mont @ p256_mul_mont(U1, in1_x, Z2sqr);
1483 add $a_ptr,sp,#$in2_x
1484 add $b_ptr,sp,#$Z1sqr
1486 bl __ecp_nistz256_mul_mont @ p256_mul_mont(U2, in2_x, Z1sqr);
1490 bl __ecp_nistz256_sub_from @ p256_sub(H, U2, U1);
1492 orr $a0,$a0,$a1 @ see if result is zero
1500 bne .Ladd_proceed @ is_equal(U1,U2)?
1502 ldr $t0,[sp,#32*18+4]
1503 ldr $t1,[sp,#32*18+8]
1504 ldr $t2,[sp,#32*18+12]
1506 beq .Ladd_proceed @ (in1infty || in2infty)?
1508 beq .Ladd_double @ is_equal(S1,S2)?
1510 ldr $r_ptr,[sp,#32*18+16]
1519 stmia $r_ptr!,{r4-r11}
1520 stmia $r_ptr!,{r4-r11}
1521 stmia $r_ptr!,{r4-r11}
1526 ldr $a_ptr,[sp,#32*18+20]
1527 add sp,sp,#32*(18-5)+16 @ difference in frame sizes
1528 b .Lpoint_double_shortcut
1534 add $r_ptr,sp,#$Rsqr
1535 bl __ecp_nistz256_mul_mont @ p256_sqr_mont(Rsqr, R);
1538 add $b_ptr,sp,#$in1_z
1539 add $r_ptr,sp,#$res_z
1540 bl __ecp_nistz256_mul_mont @ p256_mul_mont(res_z, H, in1_z);
1544 add $r_ptr,sp,#$Hsqr
1545 bl __ecp_nistz256_mul_mont @ p256_sqr_mont(Hsqr, H);
1547 add $a_ptr,sp,#$in2_z
1548 add $b_ptr,sp,#$res_z
1549 add $r_ptr,sp,#$res_z
1550 bl __ecp_nistz256_mul_mont @ p256_mul_mont(res_z, res_z, in2_z);
1553 add $b_ptr,sp,#$Hsqr
1554 add $r_ptr,sp,#$Hcub
1555 bl __ecp_nistz256_mul_mont @ p256_mul_mont(Hcub, Hsqr, H);
1557 add $a_ptr,sp,#$Hsqr
1560 bl __ecp_nistz256_mul_mont @ p256_mul_mont(U2, U1, Hsqr);
1562 add $r_ptr,sp,#$Hsqr
1563 bl __ecp_nistz256_add_self @ p256_mul_by_2(Hsqr, U2);
1565 add $b_ptr,sp,#$Rsqr
1566 add $r_ptr,sp,#$res_x
1567 bl __ecp_nistz256_sub_morf @ p256_sub(res_x, Rsqr, Hsqr);
1569 add $b_ptr,sp,#$Hcub
1570 bl __ecp_nistz256_sub_from @ p256_sub(res_x, res_x, Hcub);
1573 add $r_ptr,sp,#$res_y
1574 bl __ecp_nistz256_sub_morf @ p256_sub(res_y, U2, res_x);
1576 add $a_ptr,sp,#$Hcub
1579 bl __ecp_nistz256_mul_mont @ p256_mul_mont(S2, S1, Hcub);
1582 add $b_ptr,sp,#$res_y
1583 add $r_ptr,sp,#$res_y
1584 bl __ecp_nistz256_mul_mont @ p256_mul_mont(res_y, res_y, R);
1587 bl __ecp_nistz256_sub_from @ p256_sub(res_y, res_y, S2);
1589 ldr r11,[sp,#32*18+4] @ !in1intfy
1590 ldr r12,[sp,#32*18+8] @ !in2intfy
1598 ldr $r_ptr,[sp,#32*18+16]
1600 for($i=0;$i<96;$i+=8) { # conditional moves
1602 ldmia r1!,{r4-r5} @ res_x
1603 ldmia r2!,{r6-r7} @ in2_x
1604 ldmia r3!,{r8-r9} @ in1_x
1615 stmia $r_ptr!,{r4-r5}
1620 add sp,sp,#32*18+16+16 @ +16 means "skip even over saved r0-r3"
1621 #if __ARM_ARCH__>=5 || defined(__thumb__)
1622 ldmia sp!,{r4-r12,pc}
1624 ldmia sp!,{r4-r12,lr}
1625 bx lr @ interoperable with Thumb ISA:-)
1627 .size ecp_nistz256_point_add,.-ecp_nistz256_point_add
1631 ########################################################################
1632 # void ecp_nistz256_point_add_affine(P256_POINT *out,const P256_POINT *in1,
1633 # const P256_POINT_AFFINE *in2);
1635 my ($res_x,$res_y,$res_z,
1636 $in1_x,$in1_y,$in1_z,
1638 $U2,$S2,$H,$R,$Hsqr,$Hcub,$Rsqr)=map(32*$_,(0..14));
1640 # above map() describes stack layout with 18 temporary
1641 # 256-bit vectors on top. Then note that we push
1642 # starting from r0, which means that we have copy of
1643 # input arguments just below these temporary vectors.
1644 # We use two of them for !in1infty, !in2intfy.
1646 my @ONE_mont=(1,0,0,-1,-1,-1,-2,0);
1649 .globl ecp_nistz256_point_add_affine
1650 .type ecp_nistz256_point_add_affine,%function
1652 ecp_nistz256_point_add_affine:
1653 stmdb sp!,{r0-r12,lr} @ push from r0, unusual, but intentional
1656 ldmia $a_ptr!,{r4-r11} @ copy in1
1666 ldmia $a_ptr!,{r4-r11}
1676 ldmia $a_ptr,{r4-r11}
1683 str r12,[sp,#32*15+4] @ !in1infty
1685 ldmia $b_ptr!,{r4-r11} @ copy in2
1695 ldmia $b_ptr!,{r4-r11}
1710 str r12,[sp,#32*15+8] @ !in2infty
1712 add $a_ptr,sp,#$in1_z
1713 add $b_ptr,sp,#$in1_z
1714 add $r_ptr,sp,#$Z1sqr
1715 bl __ecp_nistz256_mul_mont @ p256_sqr_mont(Z1sqr, in1_z);
1717 add $a_ptr,sp,#$Z1sqr
1718 add $b_ptr,sp,#$in2_x
1720 bl __ecp_nistz256_mul_mont @ p256_mul_mont(U2, Z1sqr, in2_x);
1722 add $b_ptr,sp,#$in1_x
1724 bl __ecp_nistz256_sub_from @ p256_sub(H, U2, in1_x);
1726 add $a_ptr,sp,#$Z1sqr
1727 add $b_ptr,sp,#$in1_z
1729 bl __ecp_nistz256_mul_mont @ p256_mul_mont(S2, Z1sqr, in1_z);
1732 add $b_ptr,sp,#$in1_z
1733 add $r_ptr,sp,#$res_z
1734 bl __ecp_nistz256_mul_mont @ p256_mul_mont(res_z, H, in1_z);
1736 add $a_ptr,sp,#$in2_y
1739 bl __ecp_nistz256_mul_mont @ p256_mul_mont(S2, S2, in2_y);
1741 add $b_ptr,sp,#$in1_y
1743 bl __ecp_nistz256_sub_from @ p256_sub(R, S2, in1_y);
1747 add $r_ptr,sp,#$Hsqr
1748 bl __ecp_nistz256_mul_mont @ p256_sqr_mont(Hsqr, H);
1752 add $r_ptr,sp,#$Rsqr
1753 bl __ecp_nistz256_mul_mont @ p256_sqr_mont(Rsqr, R);
1756 add $b_ptr,sp,#$Hsqr
1757 add $r_ptr,sp,#$Hcub
1758 bl __ecp_nistz256_mul_mont @ p256_mul_mont(Hcub, Hsqr, H);
1760 add $a_ptr,sp,#$Hsqr
1761 add $b_ptr,sp,#$in1_x
1763 bl __ecp_nistz256_mul_mont @ p256_mul_mont(U2, in1_x, Hsqr);
1765 add $r_ptr,sp,#$Hsqr
1766 bl __ecp_nistz256_add_self @ p256_mul_by_2(Hsqr, U2);
1768 add $b_ptr,sp,#$Rsqr
1769 add $r_ptr,sp,#$res_x
1770 bl __ecp_nistz256_sub_morf @ p256_sub(res_x, Rsqr, Hsqr);
1772 add $b_ptr,sp,#$Hcub
1773 bl __ecp_nistz256_sub_from @ p256_sub(res_x, res_x, Hcub);
1776 add $r_ptr,sp,#$res_y
1777 bl __ecp_nistz256_sub_morf @ p256_sub(res_y, U2, res_x);
1779 add $a_ptr,sp,#$Hcub
1780 add $b_ptr,sp,#$in1_y
1782 bl __ecp_nistz256_mul_mont @ p256_mul_mont(S2, in1_y, Hcub);
1785 add $b_ptr,sp,#$res_y
1786 add $r_ptr,sp,#$res_y
1787 bl __ecp_nistz256_mul_mont @ p256_mul_mont(res_y, res_y, R);
1790 bl __ecp_nistz256_sub_from @ p256_sub(res_y, res_y, S2);
1792 ldr r11,[sp,#32*15+4] @ !in1intfy
1793 ldr r12,[sp,#32*15+8] @ !in2intfy
1801 ldr $r_ptr,[sp,#32*15]
1803 for($i=0;$i<64;$i+=8) { # conditional moves
1805 ldmia r1!,{r4-r5} @ res_x
1806 ldmia r2!,{r6-r7} @ in2_x
1807 ldmia r3!,{r8-r9} @ in1_x
1818 stmia $r_ptr!,{r4-r5}
1824 ldmia r1!,{r4-r5} @ res_z
1825 ldmia r3!,{r8-r9} @ in1_z
1828 and r6,r11,#@ONE_mont[$j]
1829 and r7,r11,#@ONE_mont[$j+1]
1836 stmia $r_ptr!,{r4-r5}
1840 add sp,sp,#32*15+16 @ +16 means "skip even over saved r0-r3"
1841 #if __ARM_ARCH__>=5 || !defined(__thumb__)
1842 ldmia sp!,{r4-r12,pc}
1844 ldmia sp!,{r4-r12,lr}
1845 bx lr @ interoperable with Thumb ISA:-)
1847 .size ecp_nistz256_point_add_affine,.-ecp_nistz256_point_add_affine
1851 foreach (split("\n",$code)) {
1852 s/\`([^\`]*)\`/eval $1/geo;
1854 s/\bq([0-9]+)#(lo|hi)/sprintf "d%d",2*$1+($2 eq "hi")/geo;
1858 close STDOUT; # enforce flush