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";
50 ########################################################################
51 # Convert ecp_nistz256_table.c to layout expected by ecp_nistz_gather_w7
53 $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
54 open TABLE,"<ecp_nistz256_table.c" or
55 open TABLE,"<${dir}../ecp_nistz256_table.c" or
56 die "failed to open ecp_nistz256_table.c:",$!;
61 s/TOBN\(\s*(0x[0-9a-f]+),\s*(0x[0-9a-f]+)\s*\)/push @arr,hex($2),hex($1)/geo;
65 # See ecp_nistz256_table.c for explanation for why it's 64*16*37.
66 # 64*16*37-1 is because $#arr returns last valid index or @arr, not
68 die "insane number of elements" if ($#arr != 64*16*37-1);
71 .globl ecp_nistz256_precomputed
72 .type ecp_nistz256_precomputed,%object
74 ecp_nistz256_precomputed:
76 ########################################################################
77 # this conversion smashes P256_POINT_AFFINE by individual bytes with
78 # 64 byte interval, similar to
82 @tbl = splice(@arr,0,64*16);
83 for($i=0;$i<64;$i++) {
85 for($j=0;$j<64;$j++) {
86 push @line,(@tbl[$j*16+$i/4]>>(($i%4)*8))&0xff;
89 $code.=join(',',map { sprintf "0x%02x",$_} @line);
94 .size ecp_nistz256_precomputed,.-ecp_nistz256_precomputed
96 .LRR: @ 2^512 mod P precomputed for NIST P256 polynomial
97 .long 0x00000003, 0x00000000, 0xffffffff, 0xfffffffb
98 .long 0xfffffffe, 0xffffffff, 0xfffffffd, 0x00000004
100 .long 1,0,0,0,0,0,0,0
101 .asciz "ECP_NISTZ256 for ARMv4, CRYPTOGAMS by <appro\@openssl.org>"
105 ########################################################################
106 # common register layout, note that $t2 is link register, so that if
107 # internal subroutine uses $t2, then it has to offload lr...
109 ($r_ptr,$a_ptr,$b_ptr,$ff,$a0,$a1,$a2,$a3,$a4,$a5,$a6,$a7,$t1,$t2)=
110 map("r$_",(0..12,14));
111 ($t0,$t3)=($ff,$a_ptr);
114 @ void ecp_nistz256_to_mont(BN_ULONG r0[8],const BN_ULONG r1[8]);
115 .globl ecp_nistz256_to_mont
116 .type ecp_nistz256_to_mont,%function
117 ecp_nistz256_to_mont:
119 b .Lecp_nistz256_mul_mont
120 .size ecp_nistz256_to_mont,.-ecp_nistz256_to_mont
122 @ void ecp_nistz256_from_mont(BN_ULONG r0[8],const BN_ULONG r1[8]);
123 .globl ecp_nistz256_from_mont
124 .type ecp_nistz256_from_mont,%function
125 ecp_nistz256_from_mont:
127 b .Lecp_nistz256_mul_mont
128 .size ecp_nistz256_from_mont,.-ecp_nistz256_from_mont
130 @ void ecp_nistz256_mul_by_2(BN_ULONG r0[8],const BN_ULONG r1[8]);
131 .globl ecp_nistz256_mul_by_2
132 .type ecp_nistz256_mul_by_2,%function
134 ecp_nistz256_mul_by_2:
135 stmdb sp!,{r4-r12,lr}
136 bl __ecp_nistz256_mul_by_2
137 #if __ARM_ARCH__>=5 || !defined(__thumb__)
138 ldmia sp!,{r4-r12,pc}
140 ldmia sp!,{r4-r12,lr}
141 bx lr @ interoperable with Thumb ISA:-)
143 .size ecp_nistz256_mul_by_2,.-ecp_nistz256_mul_by_2
145 .type __ecp_nistz256_mul_by_2,%function
147 __ecp_nistz256_mul_by_2:
151 adds $a0,$a0,$a0 @ a[0:7]+=a[0:7], i.e. add with itself
165 movcs $ff,#-1 @ $ff = carry ? -1 : 0
168 .size __ecp_nistz256_mul_by_2,.-__ecp_nistz256_mul_by_2
170 @ void ecp_nistz256_add(BN_ULONG r0[8],const BN_ULONG r1[8],
171 @ const BN_ULONG r2[8]);
172 .globl ecp_nistz256_add
173 .type ecp_nistz256_add,%function
176 stmdb sp!,{r4-r12,lr}
177 bl __ecp_nistz256_add
178 #if __ARM_ARCH__>=5 || !defined(__thumb__)
179 ldmia sp!,{r4-r12,pc}
181 ldmia sp!,{r4-r12,lr}
182 bx lr @ interoperable with Thumb ISA:-)
184 .size ecp_nistz256_add,.-ecp_nistz256_add
186 .type __ecp_nistz256_add,%function
189 str lr,[sp,#-4]! @ push lr
216 movcs $ff,#-1 @ $ff = carry ? -1 : 0, "broadcast" carry
217 ldr lr,[sp],#4 @ pop lr
221 @ if a+b carries, subtract modulus.
223 @ Note that because mod has special form, i.e. consists of
224 @ 0xffffffff, 1 and 0s, we can conditionally synthesize it by
225 @ using value of broadcasted carry as a whole or extracting
226 @ single bit. Follow $ff register...
228 subs $a0,$a0,$ff @ subtract synthesized modulus
239 sbcs $a6,$a6,$ff,lsr#31
246 .size __ecp_nistz256_add,.-__ecp_nistz256_add
248 @ void ecp_nistz256_mul_by_3(BN_ULONG r0[8],const BN_ULONG r1[8]);
249 .globl ecp_nistz256_mul_by_3
250 .type ecp_nistz256_mul_by_3,%function
252 ecp_nistz256_mul_by_3:
253 stmdb sp!,{r4-r12,lr}
254 bl __ecp_nistz256_mul_by_3
255 #if __ARM_ARCH__>=5 || !defined(__thumb__)
256 ldmia sp!,{r4-r12,pc}
258 ldmia sp!,{r4-r12,lr}
259 bx lr @ interoperable with Thumb ISA:-)
261 .size ecp_nistz256_mul_by_3,.-ecp_nistz256_mul_by_3
263 .type __ecp_nistz256_mul_by_3,%function
265 __ecp_nistz256_mul_by_3:
266 str lr,[sp,#-4]! @ push lr
268 @ As multiplication by 3 is performed as 2*n+n, below are inline
269 @ copies of __ecp_nistz256_mul_by_2 and __ecp_nistz256_add, see
270 @ corresponding subroutines for details.
275 adds $a0,$a0,$a0 @ a[0:7]+=a[0:7]
289 movcs $ff,#-1 @ $ff = carry ? -1 : 0, "broadcast" carry
291 subs $a0,$a0,$ff @ subtract synthesized modulus, see
292 @ .Lreduce_by_sub for details, except
293 @ that we don't write anything to
294 @ memory, but keep intermediate
295 @ results in registers...
300 ldr $b_ptr,[$a_ptr,#0]
303 sbcs $a6,$a6,$ff,lsr#31
308 adds $a0,$a0,$b_ptr @ 2*a[0:7]+=a[0:7]
309 ldr $b_ptr,[$a_ptr,#16]
321 movcs $ff,#-1 @ $ff = carry ? -1 : 0, "broadcast" carry
322 ldr lr,[sp],#4 @ pop lr
325 .size ecp_nistz256_mul_by_3,.-ecp_nistz256_mul_by_3
327 @ void ecp_nistz256_div_by_2(BN_ULONG r0[8],const BN_ULONG r1[8]);
328 .globl ecp_nistz256_div_by_2
329 .type ecp_nistz256_div_by_2,%function
331 ecp_nistz256_div_by_2:
332 stmdb sp!,{r4-r12,lr}
333 bl __ecp_nistz256_div_by_2
334 #if __ARM_ARCH__>=5 || !defined(__thumb__)
335 ldmia sp!,{r4-r12,pc}
337 ldmia sp!,{r4-r12,lr}
338 bx lr @ interoperable with Thumb ISA:-)
340 .size ecp_nistz256_div_by_2,.-ecp_nistz256_div_by_2
342 .type __ecp_nistz256_div_by_2,%function
344 __ecp_nistz256_div_by_2:
345 @ ret = (a is odd ? a+mod : a) >> 1
350 mov $ff,$a0,lsl#31 @ place least significant bit to most
351 @ significant position, now arithmetic
352 @ right shift by 31 will produce -1 or
353 @ 0, while logical rigth shift 1 or 0,
354 @ this is how modulus is conditionally
355 @ synthesized in this case...
357 adds $a0,$a0,$ff,asr#31
359 adcs $a1,$a1,$ff,asr#31
361 adcs $a2,$a2,$ff,asr#31
366 mov $a0,$a0,lsr#1 @ a[0:7]>>=1, we can start early
367 @ because it doesn't affect flags
369 orr $a0,$a0,$a1,lsl#31
370 adcs $a6,$a6,$ff,lsr#31
372 adcs $a7,$a7,$ff,asr#31
374 adc $b_ptr,$b_ptr,#0 @ top-most carry bit from addition
376 orr $a1,$a1,$a2,lsl#31
379 orr $a2,$a2,$a3,lsl#31
382 orr $a3,$a3,$a4,lsl#31
385 orr $a4,$a4,$a5,lsl#31
388 orr $a5,$a5,$a6,lsl#31
391 orr $a6,$a6,$a7,lsl#31
394 orr $a7,$a7,$b_ptr,lsl#31 @ don't forget the top-most carry bit
399 .size __ecp_nistz256_div_by_2,.-__ecp_nistz256_div_by_2
401 @ void ecp_nistz256_sub(BN_ULONG r0[8],const BN_ULONG r1[8],
402 @ const BN_ULONG r2[8]);
403 .globl ecp_nistz256_sub
404 .type ecp_nistz256_sub,%function
407 stmdb sp!,{r4-r12,lr}
408 bl __ecp_nistz256_sub
409 #if __ARM_ARCH__>=5 || !defined(__thumb__)
410 ldmia sp!,{r4-r12,pc}
412 ldmia sp!,{r4-r12,lr}
413 bx lr @ interoperable with Thumb ISA:-)
415 .size ecp_nistz256_sub,.-ecp_nistz256_sub
417 .type __ecp_nistz256_sub,%function
420 str lr,[sp,#-4]! @ push lr
446 sbc $ff,$ff,$ff @ broadcast borrow bit
447 ldr lr,[sp],#4 @ pop lr
451 @ if a-b borrows, add modulus.
453 @ Note that because mod has special form, i.e. consists of
454 @ 0xffffffff, 1 and 0s, we can conditionally synthesize it by
455 @ broadcasting borrow bit to a register, $ff, and using it as
456 @ a whole or extracting single bit.
458 adds $a0,$a0,$ff @ add synthesized modulus
469 adcs $a6,$a6,$ff,lsr#31
476 .size __ecp_nistz256_sub,.-__ecp_nistz256_sub
478 @ void ecp_nistz256_neg(BN_ULONG r0[8],const BN_ULONG r1[8]);
479 .globl ecp_nistz256_neg
480 .type ecp_nistz256_neg,%function
483 stmdb sp!,{r4-r12,lr}
484 bl __ecp_nistz256_neg
485 #if __ARM_ARCH__>=5 || !defined(__thumb__)
486 ldmia sp!,{r4-r12,pc}
488 ldmia sp!,{r4-r12,lr}
489 bx lr @ interoperable with Thumb ISA:-)
491 .size ecp_nistz256_neg,.-ecp_nistz256_neg
493 .type __ecp_nistz256_neg,%function
516 .size __ecp_nistz256_neg,.-__ecp_nistz256_neg
519 my @acc=map("r$_",(3..11));
520 my ($t0,$t1,$bj,$t2,$t3)=map("r$_",(0,1,2,12,14));
523 @ void ecp_nistz256_sqr_mont(BN_ULONG r0[8],const BN_ULONG r1[8]);
524 .globl ecp_nistz256_sqr_mont
525 .type ecp_nistz256_sqr_mont,%function
527 ecp_nistz256_sqr_mont:
529 b .Lecp_nistz256_mul_mont
530 .size ecp_nistz256_sqr_mont,.-ecp_nistz256_sqr_mont
532 @ void ecp_nistz256_mul_mont(BN_ULONG r0[8],const BN_ULONG r1[8],
533 @ const BN_ULONG r2[8]);
534 .globl ecp_nistz256_mul_mont
535 .type ecp_nistz256_mul_mont,%function
537 ecp_nistz256_mul_mont:
538 .Lecp_nistz256_mul_mont:
539 stmdb sp!,{r4-r12,lr}
540 bl __ecp_nistz256_mul_mont
541 #if __ARM_ARCH__>=5 || !defined(__thumb__)
542 ldmia sp!,{r4-r12,pc}
544 ldmia sp!,{r4-r12,lr}
545 bx lr @ interoperable with Thumb ISA:-)
547 .size ecp_nistz256_mul_mont,.-ecp_nistz256_mul_mont
549 .type __ecp_nistz256_mul_mont,%function
551 __ecp_nistz256_mul_mont:
552 stmdb sp!,{r0-r2,lr} @ make a copy of arguments too
554 ldr $bj,[$b_ptr,#0] @ b[0]
555 ldmia $a_ptr,{@acc[1]-@acc[8]}
557 umull @acc[0],$t3,@acc[1],$bj @ r[0]=a[0]*b[0]
558 stmdb sp!,{$acc[1]-@acc[8]} @ copy a[0-7] to stack, so
559 @ that it can be addressed
560 @ without spending register
562 umull @acc[1],$t0,@acc[2],$bj @ r[1]=a[1]*b[0]
563 umull @acc[2],$t1,@acc[3],$bj
564 adds @acc[1],@acc[1],$t3 @ accumulate high part of mult
565 umull @acc[3],$t2,@acc[4],$bj
566 adcs @acc[2],@acc[2],$t0
567 umull @acc[4],$t3,@acc[5],$bj
568 adcs @acc[3],@acc[3],$t1
569 umull @acc[5],$t0,@acc[6],$bj
570 adcs @acc[4],@acc[4],$t2
571 umull @acc[6],$t1,@acc[7],$bj
572 adcs @acc[5],@acc[5],$t3
573 umull @acc[7],$t2,@acc[8],$bj
574 adcs @acc[6],@acc[6],$t0
575 adcs @acc[7],@acc[7],$t1
576 eor $t3,$t3,$t3 @ first overflow bit is zero
579 for(my $i=1;$i<8;$i++) {
582 # Reduction iteration is normally performed by accumulating
583 # result of multiplication of modulus by "magic" digit [and
584 # omitting least significant word, which is guaranteed to
585 # be 0], but thanks to special form of modulus and "magic"
586 # digit being equal to least significant word, it can be
587 # performed with additions and subtractions alone. Indeed:
589 # ffff.0001.0000.0000.0000.ffff.ffff.ffff
591 # + xxxx.xxxx.xxxx.xxxx.xxxx.xxxx.xxxx.xxxx.abcd
593 # Now observing that ff..ff*x = (2^n-1)*x = 2^n*x-x, we
596 # xxxx.xxxx.xxxx.xxxx.xxxx.xxxx.xxxx.xxxx.abcd
597 # + abcd.0000.abcd.0000.0000.abcd.0000.0000.0000
598 # - abcd.0000.0000.0000.0000.0000.0000.abcd
600 # or marking redundant operations:
602 # xxxx.xxxx.xxxx.xxxx.xxxx.xxxx.xxxx.xxxx.----
603 # + abcd.0000.abcd.0000.0000.abcd.----.----.----
604 # - abcd.----.----.----.----.----.----.----
607 @ multiplication-less reduction $i
608 adds @acc[3],@acc[3],@acc[0] @ r[3]+=r[0]
609 ldr $bj,[sp,#40] @ restore b_ptr
610 adcs @acc[4],@acc[4],#0 @ r[4]+=0
611 adcs @acc[5],@acc[5],#0 @ r[5]+=0
612 adcs @acc[6],@acc[6],@acc[0] @ r[6]+=r[0]
613 ldr $t1,[sp,#0] @ load a[0]
614 adcs @acc[7],@acc[7],#0 @ r[7]+=0
615 ldr $bj,[$bj,#4*$i] @ load b[i]
616 adcs @acc[8],@acc[8],@acc[0] @ r[8]+=r[0]
618 adc $t3,$t3,#0 @ overflow bit
619 subs @acc[7],@acc[7],@acc[0] @ r[7]-=r[0]
620 ldr $t2,[sp,#4] @ a[1]
621 sbcs @acc[8],@acc[8],#0 @ r[8]-=0
622 umlal @acc[1],$t0,$t1,$bj @ "r[0]"+=a[0]*b[i]
624 sbc @acc[0],$t3,#0 @ overflow bit, keep in mind
625 @ that netto result is
626 @ addition of a value which
627 @ makes underflow impossible
629 ldr $t3,[sp,#8] @ a[2]
630 umlal @acc[2],$t1,$t2,$bj @ "r[1]"+=a[1]*b[i]
631 str @acc[0],[sp,#36] @ temporarily offload overflow
633 ldr $t4,[sp,#12] @ a[3], $t4 is alias @acc[0]
634 umlal @acc[3],$t2,$t3,$bj @ "r[2]"+=a[2]*b[i]
636 adds @acc[2],@acc[2],$t0 @ accumulate high part of mult
637 ldr $t0,[sp,#16] @ a[4]
638 umlal @acc[4],$t3,$t4,$bj @ "r[3]"+=a[3]*b[i]
640 adcs @acc[3],@acc[3],$t1
641 ldr $t1,[sp,#20] @ a[5]
642 umlal @acc[5],$t4,$t0,$bj @ "r[4]"+=a[4]*b[i]
644 adcs @acc[4],@acc[4],$t2
645 ldr $t2,[sp,#24] @ a[6]
646 umlal @acc[6],$t0,$t1,$bj @ "r[5]"+=a[5]*b[i]
648 adcs @acc[5],@acc[5],$t3
649 ldr $t3,[sp,#28] @ a[7]
650 umlal @acc[7],$t1,$t2,$bj @ "r[6]"+=a[6]*b[i]
652 adcs @acc[6],@acc[6],$t4
653 ldr @acc[0],[sp,#36] @ restore overflow bit
654 umlal @acc[8],$t2,$t3,$bj @ "r[7]"+=a[7]*b[i]
656 adcs @acc[7],@acc[7],$t0
657 adcs @acc[8],@acc[8],$t1
658 adcs @acc[0],$acc[0],$t2
659 adc $t3,$t3,#0 @ new overflow bit
661 push(@acc,shift(@acc)); # rotate registers, so that
662 # "r[i]" becomes r[i]
665 @ last multiplication-less reduction
666 adds @acc[3],@acc[3],@acc[0]
667 ldr $r_ptr,[sp,#32] @ restore r_ptr
668 adcs @acc[4],@acc[4],#0
669 adcs @acc[5],@acc[5],#0
670 adcs @acc[6],@acc[6],@acc[0]
671 adcs @acc[7],@acc[7],#0
672 adcs @acc[8],@acc[8],@acc[0]
674 subs @acc[7],@acc[7],@acc[0]
675 sbcs @acc[8],@acc[8],#0
676 sbc @acc[0],$t3,#0 @ overflow bit
678 @ Final step is "if result > mod, subtract mod", but we do it
679 @ "other way around", namely subtract modulus from result
680 @ and if it borrowed, add modulus back.
682 adds @acc[1],@acc[1],#1 @ subs @acc[1],@acc[1],#-1
683 adcs @acc[2],@acc[2],#0 @ sbcs @acc[2],@acc[2],#-1
684 adcs @acc[3],@acc[3],#0 @ sbcs @acc[3],@acc[3],#-1
685 sbcs @acc[4],@acc[4],#0
686 sbcs @acc[5],@acc[5],#0
687 sbcs @acc[6],@acc[6],#0
688 sbcs @acc[7],@acc[7],#1
689 adcs @acc[8],@acc[8],#0 @ sbcs @acc[8],@acc[8],#-1
690 ldr lr,[sp,#44] @ restore lr
691 sbc @acc[0],@acc[0],#0 @ broadcast borrow bit
694 @ Note that because mod has special form, i.e. consists of
695 @ 0xffffffff, 1 and 0s, we can conditionally synthesize it by
696 @ broadcasting borrow bit to a register, @acc[0], and using it as
697 @ a whole or extracting single bit.
699 adds @acc[1],@acc[1],@acc[0] @ add modulus or zero
700 adcs @acc[2],@acc[2],@acc[0]
701 str @acc[1],[$r_ptr,#0]
702 adcs @acc[3],@acc[3],@acc[0]
703 str @acc[2],[$r_ptr,#4]
704 adcs @acc[4],@acc[4],#0
705 str @acc[3],[$r_ptr,#8]
706 adcs @acc[5],@acc[5],#0
707 str @acc[4],[$r_ptr,#12]
708 adcs @acc[6],@acc[6],#0
709 str @acc[5],[$r_ptr,#16]
710 adcs @acc[7],@acc[7],@acc[0],lsr#31
711 str @acc[6],[$r_ptr,#20]
712 adc @acc[8],@acc[8],@acc[0]
713 str @acc[7],[$r_ptr,#24]
714 str @acc[8],[$r_ptr,#28]
717 .size __ecp_nistz256_mul_mont,.-__ecp_nistz256_mul_mont
722 my ($out,$inp,$index,$mask)=map("r$_",(0..3));
724 @ void ecp_nistz256_scatter_w5(void *r0,const P256_POINT *r1,
726 .globl ecp_nistz256_scatter_w5
727 .type ecp_nistz256_scatter_w5,%function
729 ecp_nistz256_scatter_w5:
732 add $out,$out,$index,lsl#2
734 ldmia $inp!,{r4-r11} @ X
735 str r4,[$out,#64*0-4]
736 str r5,[$out,#64*1-4]
737 str r6,[$out,#64*2-4]
738 str r7,[$out,#64*3-4]
739 str r8,[$out,#64*4-4]
740 str r9,[$out,#64*5-4]
741 str r10,[$out,#64*6-4]
742 str r11,[$out,#64*7-4]
745 ldmia $inp!,{r4-r11} @ Y
746 str r4,[$out,#64*0-4]
747 str r5,[$out,#64*1-4]
748 str r6,[$out,#64*2-4]
749 str r7,[$out,#64*3-4]
750 str r8,[$out,#64*4-4]
751 str r9,[$out,#64*5-4]
752 str r10,[$out,#64*6-4]
753 str r11,[$out,#64*7-4]
756 ldmia $inp,{r4-r11} @ Z
757 str r4,[$out,#64*0-4]
758 str r5,[$out,#64*1-4]
759 str r6,[$out,#64*2-4]
760 str r7,[$out,#64*3-4]
761 str r8,[$out,#64*4-4]
762 str r9,[$out,#64*5-4]
763 str r10,[$out,#64*6-4]
764 str r11,[$out,#64*7-4]
767 #if __ARM_ARCH__>=5 || defined(__thumb__)
772 .size ecp_nistz256_scatter_w5,.-ecp_nistz256_scatter_w5
774 @ void ecp_nistz256_gather_w5(P256_POINT *r0,const void *r1,
776 .globl ecp_nistz256_gather_w5
777 .type ecp_nistz256_gather_w5,%function
779 ecp_nistz256_gather_w5:
784 subne $index,$index,#1
786 add $inp,$inp,$index,lsl#2
805 stmia $out!,{r4-r11} @ X
824 stmia $out!,{r4-r11} @ Y
842 stmia $out,{r4-r11} @ Z
845 #if __ARM_ARCH__>=5 || defined(__thumb__)
850 .size ecp_nistz256_gather_w5,.-ecp_nistz256_gather_w5
852 @ void ecp_nistz256_scatter_w7(void *r0,const P256_POINT_AFFINE *r1,
854 .globl ecp_nistz256_scatter_w7
855 .type ecp_nistz256_scatter_w7,%function
857 ecp_nistz256_scatter_w7:
862 subs $index,$index,#1
863 strb $mask,[$out,#64*0-1]
864 mov $mask,$mask,lsr#8
865 strb $mask,[$out,#64*1-1]
866 mov $mask,$mask,lsr#8
867 strb $mask,[$out,#64*2-1]
868 mov $mask,$mask,lsr#8
869 strb $mask,[$out,#64*3-1]
873 #if __ARM_ARCH__>=5 || defined(__thumb__)
878 .size ecp_nistz256_scatter_w7,.-ecp_nistz256_scatter_w7
880 @ void ecp_nistz256_gather_w7(P256_POINT_AFFINE *r0,const void *r1,
882 .globl ecp_nistz256_gather_w7
883 .type ecp_nistz256_gather_w7,%function
885 ecp_nistz256_gather_w7:
890 subne $index,$index,#1
897 subs $index,$index,#1
910 #if __ARM_ARCH__>=5 || defined(__thumb__)
915 .size ecp_nistz256_gather_w7,.-ecp_nistz256_gather_w7
919 # In comparison to integer-only equivalent of below subroutine:
925 # As not all time is spent in multiplication, overall impact is deemed
926 # too low to care about.
928 my ($A0,$A1,$A2,$A3,$Bi,$zero,$temp)=map("d$_",(0..7));
931 my @AxB=map("q$_",(8..15));
933 my ($rptr,$aptr,$bptr,$toutptr)=map("r$_",(0..3));
939 .globl ecp_nistz256_mul_mont_neon
940 .type ecp_nistz256_mul_mont_neon,%function
942 ecp_nistz256_mul_mont_neon:
945 vstmdb sp!,{q4-q5} @ ABI specification says so
948 vld1.32 {${Bi}[0]},[$bptr,:32]!
949 veor $zero,$zero,$zero
950 vld1.32 {$A0-$A3}, [$aptr] @ can't specify :32 :-(
952 mov sp,$toutptr @ alloca
953 vmov.i64 $mask,#0xffff
955 vmull.u32 @AxB[0],$Bi,${A0}[0]
956 vmull.u32 @AxB[1],$Bi,${A0}[1]
957 vmull.u32 @AxB[2],$Bi,${A1}[0]
958 vmull.u32 @AxB[3],$Bi,${A1}[1]
959 vshr.u64 $temp,@AxB[0]#lo,#16
960 vmull.u32 @AxB[4],$Bi,${A2}[0]
961 vadd.u64 @AxB[0]#hi,@AxB[0]#hi,$temp
962 vmull.u32 @AxB[5],$Bi,${A2}[1]
963 vshr.u64 $temp,@AxB[0]#hi,#16 @ upper 32 bits of a[0]*b[0]
964 vmull.u32 @AxB[6],$Bi,${A3}[0]
965 vand.u64 @AxB[0],@AxB[0],$mask @ lower 32 bits of a[0]*b[0]
966 vmull.u32 @AxB[7],$Bi,${A3}[1]
968 for($i=1;$i<8;$i++) {
970 vld1.32 {${Bi}[0]},[$bptr,:32]!
971 veor $zero,$zero,$zero
972 vadd.u64 @AxB[1]#lo,@AxB[1]#lo,$temp @ reduction
973 vshl.u64 $mult,@AxB[0],#32
974 vadd.u64 @AxB[3],@AxB[3],@AxB[0]
975 vsub.u64 $mult,$mult,@AxB[0]
977 vadd.u64 @AxB[6],@AxB[6],@AxB[0]
978 vadd.u64 @AxB[7],@AxB[7],$mult
980 push(@AxB,shift(@AxB));
982 vmlal.u32 @AxB[0],$Bi,${A0}[0]
983 vmlal.u32 @AxB[1],$Bi,${A0}[1]
984 vmlal.u32 @AxB[2],$Bi,${A1}[0]
985 vmlal.u32 @AxB[3],$Bi,${A1}[1]
986 vshr.u64 $temp,@AxB[0]#lo,#16
987 vmlal.u32 @AxB[4],$Bi,${A2}[0]
988 vadd.u64 @AxB[0]#hi,@AxB[0]#hi,$temp
989 vmlal.u32 @AxB[5],$Bi,${A2}[1]
990 vshr.u64 $temp,@AxB[0]#hi,#16 @ upper 33 bits of a[0]*b[i]+t[0]
991 vmlal.u32 @AxB[6],$Bi,${A3}[0]
992 vand.u64 @AxB[0],@AxB[0],$mask @ lower 32 bits of a[0]*b[0]
993 vmull.u32 @AxB[7],$Bi,${A3}[1]
997 vadd.u64 @AxB[1]#lo,@AxB[1]#lo,$temp @ last reduction
998 vshl.u64 $mult,@AxB[0],#32
999 vadd.u64 @AxB[3],@AxB[3],@AxB[0]
1000 vsub.u64 $mult,$mult,@AxB[0]
1001 vadd.u64 @AxB[6],@AxB[6],@AxB[0]
1002 vadd.u64 @AxB[7],@AxB[7],$mult
1004 vshr.u64 $temp,@AxB[1]#lo,#16 @ convert
1005 vadd.u64 @AxB[1]#hi,@AxB[1]#hi,$temp
1006 vshr.u64 $temp,@AxB[1]#hi,#16
1007 vzip.16 @AxB[1]#lo,@AxB[1]#hi
1011 vadd.u64 @AxB[$_]#lo,@AxB[$_]#lo,$temp
1012 vst1.32 {@AxB[$_-1]#lo[0]},[$toutptr,:32]!
1013 vshr.u64 $temp,@AxB[$_]#lo,#16
1014 vadd.u64 @AxB[$_]#hi,@AxB[$_]#hi,$temp
1015 vshr.u64 $temp,@AxB[$_]#hi,#16
1016 vzip.16 @AxB[$_]#lo,@AxB[$_]#hi
1020 vst1.32 {@AxB[7]#lo[0]},[$toutptr,:32]!
1021 vst1.32 {$temp},[$toutptr] @ upper 33 bits
1037 ldr r9,[sp,#32] @ top-most bit
1055 adcs r7,r7,r9,lsr#31
1063 .size ecp_nistz256_mul_mont_neon,.-ecp_nistz256_mul_mont_neon
1069 ########################################################################
1070 # Below $aN assignment matches order in which 256-bit result appears in
1071 # register bank at return from __ecp_nistz256_mul_mont, so that we can
1072 # skip over reloading it from memory. This means that below functions
1073 # use custom calling sequence accepting 256-bit input in registers,
1074 # output pointer in r0, $r_ptr, and optional pointer in r2, $b_ptr.
1076 # See their "normal" counterparts for insights on calculations.
1078 my ($a0,$a1,$a2,$a3,$a4,$a5,$a6,$a7,
1079 $t0,$t1,$t2,$t3)=map("r$_",(11,3..10,12,14,1));
1083 .type __ecp_nistz256_sub_from,%function
1085 __ecp_nistz256_sub_from:
1086 str lr,[sp,#-4]! @ push lr
1091 ldr $t3,[$b_ptr,#12]
1093 ldr $t0,[$b_ptr,#16]
1095 ldr $t1,[$b_ptr,#20]
1097 ldr $t2,[$b_ptr,#24]
1099 ldr $t3,[$b_ptr,#28]
1104 sbc $ff,$ff,$ff @ broadcast borrow bit
1105 ldr lr,[sp],#4 @ pop lr
1107 adds $a0,$a0,$ff @ add synthesized modulus
1115 str $a3,[$r_ptr,#12]
1117 str $a4,[$r_ptr,#16]
1118 adcs $a6,$a6,$ff,lsr#31
1119 str $a5,[$r_ptr,#20]
1121 str $a6,[$r_ptr,#24]
1122 str $a7,[$r_ptr,#28]
1125 .size __ecp_nistz256_sub_from,.-__ecp_nistz256_sub_from
1127 .type __ecp_nistz256_sub_morf,%function
1129 __ecp_nistz256_sub_morf:
1130 str lr,[sp,#-4]! @ push lr
1135 ldr $t3,[$b_ptr,#12]
1137 ldr $t0,[$b_ptr,#16]
1139 ldr $t1,[$b_ptr,#20]
1141 ldr $t2,[$b_ptr,#24]
1143 ldr $t3,[$b_ptr,#28]
1148 sbc $ff,$ff,$ff @ broadcast borrow bit
1149 ldr lr,[sp],#4 @ pop lr
1151 adds $a0,$a0,$ff @ add synthesized modulus
1159 str $a3,[$r_ptr,#12]
1161 str $a4,[$r_ptr,#16]
1162 adcs $a6,$a6,$ff,lsr#31
1163 str $a5,[$r_ptr,#20]
1165 str $a6,[$r_ptr,#24]
1166 str $a7,[$r_ptr,#28]
1169 .size __ecp_nistz256_sub_morf,.-__ecp_nistz256_sub_morf
1171 .type __ecp_nistz256_add_self,%function
1173 __ecp_nistz256_add_self:
1174 adds $a0,$a0,$a0 @ a[0:7]+=a[0:7]
1183 movcs $ff,#-1 @ $ff = carry ? -1 : 0
1185 subs $a0,$a0,$ff @ subtract synthesized modulus
1193 str $a3,[$r_ptr,#12]
1195 str $a4,[$r_ptr,#16]
1196 sbcs $a6,$a6,$ff,lsr#31
1197 str $a5,[$r_ptr,#20]
1199 str $a6,[$r_ptr,#24]
1200 str $a7,[$r_ptr,#28]
1203 .size __ecp_nistz256_add_self,.-__ecp_nistz256_add_self
1207 ########################################################################
1208 # following subroutines are "literal" implemetation of those found in
1211 ########################################################################
1212 # void ecp_nistz256_point_double(P256_POINT *out,const P256_POINT *inp);
1215 my ($S,$M,$Zsqr,$in_x,$tmp0)=map(32*$_,(0..4));
1216 # above map() describes stack layout with 5 temporary
1217 # 256-bit vectors on top. Then note that we push
1218 # starting from r0, which means that we have copy of
1219 # input arguments just below these temporary vectors.
1222 .globl ecp_nistz256_point_double
1223 .type ecp_nistz256_point_double,%function
1225 ecp_nistz256_point_double:
1226 stmdb sp!,{r0-r12,lr} @ push from r0, unusual, but intentional
1230 ldmia $a_ptr!,{r4-r11} @ copy in_x
1234 bl __ecp_nistz256_mul_by_2 @ p256_mul_by_2(S, in_y);
1236 add $b_ptr,$a_ptr,#32
1237 add $a_ptr,$a_ptr,#32
1238 add $r_ptr,sp,#$Zsqr
1239 bl __ecp_nistz256_mul_mont @ p256_sqr_mont(Zsqr, in_z);
1244 bl __ecp_nistz256_mul_mont @ p256_sqr_mont(S, S);
1246 ldr $b_ptr,[sp,#32*5+4]
1247 add $a_ptr,$b_ptr,#32
1248 add $b_ptr,$b_ptr,#64
1249 add $r_ptr,sp,#$tmp0
1250 bl __ecp_nistz256_mul_mont @ p256_mul_mont(tmp0, in_z, in_y);
1252 ldr $r_ptr,[sp,#32*5]
1253 add $r_ptr,$r_ptr,#64
1254 bl __ecp_nistz256_add_self @ p256_mul_by_2(res_z, tmp0);
1256 add $a_ptr,sp,#$in_x
1257 add $b_ptr,sp,#$Zsqr
1259 bl __ecp_nistz256_add @ p256_add(M, in_x, Zsqr);
1261 add $a_ptr,sp,#$in_x
1262 add $b_ptr,sp,#$Zsqr
1263 add $r_ptr,sp,#$Zsqr
1264 bl __ecp_nistz256_sub @ p256_sub(Zsqr, in_x, Zsqr);
1268 add $r_ptr,sp,#$tmp0
1269 bl __ecp_nistz256_mul_mont @ p256_sqr_mont(tmp0, S);
1271 add $a_ptr,sp,#$Zsqr
1274 bl __ecp_nistz256_mul_mont @ p256_mul_mont(M, M, Zsqr);
1276 ldr $r_ptr,[sp,#32*5]
1277 add $a_ptr,sp,#$tmp0
1278 add $r_ptr,$r_ptr,#32
1279 bl __ecp_nistz256_div_by_2 @ p256_div_by_2(res_y, tmp0);
1283 bl __ecp_nistz256_mul_by_3 @ p256_mul_by_3(M, M);
1285 add $a_ptr,sp,#$in_x
1288 bl __ecp_nistz256_mul_mont @ p256_mul_mont(S, S, in_x);
1290 add $r_ptr,sp,#$tmp0
1291 bl __ecp_nistz256_add_self @ p256_mul_by_2(tmp0, S);
1293 ldr $r_ptr,[sp,#32*5]
1296 bl __ecp_nistz256_mul_mont @ p256_sqr_mont(res_x, M);
1298 add $b_ptr,sp,#$tmp0
1299 bl __ecp_nistz256_sub_from @ p256_sub(res_x, res_x, tmp0);
1303 bl __ecp_nistz256_sub_morf @ p256_sub(S, S, res_x);
1307 bl __ecp_nistz256_mul_mont @ p256_mul_mont(S, S, M);
1309 ldr $r_ptr,[sp,#32*5]
1310 add $b_ptr,$r_ptr,#32
1311 add $r_ptr,$r_ptr,#32
1312 bl __ecp_nistz256_sub_from @ p256_sub(res_y, S, res_y);
1314 add sp,sp,#32*5+16 @ +16 means "skip even over saved r0-r3"
1315 #if __ARM_ARCH__>=5 || !defined(__thumb__)
1316 ldmia sp!,{r4-r12,pc}
1318 ldmia sp!,{r4-r12,lr}
1319 bx lr @ interoperable with Thumb ISA:-)
1321 .size ecp_nistz256_point_double,.-ecp_nistz256_point_double
1325 ########################################################################
1326 # void ecp_nistz256_point_add(P256_POINT *out,const P256_POINT *in1,
1327 # const P256_POINT *in2);
1329 my ($res_x,$res_y,$res_z,
1330 $in1_x,$in1_y,$in1_z,
1331 $in2_x,$in2_y,$in2_z,
1332 $H,$Hsqr,$R,$Rsqr,$Hcub,
1333 $U1,$U2,$S1,$S2)=map(32*$_,(0..17));
1334 my ($Z1sqr, $Z2sqr) = ($Hsqr, $Rsqr);
1335 # above map() describes stack layout with 18 temporary
1336 # 256-bit vectors on top. Then note that we push
1337 # starting from r0, which means that we have copy of
1338 # input arguments just below these temporary vectors.
1339 # We use three of them for !in1infty, !in2intfy and
1340 # result of check for zero.
1343 .globl ecp_nistz256_point_add
1344 .type ecp_nistz256_point_add,%function
1346 ecp_nistz256_point_add:
1347 stmdb sp!,{r0-r12,lr} @ push from r0, unusual, but intentional
1350 ldmia $b_ptr!,{r4-r11} @ copy in2
1360 ldmia $b_ptr!,{r4-r11}
1370 ldmia $b_ptr,{r4-r11}
1374 str r12,[sp,#32*18+8] @ !in2infty
1376 ldmia $a_ptr!,{r4-r11} @ copy in1
1386 ldmia $a_ptr!,{r4-r11}
1396 ldmia $a_ptr,{r4-r11}
1400 str r12,[sp,#32*18+4] @ !in1infty
1402 add $a_ptr,sp,#$in2_z
1403 add $b_ptr,sp,#$in2_z
1404 add $r_ptr,sp,#$Z2sqr
1405 bl __ecp_nistz256_mul_mont @ p256_sqr_mont(Z2sqr, in2_z);
1407 add $a_ptr,sp,#$in1_z
1408 add $b_ptr,sp,#$in1_z
1409 add $r_ptr,sp,#$Z1sqr
1410 bl __ecp_nistz256_mul_mont @ p256_sqr_mont(Z1sqr, in1_z);
1412 add $a_ptr,sp,#$in2_z
1413 add $b_ptr,sp,#$Z2sqr
1415 bl __ecp_nistz256_mul_mont @ p256_mul_mont(S1, Z2sqr, in2_z);
1417 add $a_ptr,sp,#$in1_z
1418 add $b_ptr,sp,#$Z1sqr
1420 bl __ecp_nistz256_mul_mont @ p256_mul_mont(S2, Z1sqr, in1_z);
1422 add $a_ptr,sp,#$in1_y
1425 bl __ecp_nistz256_mul_mont @ p256_mul_mont(S1, S1, in1_y);
1427 add $a_ptr,sp,#$in2_y
1430 bl __ecp_nistz256_mul_mont @ p256_mul_mont(S2, S2, in2_y);
1434 bl __ecp_nistz256_sub_from @ p256_sub(R, S2, S1);
1436 orr $a0,$a0,$a1 @ see if result is zero
1442 add $a_ptr,sp,#$in1_x
1444 add $b_ptr,sp,#$Z2sqr
1445 str $a0,[sp,#32*18+12]
1448 bl __ecp_nistz256_mul_mont @ p256_mul_mont(U1, in1_x, Z2sqr);
1450 add $a_ptr,sp,#$in2_x
1451 add $b_ptr,sp,#$Z1sqr
1453 bl __ecp_nistz256_mul_mont @ p256_mul_mont(U2, in2_x, Z1sqr);
1457 bl __ecp_nistz256_sub_from @ p256_sub(H, U2, U1);
1459 orr $a0,$a0,$a1 @ see if result is zero
1467 bne .Ladd_proceed @ is_equal(U1,U2)?
1469 ldr $t0,[sp,#32*18+4]
1470 ldr $t1,[sp,#32*18+8]
1471 ldr $t2,[sp,#32*18+12]
1473 beq .Ladd_proceed @ (in1infty || in2infty)?
1475 beq .Ladd_proceed @ is_equal(S1,S2)?
1477 ldr $r_ptr,[sp,#32*18]
1486 stmia $r_ptr!,{r4-r11}
1487 stmia $r_ptr!,{r4-r11}
1488 stmia $r_ptr!,{r4-r11}
1495 add $r_ptr,sp,#$Rsqr
1496 bl __ecp_nistz256_mul_mont @ p256_sqr_mont(Rsqr, R);
1499 add $b_ptr,sp,#$in1_z
1500 add $r_ptr,sp,#$res_z
1501 bl __ecp_nistz256_mul_mont @ p256_mul_mont(res_z, H, in1_z);
1505 add $r_ptr,sp,#$Hsqr
1506 bl __ecp_nistz256_mul_mont @ p256_sqr_mont(Hsqr, H);
1508 add $a_ptr,sp,#$in2_z
1509 add $b_ptr,sp,#$res_z
1510 add $r_ptr,sp,#$res_z
1511 bl __ecp_nistz256_mul_mont @ p256_mul_mont(res_z, res_z, in2_z);
1514 add $b_ptr,sp,#$Hsqr
1515 add $r_ptr,sp,#$Hcub
1516 bl __ecp_nistz256_mul_mont @ p256_mul_mont(Hcub, Hsqr, H);
1518 add $a_ptr,sp,#$Hsqr
1521 bl __ecp_nistz256_mul_mont @ p256_mul_mont(U2, U1, Hsqr);
1523 add $r_ptr,sp,#$Hsqr
1524 bl __ecp_nistz256_add_self @ p256_mul_by_2(Hsqr, U2);
1526 add $b_ptr,sp,#$Rsqr
1527 add $r_ptr,sp,#$res_x
1528 bl __ecp_nistz256_sub_morf @ p256_sub(res_x, Rsqr, Hsqr);
1530 add $b_ptr,sp,#$Hcub
1531 bl __ecp_nistz256_sub_from @ p256_sub(res_x, res_x, Hcub);
1534 add $r_ptr,sp,#$res_y
1535 bl __ecp_nistz256_sub_morf @ p256_sub(res_y, U2, res_x);
1537 add $a_ptr,sp,#$Hcub
1540 bl __ecp_nistz256_mul_mont @ p256_mul_mont(S2, S1, Hcub);
1543 add $b_ptr,sp,#$res_y
1544 add $r_ptr,sp,#$res_y
1545 bl __ecp_nistz256_mul_mont @ p256_mul_mont(res_y, res_y, R);
1548 bl __ecp_nistz256_sub_from @ p256_sub(res_y, res_y, S2);
1550 ldr r11,[sp,#32*18+4] @ !in1intfy
1551 ldr r12,[sp,#32*18+8] @ !in2intfy
1559 ldr $r_ptr,[sp,#32*18]
1561 for($i=0;$i<96;$i+=8) { # conditional moves
1563 ldmia r1!,{r4-r5} @ res_x
1564 ldmia r2!,{r6-r7} @ in2_x
1565 ldmia r3!,{r8-r9} @ in1_x
1576 stmia $r_ptr!,{r4-r5}
1581 add sp,sp,#32*18+16 @ +16 means "skip even over saved r0-r3"
1582 #if __ARM_ARCH__>=5 || defined(__thumb__)
1583 ldmia sp!,{r4-r12,pc}
1585 ldmia sp!,{r4-r12,lr}
1586 bx lr @ interoperable with Thumb ISA:-)
1588 .size ecp_nistz256_point_add,.-ecp_nistz256_point_add
1592 ########################################################################
1593 # void ecp_nistz256_point_add_affine(P256_POINT *out,const P256_POINT *in1,
1594 # const P256_POINT_AFFINE *in2);
1596 my ($res_x,$res_y,$res_z,
1597 $in1_x,$in1_y,$in1_z,
1599 $U2,$S2,$H,$R,$Hsqr,$Hcub,$Rsqr)=map(32*$_,(0..14));
1601 # above map() describes stack layout with 18 temporary
1602 # 256-bit vectors on top. Then note that we push
1603 # starting from r0, which means that we have copy of
1604 # input arguments just below these temporary vectors.
1605 # We use two of them for !in1infty, !in2intfy.
1607 my @ONE_mont=(1,0,0,-1,-1,-1,-2,0);
1610 .globl ecp_nistz256_point_add_affine
1611 .type ecp_nistz256_point_add_affine,%function
1613 ecp_nistz256_point_add_affine:
1614 stmdb sp!,{r0-r12,lr} @ push from r0, unusual, but intentional
1617 ldmia $a_ptr!,{r4-r11} @ copy in1
1627 ldmia $a_ptr!,{r4-r11}
1637 ldmia $a_ptr,{r4-r11}
1641 str r12,[sp,#32*15+4] @ !in1infty
1643 ldmia $b_ptr!,{r4-r11} @ copy in2
1653 ldmia $b_ptr!,{r4-r11}
1665 str r12,[sp,#32*15+8] @ !in2infty
1667 add $a_ptr,sp,#$in1_z
1668 add $b_ptr,sp,#$in1_z
1669 add $r_ptr,sp,#$Z1sqr
1670 bl __ecp_nistz256_mul_mont @ p256_sqr_mont(Z1sqr, in1_z);
1672 add $a_ptr,sp,#$Z1sqr
1673 add $b_ptr,sp,#$in2_x
1675 bl __ecp_nistz256_mul_mont @ p256_mul_mont(U2, Z1sqr, in2_x);
1677 add $b_ptr,sp,#$in1_x
1679 bl __ecp_nistz256_sub_from @ p256_sub(H, U2, in1_x);
1681 add $a_ptr,sp,#$Z1sqr
1682 add $b_ptr,sp,#$in1_z
1684 bl __ecp_nistz256_mul_mont @ p256_mul_mont(S2, Z1sqr, in1_z);
1687 add $b_ptr,sp,#$in1_z
1688 add $r_ptr,sp,#$res_z
1689 bl __ecp_nistz256_mul_mont @ p256_mul_mont(res_z, H, in1_z);
1691 add $a_ptr,sp,#$in2_y
1694 bl __ecp_nistz256_mul_mont @ p256_mul_mont(S2, S2, in2_y);
1696 add $b_ptr,sp,#$in1_y
1698 bl __ecp_nistz256_sub_from @ p256_sub(R, S2, in1_y);
1702 add $r_ptr,sp,#$Hsqr
1703 bl __ecp_nistz256_mul_mont @ p256_sqr_mont(Hsqr, H);
1707 add $r_ptr,sp,#$Rsqr
1708 bl __ecp_nistz256_mul_mont @ p256_sqr_mont(Rsqr, R);
1711 add $b_ptr,sp,#$Hsqr
1712 add $r_ptr,sp,#$Hcub
1713 bl __ecp_nistz256_mul_mont @ p256_mul_mont(Hcub, Hsqr, H);
1715 add $a_ptr,sp,#$Hsqr
1716 add $b_ptr,sp,#$in1_x
1718 bl __ecp_nistz256_mul_mont @ p256_mul_mont(U2, in1_x, Hsqr);
1720 add $r_ptr,sp,#$Hsqr
1721 bl __ecp_nistz256_add_self @ p256_mul_by_2(Hsqr, U2);
1723 add $b_ptr,sp,#$Rsqr
1724 add $r_ptr,sp,#$res_x
1725 bl __ecp_nistz256_sub_morf @ p256_sub(res_x, Rsqr, Hsqr);
1727 add $b_ptr,sp,#$Hcub
1728 bl __ecp_nistz256_sub_from @ p256_sub(res_x, res_x, Hcub);
1731 add $r_ptr,sp,#$res_y
1732 bl __ecp_nistz256_sub_morf @ p256_sub(res_y, U2, res_x);
1734 add $a_ptr,sp,#$Hcub
1735 add $b_ptr,sp,#$in1_y
1737 bl __ecp_nistz256_mul_mont @ p256_mul_mont(S2, in1_y, Hcub);
1740 add $b_ptr,sp,#$res_y
1741 add $r_ptr,sp,#$res_y
1742 bl __ecp_nistz256_mul_mont @ p256_mul_mont(res_y, res_y, R);
1745 bl __ecp_nistz256_sub_from @ p256_sub(res_y, res_y, S2);
1747 ldr r11,[sp,#32*15+4] @ !in1intfy
1748 ldr r12,[sp,#32*15+8] @ !in2intfy
1756 ldr $r_ptr,[sp,#32*15]
1758 for($i=0;$i<64;$i+=8) { # conditional moves
1760 ldmia r1!,{r4-r5} @ res_x
1761 ldmia r2!,{r6-r7} @ in2_x
1762 ldmia r3!,{r8-r9} @ in1_x
1773 stmia $r_ptr!,{r4-r5}
1779 ldmia r1!,{r4-r5} @ res_z
1780 ldmia r3!,{r8-r9} @ in1_z
1783 and r6,r11,#@ONE_mont[$j]
1784 and r7,r11,#@ONE_mont[$j+1]
1791 stmia $r_ptr!,{r4-r5}
1795 add sp,sp,#32*15+16 @ +16 means "skip even over saved r0-r3"
1796 #if __ARM_ARCH__>=5 || !defined(__thumb__)
1797 ldmia sp!,{r4-r12,pc}
1799 ldmia sp!,{r4-r12,lr}
1800 bx lr @ interoperable with Thumb ISA:-)
1802 .size ecp_nistz256_point_add_affine,.-ecp_nistz256_point_add_affine
1806 foreach (split("\n",$code)) {
1807 s/\`([^\`]*)\`/eval $1/geo;
1809 s/\bq([0-9]+)#(lo|hi)/sprintf "d%d",2*$1+($2 eq "hi")/geo;
1813 close STDOUT; # enforce flush