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 while (($output=shift) && ($output!~/^\w[\w\-]*\.\w+$/)) {}
32 $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
33 ( $xlate="${dir}arm-xlate.pl" and -f $xlate ) or
34 ( $xlate="${dir}../../perlasm/arm-xlate.pl" and -f $xlate) or
35 die "can't locate arm-xlate.pl";
37 open OUT,"| \"$^X\" $xlate $flavour $output";
46 ########################################################################
47 # Convert ecp_nistz256_table.c to layout expected by ecp_nistz_gather_w7
49 $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
50 open TABLE,"<ecp_nistz256_table.c" or
51 open TABLE,"<${dir}../ecp_nistz256_table.c" or
52 die "failed to open ecp_nistz256_table.c:",$!;
57 s/TOBN\(\s*(0x[0-9a-f]+),\s*(0x[0-9a-f]+)\s*\)/push @arr,hex($2),hex($1)/geo;
61 # See ecp_nistz256_table.c for explanation for why it's 64*16*37.
62 # 64*16*37-1 is because $#arr returns last valid index or @arr, not
64 die "insane number of elements" if ($#arr != 64*16*37-1);
67 .globl ecp_nistz256_precomputed
68 .type ecp_nistz256_precomputed,%object
70 ecp_nistz256_precomputed:
72 ########################################################################
73 # this conversion smashes P256_POINT_AFFINE by individual bytes with
74 # 64 byte interval, similar to
78 @tbl = splice(@arr,0,64*16);
79 for($i=0;$i<64;$i++) {
81 for($j=0;$j<64;$j++) {
82 push @line,(@tbl[$j*16+$i/4]>>(($i%4)*8))&0xff;
85 $code.=join(',',map { sprintf "0x%02x",$_} @line);
90 .size ecp_nistz256_precomputed,.-ecp_nistz256_precomputed
92 .LRR: @ 2^512 mod P precomputed for NIST P256 polynomial
93 .long 0x00000003, 0x00000000, 0xffffffff, 0xfffffffb
94 .long 0xfffffffe, 0xffffffff, 0xfffffffd, 0x00000004
97 .asciz "ECP_NISTZ256 for ARMv4, CRYPTOGAMS by <appro\@openssl.org>"
101 ########################################################################
102 # common register layout, note that $t2 is link register, so that if
103 # internal subroutine uses $t2, then it has to offload lr...
105 ($r_ptr,$a_ptr,$b_ptr,$ff,$a0,$a1,$a2,$a3,$a4,$a5,$a6,$a7,$t1,$t2)=
106 map("r$_",(0..12,14));
107 ($t0,$t3)=($ff,$a_ptr);
110 @ void ecp_nistz256_to_mont(BN_ULONG r0[8],const BN_ULONG r1[8]);
111 .globl ecp_nistz256_to_mont
112 .type ecp_nistz256_to_mont,%function
113 ecp_nistz256_to_mont:
115 b .Lecp_nistz256_mul_mont
116 .size ecp_nistz256_to_mont,.-ecp_nistz256_to_mont
118 @ void ecp_nistz256_from_mont(BN_ULONG r0[8],const BN_ULONG r1[8]);
119 .globl ecp_nistz256_from_mont
120 .type ecp_nistz256_from_mont,%function
121 ecp_nistz256_from_mont:
123 b .Lecp_nistz256_mul_mont
124 .size ecp_nistz256_from_mont,.-ecp_nistz256_from_mont
126 @ void ecp_nistz256_mul_by_2(BN_ULONG r0[8],const BN_ULONG r1[8]);
127 .globl ecp_nistz256_mul_by_2
128 .type ecp_nistz256_mul_by_2,%function
130 ecp_nistz256_mul_by_2:
131 stmdb sp!,{r4-r12,lr}
132 bl _ecp_nistz256_mul_by_2
133 #if __ARM_ARCH__>=5 || !defined(__thumb__)
134 ldmia sp!,{r4-r12,pc}
136 ldmia sp!,{r4-r12,lr}
137 bx lr @ interoperable with Thumb ISA:-)
139 .size ecp_nistz256_mul_by_2,.-ecp_nistz256_mul_by_2
141 .type _ecp_nistz256_mul_by_2,%function
143 _ecp_nistz256_mul_by_2:
147 adds $a0,$a0,$a0 @ a[0:7]+=a[0:7], i.e. add with itself
161 movcs $ff,#-1 @ $ff = carry ? -1 : 0
164 .size _ecp_nistz256_mul_by_2,.-_ecp_nistz256_mul_by_2
166 @ void ecp_nistz256_add(BN_ULONG r0[8],const BN_ULONG r1[8],
167 @ const BN_ULONG r2[8]);
168 .globl ecp_nistz256_add
169 .type ecp_nistz256_add,%function
172 stmdb sp!,{r4-r12,lr}
174 #if __ARM_ARCH__>=5 || !defined(__thumb__)
175 ldmia sp!,{r4-r12,pc}
177 ldmia sp!,{r4-r12,lr}
178 bx lr @ interoperable with Thumb ISA:-)
180 .size ecp_nistz256_add,.-ecp_nistz256_add
182 .type _ecp_nistz256_add,%function
185 str lr,[sp,#-4]! @ push lr
212 movcs $ff,#-1 @ $ff = carry ? -1 : 0, "broadcast" carry
213 ldr lr,[sp],#4 @ pop lr
217 @ if a+b carries, subtract modulus.
219 @ Note that because mod has special form, i.e. consists of
220 @ 0xffffffff, 1 and 0s, we can conditionally synthesize it by
221 @ using value of broadcasted carry as a whole or extracting
222 @ single bit. Follow $ff register...
224 subs $a0,$a0,$ff @ subtract synthesized modulus
235 sbcs $a6,$a6,$ff,lsr#31
242 .size _ecp_nistz256_add,.-_ecp_nistz256_add
244 @ void ecp_nistz256_mul_by_3(BN_ULONG r0[8],const BN_ULONG r1[8]);
245 .globl ecp_nistz256_mul_by_3
246 .type ecp_nistz256_mul_by_3,%function
248 ecp_nistz256_mul_by_3:
249 stmdb sp!,{r4-r12,lr}
250 bl _ecp_nistz256_mul_by_3
251 #if __ARM_ARCH__>=5 || !defined(__thumb__)
252 ldmia sp!,{r4-r12,pc}
254 ldmia sp!,{r4-r12,lr}
255 bx lr @ interoperable with Thumb ISA:-)
257 .size ecp_nistz256_mul_by_3,.-ecp_nistz256_mul_by_3
259 .type _ecp_nistz256_mul_by_3,%function
261 _ecp_nistz256_mul_by_3:
262 str lr,[sp,#-4]! @ push lr
264 @ As multiplication by 3 is performed as 2*n+n, below are inline
265 @ copies of _ecp_nistz256_mul_by_2 and _ecp_nistz256_add, see
266 @ corresponding subroutines for details.
271 adds $a0,$a0,$a0 @ a[0:7]+=a[0:7]
285 movcs $ff,#-1 @ $ff = carry ? -1 : 0, "broadcast" carry
287 subs $a0,$a0,$ff @ subtract synthesized modulus, see
288 @ .Lreduce_by_sub for details, except
289 @ that we don't write anything to
290 @ memory, but keep intermediate
291 @ results in registers...
296 ldr $b_ptr,[$a_ptr,#0]
299 sbcs $a6,$a6,$ff,lsr#31
304 adds $a0,$a0,$b_ptr @ 2*a[0:7]+=a[0:7]
305 ldr $b_ptr,[$a_ptr,#16]
317 movcs $ff,#-1 @ $ff = carry ? -1 : 0, "broadcast" carry
318 ldr lr,[sp],#4 @ pop lr
321 .size ecp_nistz256_mul_by_3,.-ecp_nistz256_mul_by_3
323 @ void ecp_nistz256_div_by_2(BN_ULONG r0[8],const BN_ULONG r1[8]);
324 .globl ecp_nistz256_div_by_2
325 .type ecp_nistz256_div_by_2,%function
327 ecp_nistz256_div_by_2:
328 stmdb sp!,{r4-r12,lr}
329 bl _ecp_nistz256_div_by_2
330 #if __ARM_ARCH__>=5 || !defined(__thumb__)
331 ldmia sp!,{r4-r12,pc}
333 ldmia sp!,{r4-r12,lr}
334 bx lr @ interoperable with Thumb ISA:-)
336 .size ecp_nistz256_div_by_2,.-ecp_nistz256_div_by_2
338 .type _ecp_nistz256_div_by_2,%function
340 _ecp_nistz256_div_by_2:
341 @ ret = (a is odd ? a+mod : a) >> 1
346 mov $ff,$a0,lsl#31 @ place least significant bit to most
347 @ significant position, now arithmetic
348 @ right shift by 31 will produce -1 or
349 @ 0, while logical rigth shift 1 or 0,
350 @ this is how modulus is conditionally
351 @ synthesized in this case...
353 adds $a0,$a0,$ff,asr#31
355 adcs $a1,$a1,$ff,asr#31
357 adcs $a2,$a2,$ff,asr#31
362 mov $a0,$a0,lsr#1 @ a[0:7]>>=1, we can start early
363 @ because it doesn't affect flags
365 orr $a0,$a0,$a1,lsl#31
366 adcs $a6,$a6,$ff,lsr#31
368 adcs $a7,$a7,$ff,asr#31
370 adc $b_ptr,$b_ptr,#0 @ top-most carry bit from addition
372 orr $a1,$a1,$a2,lsl#31
375 orr $a2,$a2,$a3,lsl#31
378 orr $a3,$a3,$a4,lsl#31
381 orr $a4,$a4,$a5,lsl#31
384 orr $a5,$a5,$a6,lsl#31
387 orr $a6,$a6,$a7,lsl#31
390 orr $a7,$a7,$b_ptr,lsl#31 @ don't forget the top-most carry bit
395 .size _ecp_nistz256_div_by_2,.-_ecp_nistz256_div_by_2
397 @ void ecp_nistz256_sub(BN_ULONG r0[8],const BN_ULONG r1[8],
398 @ const BN_ULONG r2[8]);
399 .globl ecp_nistz256_sub
400 .type ecp_nistz256_sub,%function
403 stmdb sp!,{r4-r12,lr}
405 #if __ARM_ARCH__>=5 || !defined(__thumb__)
406 ldmia sp!,{r4-r12,pc}
408 ldmia sp!,{r4-r12,lr}
409 bx lr @ interoperable with Thumb ISA:-)
411 .size ecp_nistz256_sub,.-ecp_nistz256_sub
413 .type _ecp_nistz256_sub,%function
416 str lr,[sp,#-4]! @ push lr
442 sbc $ff,$ff,$ff @ broadcast borrow bit
443 ldr lr,[sp],#4 @ pop lr
447 @ if a-b borrows, add modulus.
449 @ Note that because mod has special form, i.e. consists of
450 @ 0xffffffff, 1 and 0s, we can conditionally synthesize it by
451 @ broadcasting borrow bit to a register, $ff, and using it as
452 @ a whole or extracting single bit.
454 adds $a0,$a0,$ff @ add synthesized modulus
465 adcs $a6,$a6,$ff,lsr#31
472 .size _ecp_nistz256_sub,.-_ecp_nistz256_sub
474 @ void ecp_nistz256_neg(BN_ULONG r0[8],const BN_ULONG r1[8]);
475 .globl ecp_nistz256_neg
476 .type ecp_nistz256_neg,%function
479 stmdb sp!,{r4-r12,lr}
481 #if __ARM_ARCH__>=5 || !defined(__thumb__)
482 ldmia sp!,{r4-r12,pc}
484 ldmia sp!,{r4-r12,lr}
485 bx lr @ interoperable with Thumb ISA:-)
487 .size ecp_nistz256_neg,.-ecp_nistz256_neg
489 .type _ecp_nistz256_neg,%function
512 .size _ecp_nistz256_neg,.-_ecp_nistz256_neg
515 my @acc=map("r$_",(3..11));
516 my ($t0,$t1,$bj,$t2,$t3)=map("r$_",(0,1,2,12,14));
519 @ void ecp_nistz256_sqr_mont(BN_ULONG r0[8],const BN_ULONG r1[8]);
520 .globl ecp_nistz256_sqr_mont
521 .type ecp_nistz256_sqr_mont,%function
523 ecp_nistz256_sqr_mont:
525 b .Lecp_nistz256_mul_mont
526 .size ecp_nistz256_sqr_mont,.-ecp_nistz256_sqr_mont
528 @ void ecp_nistz256_mul_mont(BN_ULONG r0[8],const BN_ULONG r1[8],
529 @ const BN_ULONG r2[8]);
530 .globl ecp_nistz256_mul_mont
531 .type ecp_nistz256_mul_mont,%function
533 ecp_nistz256_mul_mont:
534 .Lecp_nistz256_mul_mont:
535 stmdb sp!,{r4-r12,lr}
536 bl _ecp_nistz256_mul_mont
537 #if __ARM_ARCH__>=5 || !defined(__thumb__)
538 ldmia sp!,{r4-r12,pc}
540 ldmia sp!,{r4-r12,lr}
541 bx lr @ interoperable with Thumb ISA:-)
543 .size ecp_nistz256_mul_mont,.-ecp_nistz256_mul_mont
545 .type _ecp_nistz256_mul_mont,%function
547 _ecp_nistz256_mul_mont:
548 stmdb sp!,{r0-r2,lr} @ make a copy of arguments too
550 ldr $bj,[$b_ptr,#0] @ b[0]
551 ldmia $a_ptr,{@acc[1]-@acc[8]}
553 umull @acc[0],$t3,@acc[1],$bj @ r[0]=a[0]*b[0]
554 stmdb sp!,{$acc[1]-@acc[8]} @ copy a[0-7] to stack, so
555 @ that it can be addressed
556 @ without spending register
558 umull @acc[1],$t0,@acc[2],$bj @ r[1]=a[1]*b[0]
559 umull @acc[2],$t1,@acc[3],$bj
560 adds @acc[1],@acc[1],$t3 @ accumulate high part of mult
561 umull @acc[3],$t2,@acc[4],$bj
562 adcs @acc[2],@acc[2],$t0
563 umull @acc[4],$t3,@acc[5],$bj
564 adcs @acc[3],@acc[3],$t1
565 umull @acc[5],$t0,@acc[6],$bj
566 adcs @acc[4],@acc[4],$t2
567 umull @acc[6],$t1,@acc[7],$bj
568 adcs @acc[5],@acc[5],$t3
569 umull @acc[7],$t2,@acc[8],$bj
570 adcs @acc[6],@acc[6],$t0
571 adcs @acc[7],@acc[7],$t1
572 eor $t3,$t3,$t3 @ first overflow bit is zero
575 for(my $i=1;$i<8;$i++) {
578 # Reduction iteration is normally performed by accumulating
579 # result of multiplication of modulus by "magic" digit [and
580 # omitting least significant word, which is guaranteed to
581 # be 0], but thanks to special form of modulus and "magic"
582 # digit being equal to least significant word, it can be
583 # performed with additions and subtractions alone. Indeed:
585 # ffff.0001.0000.0000.0000.ffff.ffff.ffff
587 # + xxxx.xxxx.xxxx.xxxx.xxxx.xxxx.xxxx.xxxx.abcd
589 # Now observing that ff..ff*x = (2^n-1)*x = 2^n*x-x, we
592 # xxxx.xxxx.xxxx.xxxx.xxxx.xxxx.xxxx.xxxx.abcd
593 # + abcd.0000.abcd.0000.0000.abcd.0000.0000.0000
594 # - abcd.0000.0000.0000.0000.0000.0000.abcd
596 # or marking redundant operations:
598 # xxxx.xxxx.xxxx.xxxx.xxxx.xxxx.xxxx.xxxx.----
599 # + abcd.0000.abcd.0000.0000.abcd.----.----.----
600 # - abcd.----.----.----.----.----.----.----
603 @ multiplication-less reduction $i
604 adds @acc[3],@acc[3],@acc[0] @ r[3]+=r[0]
605 ldr $bj,[sp,#40] @ restore b_ptr
606 adcs @acc[4],@acc[4],#0 @ r[4]+=0
607 adcs @acc[5],@acc[5],#0 @ r[5]+=0
608 adcs @acc[6],@acc[6],@acc[0] @ r[6]+=r[0]
609 ldr $t1,[sp,#0] @ load a[0]
610 adcs @acc[7],@acc[7],#0 @ r[7]+=0
611 ldr $bj,[$bj,#4*$i] @ load b[i]
612 adcs @acc[8],@acc[8],@acc[0] @ r[8]+=r[0]
614 adc $t3,$t3,#0 @ overflow bit
615 subs @acc[7],@acc[7],@acc[0] @ r[7]-=r[0]
616 ldr $t2,[sp,#4] @ a[1]
617 sbcs @acc[8],@acc[8],#0 @ r[8]-=0
618 umlal @acc[1],$t0,$t1,$bj @ "r[0]"+=a[0]*b[i]
620 sbc @acc[0],$t3,#0 @ overflow bit, keep in mind
621 @ that netto result is
622 @ addition of a value which
623 @ makes underflow impossible
625 ldr $t3,[sp,#8] @ a[2]
626 umlal @acc[2],$t1,$t2,$bj @ "r[1]"+=a[1]*b[i]
627 str @acc[0],[sp,#36] @ temporarily offload overflow
629 ldr $t4,[sp,#12] @ a[3], $t4 is alias @acc[0]
630 umlal @acc[3],$t2,$t3,$bj @ "r[2]"+=a[2]*b[i]
632 adds @acc[2],@acc[2],$t0 @ accumulate high part of mult
633 ldr $t0,[sp,#16] @ a[4]
634 umlal @acc[4],$t3,$t4,$bj @ "r[3]"+=a[3]*b[i]
636 adcs @acc[3],@acc[3],$t1
637 ldr $t1,[sp,#20] @ a[5]
638 umlal @acc[5],$t4,$t0,$bj @ "r[4]"+=a[4]*b[i]
640 adcs @acc[4],@acc[4],$t2
641 ldr $t2,[sp,#24] @ a[6]
642 umlal @acc[6],$t0,$t1,$bj @ "r[5]"+=a[5]*b[i]
644 adcs @acc[5],@acc[5],$t3
645 ldr $t3,[sp,#28] @ a[7]
646 umlal @acc[7],$t1,$t2,$bj @ "r[6]"+=a[6]*b[i]
648 adcs @acc[6],@acc[6],$t4
649 ldr @acc[0],[sp,#36] @ restore overflow bit
650 umlal @acc[8],$t2,$t3,$bj @ "r[7]"+=a[7]*b[i]
652 adcs @acc[7],@acc[7],$t0
653 adcs @acc[8],@acc[8],$t1
654 adcs @acc[0],$acc[0],$t2
655 adc $t3,$t3,#0 @ new overflow bit
657 push(@acc,shift(@acc)); # rotate registers, so that
658 # "r[i]" becomes r[i]
661 @ last multiplication-less reduction
662 adds @acc[3],@acc[3],@acc[0]
663 ldr $r_ptr,[sp,#32] @ restore r_ptr
664 adcs @acc[4],@acc[4],#0
665 adcs @acc[5],@acc[5],#0
666 adcs @acc[6],@acc[6],@acc[0]
667 adcs @acc[7],@acc[7],#0
668 adcs @acc[8],@acc[8],@acc[0]
670 subs @acc[7],@acc[7],@acc[0]
671 sbcs @acc[8],@acc[8],#0
672 sbc @acc[0],$t3,#0 @ overflow bit
674 @ Final step is "if result > mod, subtract mod", but we do it
675 @ "other way around", namely subtract modulus from result
676 @ and if it borrowed, add modulus back.
678 subs @acc[1],@acc[1],#-1 @ compare to modulus
679 sbcs @acc[2],@acc[2],#-1
680 sbcs @acc[3],@acc[3],#-1
681 sbcs @acc[4],@acc[4],#0
682 sbcs @acc[5],@acc[5],#0
683 sbcs @acc[6],@acc[6],#0
684 sbcs @acc[7],@acc[7],#1
685 sbcs @acc[8],@acc[8],#-1
686 ldr lr,[sp,#44] @ restore lr
687 sbc @acc[0],@acc[0],#0 @ broadcast borrow bit
690 @ Note that because mod has special form, i.e. consists of
691 @ 0xffffffff, 1 and 0s, we can conditionally synthesize it by
692 @ broadcasting borrow bit to a register, @acc[0], and using it as
693 @ a whole or extracting single bit.
695 adds @acc[1],@acc[1],@acc[0] @ add modulus or zero
696 adcs @acc[2],@acc[2],@acc[0]
697 str @acc[1],[$r_ptr,#0]
698 adcs @acc[3],@acc[3],@acc[0]
699 str @acc[2],[$r_ptr,#4]
700 adcs @acc[4],@acc[4],#0
701 str @acc[3],[$r_ptr,#8]
702 adcs @acc[5],@acc[5],#0
703 str @acc[4],[$r_ptr,#12]
704 adcs @acc[6],@acc[6],#0
705 str @acc[5],[$r_ptr,#16]
706 adcs @acc[7],@acc[7],@acc[0],lsr#31
707 str @acc[6],[$r_ptr,#20]
708 adc @acc[8],@acc[8],@acc[0]
709 str @acc[7],[$r_ptr,#24]
710 str @acc[8],[$r_ptr,#28]
713 .size _ecp_nistz256_mul_mont,.-_ecp_nistz256_mul_mont
718 my ($out,$inp,$index,$mask)=map("r$_",(0..3));
720 @ void ecp_nistz256_scatter_w5(void *r0,const P256_POINT *r1,
722 .globl ecp_nistz256_scatter_w5
723 .type ecp_nistz256_scatter_w5,%function
725 ecp_nistz256_scatter_w5:
728 add $out,$out,$index,lsl#2
730 ldmia $inp!,{r4-r11} @ X
731 str r4,[$out,#64*0-4]
732 str r5,[$out,#64*1-4]
733 str r6,[$out,#64*2-4]
734 str r7,[$out,#64*3-4]
735 str r8,[$out,#64*4-4]
736 str r9,[$out,#64*5-4]
737 str r10,[$out,#64*6-4]
738 str r11,[$out,#64*7-4]
741 ldmia $inp!,{r4-r11} @ Y
742 str r4,[$out,#64*0-4]
743 str r5,[$out,#64*1-4]
744 str r6,[$out,#64*2-4]
745 str r7,[$out,#64*3-4]
746 str r8,[$out,#64*4-4]
747 str r9,[$out,#64*5-4]
748 str r10,[$out,#64*6-4]
749 str r11,[$out,#64*7-4]
752 ldmia $inp,{r4-r11} @ Z
753 str r4,[$out,#64*0-4]
754 str r5,[$out,#64*1-4]
755 str r6,[$out,#64*2-4]
756 str r7,[$out,#64*3-4]
757 str r8,[$out,#64*4-4]
758 str r9,[$out,#64*5-4]
759 str r10,[$out,#64*6-4]
760 str r11,[$out,#64*7-4]
763 #if __ARM_ARCH__>=5 || defined(__thumb__)
768 .size ecp_nistz256_scatter_w5,.-ecp_nistz256_scatter_w5
770 @ void ecp_nistz256_gather_w5(P256_POINT *r0,const void *r1,
772 .globl ecp_nistz256_gather_w5
773 .type ecp_nistz256_gather_w5,%function
775 ecp_nistz256_gather_w5:
780 subne $index,$index,#1
782 add $inp,$inp,$index,lsl#2
801 stmia $out!,{r4-r11} @ X
820 stmia $out!,{r4-r11} @ Y
838 stmia $out,{r4-r11} @ Z
841 #if __ARM_ARCH__>=5 || defined(__thumb__)
846 .size ecp_nistz256_gather_w5,.-ecp_nistz256_gather_w5
848 @ void ecp_nistz256_scatter_w7(void *r0,const P256_POINT_AFFINE *r1,
850 .globl ecp_nistz256_scatter_w7
851 .type ecp_nistz256_scatter_w7,%function
853 ecp_nistz256_scatter_w7:
858 subs $index,$index,#1
859 strb $mask,[$out,#64*0-1]
860 mov $mask,$mask,lsr#8
861 strb $mask,[$out,#64*1-1]
862 mov $mask,$mask,lsr#8
863 strb $mask,[$out,#64*2-1]
864 mov $mask,$mask,lsr#8
865 strb $mask,[$out,#64*3-1]
869 #if __ARM_ARCH__>=5 || defined(__thumb__)
874 .size ecp_nistz256_scatter_w7,.-ecp_nistz256_scatter_w7
876 @ void ecp_nistz256_gather_w7(P256_POINT_AFFINE *r0,const void *r1,
878 .globl ecp_nistz256_gather_w7
879 .type ecp_nistz256_gather_w7,%function
881 ecp_nistz256_gather_w7:
886 subne $index,$index,#1
893 subs $index,$index,#1
906 #if __ARM_ARCH__>=5 || defined(__thumb__)
911 .size ecp_nistz256_gather_w7,.-ecp_nistz256_gather_w7
915 # In comparison to integer-only equivalent of below subroutine:
921 # As not all time is spent in multiplication, overall impact is deemed
922 # too low to care about.
924 my ($A0,$A1,$A2,$A3,$Bi,$zero,$temp)=map("d$_",(0..7));
927 my @AxB=map("q$_",(8..15));
929 my ($rptr,$aptr,$bptr,$toutptr)=map("r$_",(0..3));
935 .globl ecp_nistz256_mul_mont_neon
936 .type ecp_nistz256_mul_mont_neon,%function
938 ecp_nistz256_mul_mont_neon:
941 vstmdb sp!,{q4-q5} @ ABI specification says so
944 vld1.32 {${Bi}[0]},[$bptr,:32]!
945 veor $zero,$zero,$zero
946 vld1.32 {$A0-$A3}, [$aptr] @ can't specify :32 :-(
948 mov sp,$toutptr @ alloca
949 vmov.i64 $mask,#0xffff
951 vmull.u32 @AxB[0],$Bi,${A0}[0]
952 vmull.u32 @AxB[1],$Bi,${A0}[1]
953 vmull.u32 @AxB[2],$Bi,${A1}[0]
954 vmull.u32 @AxB[3],$Bi,${A1}[1]
955 vshr.u64 $temp,@AxB[0]#lo,#16
956 vmull.u32 @AxB[4],$Bi,${A2}[0]
957 vadd.u64 @AxB[0]#hi,@AxB[0]#hi,$temp
958 vmull.u32 @AxB[5],$Bi,${A2}[1]
959 vshr.u64 $temp,@AxB[0]#hi,#16 @ upper 32 bits of a[0]*b[0]
960 vmull.u32 @AxB[6],$Bi,${A3}[0]
961 vand.u64 @AxB[0],@AxB[0],$mask @ lower 32 bits of a[0]*b[0]
962 vmull.u32 @AxB[7],$Bi,${A3}[1]
964 for($i=1;$i<8;$i++) {
966 vld1.32 {${Bi}[0]},[$bptr,:32]!
967 veor $zero,$zero,$zero
968 vadd.u64 @AxB[1]#lo,@AxB[1]#lo,$temp @ reduction
969 vshl.u64 $mult,@AxB[0],#32
970 vadd.u64 @AxB[3],@AxB[3],@AxB[0]
971 vsub.u64 $mult,$mult,@AxB[0]
973 vadd.u64 @AxB[6],@AxB[6],@AxB[0]
974 vadd.u64 @AxB[7],@AxB[7],$mult
976 push(@AxB,shift(@AxB));
978 vmlal.u32 @AxB[0],$Bi,${A0}[0]
979 vmlal.u32 @AxB[1],$Bi,${A0}[1]
980 vmlal.u32 @AxB[2],$Bi,${A1}[0]
981 vmlal.u32 @AxB[3],$Bi,${A1}[1]
982 vshr.u64 $temp,@AxB[0]#lo,#16
983 vmlal.u32 @AxB[4],$Bi,${A2}[0]
984 vadd.u64 @AxB[0]#hi,@AxB[0]#hi,$temp
985 vmlal.u32 @AxB[5],$Bi,${A2}[1]
986 vshr.u64 $temp,@AxB[0]#hi,#16 @ upper 33 bits of a[0]*b[i]+t[0]
987 vmlal.u32 @AxB[6],$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]
993 vadd.u64 @AxB[1]#lo,@AxB[1]#lo,$temp @ last reduction
994 vshl.u64 $mult,@AxB[0],#32
995 vadd.u64 @AxB[3],@AxB[3],@AxB[0]
996 vsub.u64 $mult,$mult,@AxB[0]
997 vadd.u64 @AxB[6],@AxB[6],@AxB[0]
998 vadd.u64 @AxB[7],@AxB[7],$mult
1000 vshr.u64 $temp,@AxB[1]#lo,#16 @ convert
1001 vadd.u64 @AxB[1]#hi,@AxB[1]#hi,$temp
1002 vshr.u64 $temp,@AxB[1]#hi,#16
1003 vzip.16 @AxB[1]#lo,@AxB[1]#hi
1007 vadd.u64 @AxB[$_]#lo,@AxB[$_]#lo,$temp
1008 vst1.32 {@AxB[$_-1]#lo[0]},[$toutptr,:32]!
1009 vshr.u64 $temp,@AxB[$_]#lo,#16
1010 vadd.u64 @AxB[$_]#hi,@AxB[$_]#hi,$temp
1011 vshr.u64 $temp,@AxB[$_]#hi,#16
1012 vzip.16 @AxB[$_]#lo,@AxB[$_]#hi
1016 vst1.32 {@AxB[7]#lo[0]},[$toutptr,:32]!
1017 vst1.32 {$temp},[$toutptr] @ upper 33 bits
1033 ldr r9,[sp,#32] @ top-most bit
1051 adcs r7,r7,r9,lsr#31
1059 .size ecp_nistz256_mul_mont_neon,.-ecp_nistz256_mul_mont_neon
1065 ########################################################################
1066 # Below $aN assignment matches order in which 256-bit result appears in
1067 # register bank at return from _ecp_nistz256_mul_mont, so that we can
1068 # skip over reloading it from memory. This means that below functions
1069 # use custom calling sequence accepting 256-bit input in registers,
1070 # output pointer in r0, $r_ptr, and optional pointer in r2, $b_ptr.
1072 # See their "normal" counterparts for insights on calculations.
1074 my ($a0,$a1,$a2,$a3,$a4,$a5,$a6,$a7,
1075 $t0,$t1,$t2,$t3)=map("r$_",(11,3..10,12,14,1));
1079 .type __ecp_nistz256_sub_from,%function
1081 __ecp_nistz256_sub_from:
1082 str lr,[sp,#-4]! @ push lr
1087 ldr $t3,[$b_ptr,#12]
1089 ldr $t0,[$b_ptr,#16]
1091 ldr $t1,[$b_ptr,#20]
1093 ldr $t2,[$b_ptr,#24]
1095 ldr $t3,[$b_ptr,#28]
1100 sbc $ff,$ff,$ff @ broadcast borrow bit
1101 ldr lr,[sp],#4 @ pop lr
1103 adds $a0,$a0,$ff @ add synthesized modulus
1111 str $a3,[$r_ptr,#12]
1113 str $a4,[$r_ptr,#16]
1114 adcs $a6,$a6,$ff,lsr#31
1115 str $a5,[$r_ptr,#20]
1117 str $a6,[$r_ptr,#24]
1118 str $a7,[$r_ptr,#28]
1121 .size __ecp_nistz256_sub_from,.-__ecp_nistz256_sub_from
1123 .type __ecp_nistz256_sub_morf,%function
1125 __ecp_nistz256_sub_morf:
1126 str lr,[sp,#-4]! @ push lr
1131 ldr $t3,[$b_ptr,#12]
1133 ldr $t0,[$b_ptr,#16]
1135 ldr $t1,[$b_ptr,#20]
1137 ldr $t2,[$b_ptr,#24]
1139 ldr $t3,[$b_ptr,#28]
1144 sbc $ff,$ff,$ff @ broadcast borrow bit
1145 ldr lr,[sp],#4 @ pop lr
1147 adds $a0,$a0,$ff @ add synthesized modulus
1155 str $a3,[$r_ptr,#12]
1157 str $a4,[$r_ptr,#16]
1158 adcs $a6,$a6,$ff,lsr#31
1159 str $a5,[$r_ptr,#20]
1161 str $a6,[$r_ptr,#24]
1162 str $a7,[$r_ptr,#28]
1165 .size __ecp_nistz256_sub_morf,.-__ecp_nistz256_sub_morf
1167 .type __ecp_nistz256_mul_by_2,%function
1169 __ecp_nistz256_mul_by_2:
1170 adds $a0,$a0,$a0 @ a[0:7]+=a[0:7]
1179 movcs $ff,#-1 @ $ff = carry ? -1 : 0
1181 subs $a0,$a0,$ff @ subtract synthesized modulus
1189 str $a3,[$r_ptr,#12]
1191 str $a4,[$r_ptr,#16]
1192 sbcs $a6,$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_mul_by_2,.-__ecp_nistz256_mul_by_2
1203 ########################################################################
1204 # following subroutines are "literal" implemetation of those found in
1207 ########################################################################
1208 # void ecp_nistz256_point_double(P256_POINT *out,const P256_POINT *inp);
1211 my ($S,$M,$Zsqr,$in_x,$tmp0)=map(32*$_,(0..4));
1212 # above map() describes stack layout with 5 temporary
1213 # 256-bit vectors on top. Then note that we push
1214 # starting from r0, which means that we have copy of
1215 # input arguments just below these temporary vectors.
1218 .globl ecp_nistz256_point_double
1219 .type ecp_nistz256_point_double,%function
1221 ecp_nistz256_point_double:
1222 stmdb sp!,{r0-r12,lr} @ push from r0, unusual, but intentional
1226 ldmia $a_ptr!,{r4-r11} @ copy in_x
1230 bl _ecp_nistz256_mul_by_2 @ p256_mul_by_2(S, in_y);
1232 add $b_ptr,$a_ptr,#32
1233 add $a_ptr,$a_ptr,#32
1234 add $r_ptr,sp,#$Zsqr
1235 bl _ecp_nistz256_mul_mont @ p256_sqr_mont(Zsqr, in_z);
1240 bl _ecp_nistz256_mul_mont @ p256_sqr_mont(S, S);
1242 ldr $b_ptr,[sp,#32*5+4]
1243 add $a_ptr,$b_ptr,#32
1244 add $b_ptr,$b_ptr,#64
1245 add $r_ptr,sp,#$tmp0
1246 bl _ecp_nistz256_mul_mont @ p256_mul_mont(tmp0, in_z, in_y);
1248 ldr $r_ptr,[sp,#32*5]
1249 add $r_ptr,$r_ptr,#64
1250 bl __ecp_nistz256_mul_by_2 @ p256_mul_by_2(res_z, tmp0);
1252 add $a_ptr,sp,#$in_x
1253 add $b_ptr,sp,#$Zsqr
1255 bl _ecp_nistz256_add @ p256_add(M, in_x, Zsqr);
1257 add $a_ptr,sp,#$in_x
1258 add $b_ptr,sp,#$Zsqr
1259 add $r_ptr,sp,#$Zsqr
1260 bl _ecp_nistz256_sub @ p256_sub(Zsqr, in_x, Zsqr);
1264 add $r_ptr,sp,#$tmp0
1265 bl _ecp_nistz256_mul_mont @ p256_sqr_mont(tmp0, S);
1267 add $a_ptr,sp,#$Zsqr
1270 bl _ecp_nistz256_mul_mont @ p256_mul_mont(M, M, Zsqr);
1272 ldr $r_ptr,[sp,#32*5]
1273 add $a_ptr,sp,#$tmp0
1274 add $r_ptr,$r_ptr,#32
1275 bl _ecp_nistz256_div_by_2 @ p256_div_by_2(res_y, tmp0);
1279 bl _ecp_nistz256_mul_by_3 @ p256_mul_by_3(M, M);
1281 add $a_ptr,sp,#$in_x
1284 bl _ecp_nistz256_mul_mont @ p256_mul_mont(S, S, in_x);
1286 add $r_ptr,sp,#$tmp0
1287 bl __ecp_nistz256_mul_by_2 @ p256_mul_by_2(tmp0, S);
1289 ldr $r_ptr,[sp,#32*5]
1292 bl _ecp_nistz256_mul_mont @ p256_sqr_mont(res_x, M);
1294 add $b_ptr,sp,#$tmp0
1295 bl __ecp_nistz256_sub_from @ p256_sub(res_x, res_x, tmp0);
1299 bl __ecp_nistz256_sub_morf @ p256_sub(S, S, res_x);
1303 bl _ecp_nistz256_mul_mont @ p256_mul_mont(S, S, M);
1305 ldr $r_ptr,[sp,#32*5]
1306 add $b_ptr,$r_ptr,#32
1307 add $r_ptr,$r_ptr,#32
1308 bl __ecp_nistz256_sub_from @ p256_sub(res_y, S, res_y);
1310 add sp,sp,#32*5+16 @ +16 means "skip even over saved r0-r3"
1311 #if __ARM_ARCH__>=5 || !defined(__thumb__)
1312 ldmia sp!,{r4-r12,pc}
1314 ldmia sp!,{r4-r12,lr}
1315 bx lr @ interoperable with Thumb ISA:-)
1317 .size ecp_nistz256_point_double,.-ecp_nistz256_point_double
1321 ########################################################################
1322 # void ecp_nistz256_point_add(P256_POINT *out,const P256_POINT *in1,
1323 # const P256_POINT *in2);
1325 my ($res_x,$res_y,$res_z,
1326 $in1_x,$in1_y,$in1_z,
1327 $in2_x,$in2_y,$in2_z,
1328 $H,$Hsqr,$R,$Rsqr,$Hcub,
1329 $U1,$U2,$S1,$S2)=map(32*$_,(0..17));
1330 my ($Z1sqr, $Z2sqr) = ($Hsqr, $Rsqr);
1331 # above map() describes stack layout with 18 temporary
1332 # 256-bit vectors on top. Then note that we push
1333 # starting from r0, which means that we have copy of
1334 # input arguments just below these temporary vectors.
1335 # We use three of them for !in1infty, !in2intfy and
1336 # result of check for zero.
1339 .globl ecp_nistz256_point_add
1340 .type ecp_nistz256_point_add,%function
1342 ecp_nistz256_point_add:
1343 stmdb sp!,{r0-r12,lr} @ push from r0, unusual, but intentional
1346 ldmia $b_ptr!,{r4-r11} @ copy in2
1356 ldmia $b_ptr!,{r4-r11}
1366 ldmia $b_ptr,{r4-r11}
1370 str r12,[sp,#32*18+8] @ !in2infty
1372 ldmia $a_ptr!,{r4-r11} @ copy in1
1382 ldmia $a_ptr!,{r4-r11}
1392 ldmia $a_ptr,{r4-r11}
1396 str r12,[sp,#32*18+4] @ !in1infty
1398 add $a_ptr,sp,#$in2_z
1399 add $b_ptr,sp,#$in2_z
1400 add $r_ptr,sp,#$Z2sqr
1401 bl _ecp_nistz256_mul_mont @ p256_sqr_mont(Z2sqr, in2_z);
1403 add $a_ptr,sp,#$in1_z
1404 add $b_ptr,sp,#$in1_z
1405 add $r_ptr,sp,#$Z1sqr
1406 bl _ecp_nistz256_mul_mont @ p256_sqr_mont(Z1sqr, in1_z);
1408 add $a_ptr,sp,#$in2_z
1409 add $b_ptr,sp,#$Z2sqr
1411 bl _ecp_nistz256_mul_mont @ p256_mul_mont(S1, Z2sqr, in2_z);
1413 add $a_ptr,sp,#$in1_z
1414 add $b_ptr,sp,#$Z1sqr
1416 bl _ecp_nistz256_mul_mont @ p256_mul_mont(S2, Z1sqr, in1_z);
1418 add $a_ptr,sp,#$in1_y
1421 bl _ecp_nistz256_mul_mont @ p256_mul_mont(S1, S1, in1_y);
1423 add $a_ptr,sp,#$in2_y
1426 bl _ecp_nistz256_mul_mont @ p256_mul_mont(S2, S2, in2_y);
1430 bl __ecp_nistz256_sub_from @ p256_sub(R, S2, S1);
1432 orr $a0,$a0,$a1 @ see if result is zero
1438 add $a_ptr,sp,#$in1_x
1440 add $b_ptr,sp,#$Z2sqr
1441 str $a0,[sp,#32*18+12]
1444 bl _ecp_nistz256_mul_mont @ p256_mul_mont(U1, in1_x, Z2sqr);
1446 add $a_ptr,sp,#$in2_x
1447 add $b_ptr,sp,#$Z1sqr
1449 bl _ecp_nistz256_mul_mont @ p256_mul_mont(U2, in2_x, Z1sqr);
1453 bl __ecp_nistz256_sub_from @ p256_sub(H, U2, U1);
1455 orr $a0,$a0,$a1 @ see if result is zero
1463 bne .Ladd_proceed @ is_equal(U1,U2)?
1465 ldr $t0,[sp,#32*18+4]
1466 ldr $t1,[sp,#32*18+8]
1467 ldr $t2,[sp,#32*18+12]
1469 beq .Ladd_proceed @ (in1infty || in2infty)?
1471 beq .Ladd_proceed @ is_equal(S1,S2)?
1473 ldr $r_ptr,[sp,#32*18]
1482 stmia $r_ptr!,{r4-r11}
1483 stmia $r_ptr!,{r4-r11}
1484 stmia $r_ptr!,{r4-r11}
1491 add $r_ptr,sp,#$Rsqr
1492 bl _ecp_nistz256_mul_mont @ p256_sqr_mont(Rsqr, R);
1495 add $b_ptr,sp,#$in1_z
1496 add $r_ptr,sp,#$res_z
1497 bl _ecp_nistz256_mul_mont @ p256_mul_mont(res_z, H, in1_z);
1501 add $r_ptr,sp,#$Hsqr
1502 bl _ecp_nistz256_mul_mont @ p256_sqr_mont(Hsqr, H);
1504 add $a_ptr,sp,#$in2_z
1505 add $b_ptr,sp,#$res_z
1506 add $r_ptr,sp,#$res_z
1507 bl _ecp_nistz256_mul_mont @ p256_mul_mont(res_z, res_z, in2_z);
1510 add $b_ptr,sp,#$Hsqr
1511 add $r_ptr,sp,#$Hcub
1512 bl _ecp_nistz256_mul_mont @ p256_mul_mont(Hcub, Hsqr, H);
1514 add $a_ptr,sp,#$Hsqr
1517 bl _ecp_nistz256_mul_mont @ p256_mul_mont(U2, U1, Hsqr);
1519 add $r_ptr,sp,#$Hsqr
1520 bl __ecp_nistz256_mul_by_2 @ p256_mul_by_2(Hsqr, U2);
1522 add $b_ptr,sp,#$Rsqr
1523 add $r_ptr,sp,#$res_x
1524 bl __ecp_nistz256_sub_morf @ p256_sub(res_x, Rsqr, Hsqr);
1526 add $b_ptr,sp,#$Hcub
1527 bl __ecp_nistz256_sub_from @ p256_sub(res_x, res_x, Hcub);
1530 add $r_ptr,sp,#$res_y
1531 bl __ecp_nistz256_sub_morf @ p256_sub(res_y, U2, res_x);
1533 add $a_ptr,sp,#$Hcub
1536 bl _ecp_nistz256_mul_mont @ p256_mul_mont(S2, S1, Hcub);
1539 add $b_ptr,sp,#$res_y
1540 add $r_ptr,sp,#$res_y
1541 bl _ecp_nistz256_mul_mont @ p256_mul_mont(res_y, res_y, R);
1544 bl __ecp_nistz256_sub_from @ p256_sub(res_y, res_y, S2);
1546 ldr r11,[sp,#32*18+4] @ !in1intfy
1547 ldr r12,[sp,#32*18+8] @ !in2intfy
1555 ldr $r_ptr,[sp,#32*18]
1557 for($i=0;$i<96;$i+=8) { # conditional moves
1559 ldmia r1!,{r4-r5} @ res_x
1560 ldmia r2!,{r6-r7} @ in2_x
1561 ldmia r3!,{r8-r9} @ in1_x
1572 stmia $r_ptr!,{r4-r5}
1577 add sp,sp,#32*18+16 @ +16 means "skip even over saved r0-r3"
1578 #if __ARM_ARCH__>=5 || defined(__thumb__)
1579 ldmia sp!,{r4-r12,pc}
1581 ldmia sp!,{r4-r12,lr}
1582 bx lr @ interoperable with Thumb ISA:-)
1584 .size ecp_nistz256_point_add,.-ecp_nistz256_point_add
1588 ########################################################################
1589 # void ecp_nistz256_point_add_affine(P256_POINT *out,const P256_POINT *in1,
1590 # const P256_POINT_AFFINE *in2);
1592 my ($res_x,$res_y,$res_z,
1593 $in1_x,$in1_y,$in1_z,
1595 $U2,$S2,$H,$R,$Hsqr,$Hcub,$Rsqr)=map(32*$_,(0..14));
1597 # above map() describes stack layout with 18 temporary
1598 # 256-bit vectors on top. Then note that we push
1599 # starting from r0, which means that we have copy of
1600 # input arguments just below these temporary vectors.
1601 # We use two of them for !in1infty, !in2intfy.
1603 my @ONE_mont=(1,0,0,-1,-1,-1,-2,0);
1606 .globl ecp_nistz256_point_add_affine
1607 .type ecp_nistz256_point_add_affine,%function
1609 ecp_nistz256_point_add_affine:
1610 stmdb sp!,{r0-r12,lr} @ push from r0, unusual, but intentional
1613 ldmia $a_ptr!,{r4-r11} @ copy in1
1623 ldmia $a_ptr!,{r4-r11}
1633 ldmia $a_ptr,{r4-r11}
1637 str r12,[sp,#32*15+4] @ !in1infty
1639 ldmia $b_ptr!,{r4-r11} @ copy in2
1649 ldmia $b_ptr!,{r4-r11}
1661 str r12,[sp,#32*15+8] @ !in2infty
1663 add $a_ptr,sp,#$in1_z
1664 add $b_ptr,sp,#$in1_z
1665 add $r_ptr,sp,#$Z1sqr
1666 bl _ecp_nistz256_mul_mont @ p256_sqr_mont(Z1sqr, in1_z);
1668 add $a_ptr,sp,#$Z1sqr
1669 add $b_ptr,sp,#$in2_x
1671 bl _ecp_nistz256_mul_mont @ p256_mul_mont(U2, Z1sqr, in2_x);
1673 add $b_ptr,sp,#$in1_x
1675 bl __ecp_nistz256_sub_from @ p256_sub(H, U2, in1_x);
1677 add $a_ptr,sp,#$Z1sqr
1678 add $b_ptr,sp,#$in1_z
1680 bl _ecp_nistz256_mul_mont @ p256_mul_mont(S2, Z1sqr, in1_z);
1683 add $b_ptr,sp,#$in1_z
1684 add $r_ptr,sp,#$res_z
1685 bl _ecp_nistz256_mul_mont @ p256_mul_mont(res_z, H, in1_z);
1687 add $a_ptr,sp,#$in2_y
1690 bl _ecp_nistz256_mul_mont @ p256_mul_mont(S2, S2, in2_y);
1692 add $b_ptr,sp,#$in1_y
1694 bl __ecp_nistz256_sub_from @ p256_sub(R, S2, in1_y);
1698 add $r_ptr,sp,#$Hsqr
1699 bl _ecp_nistz256_mul_mont @ p256_sqr_mont(Hsqr, H);
1703 add $r_ptr,sp,#$Rsqr
1704 bl _ecp_nistz256_mul_mont @ p256_sqr_mont(Rsqr, R);
1707 add $b_ptr,sp,#$Hsqr
1708 add $r_ptr,sp,#$Hcub
1709 bl _ecp_nistz256_mul_mont @ p256_mul_mont(Hcub, Hsqr, H);
1711 add $a_ptr,sp,#$Hsqr
1712 add $b_ptr,sp,#$in1_x
1714 bl _ecp_nistz256_mul_mont @ p256_mul_mont(U2, in1_x, Hsqr);
1716 add $r_ptr,sp,#$Hsqr
1717 bl __ecp_nistz256_mul_by_2 @ p256_mul_by_2(Hsqr, U2);
1719 add $b_ptr,sp,#$Rsqr
1720 add $r_ptr,sp,#$res_x
1721 bl __ecp_nistz256_sub_morf @ p256_sub(res_x, Rsqr, Hsqr);
1723 add $b_ptr,sp,#$Hcub
1724 bl __ecp_nistz256_sub_from @ p256_sub(res_x, res_x, Hcub);
1727 add $r_ptr,sp,#$res_y
1728 bl __ecp_nistz256_sub_morf @ p256_sub(res_y, U2, res_x);
1730 add $a_ptr,sp,#$Hcub
1731 add $b_ptr,sp,#$in1_y
1733 bl _ecp_nistz256_mul_mont @ p256_mul_mont(S2, in1_y, Hcub);
1736 add $b_ptr,sp,#$res_y
1737 add $r_ptr,sp,#$res_y
1738 bl _ecp_nistz256_mul_mont @ p256_mul_mont(res_y, res_y, R);
1741 bl __ecp_nistz256_sub_from @ p256_sub(res_y, res_y, S2);
1743 ldr r11,[sp,#32*15+4] @ !in1intfy
1744 ldr r12,[sp,#32*15+8] @ !in2intfy
1752 ldr $r_ptr,[sp,#32*15]
1754 for($i=0;$i<64;$i+=8) { # conditional moves
1756 ldmia r1!,{r4-r5} @ res_x
1757 ldmia r2!,{r6-r7} @ in2_x
1758 ldmia r3!,{r8-r9} @ in1_x
1769 stmia $r_ptr!,{r4-r5}
1775 ldmia r1!,{r4-r5} @ res_z
1776 ldmia r3!,{r8-r9} @ in1_z
1779 and r6,r11,#@ONE_mont[$j]
1780 and r7,r11,#@ONE_mont[$j+1]
1787 stmia $r_ptr!,{r4-r5}
1791 add sp,sp,#32*15+16 @ +16 means "skip even over saved r0-r3"
1792 #if __ARM_ARCH__>=5 || !defined(__thumb__)
1793 ldmia sp!,{r4-r12,pc}
1795 ldmia sp!,{r4-r12,lr}
1796 bx lr @ interoperable with Thumb ISA:-)
1798 .size ecp_nistz256_point_add_affine,.-ecp_nistz256_point_add_affine
1802 foreach (split("\n",$code)) {
1803 s/\`([^\`]*)\`/eval $1/geo;
1805 s/\bq([0-9]+)#(lo|hi)/sprintf "d%d",2*$1+($2 eq "hi")/geo;
1809 close STDOUT; # enforce flush