1 /* crypto/bn/bn_asm.c */
2 /* Copyright (C) 1995-1998 Eric Young (eay@cryptsoft.com)
5 * This package is an SSL implementation written
6 * by Eric Young (eay@cryptsoft.com).
7 * The implementation was written so as to conform with Netscapes SSL.
9 * This library is free for commercial and non-commercial use as long as
10 * the following conditions are aheared to. The following conditions
11 * apply to all code found in this distribution, be it the RC4, RSA,
12 * lhash, DES, etc., code; not just the SSL code. The SSL documentation
13 * included with this distribution is covered by the same copyright terms
14 * except that the holder is Tim Hudson (tjh@cryptsoft.com).
16 * Copyright remains Eric Young's, and as such any Copyright notices in
17 * the code are not to be removed.
18 * If this package is used in a product, Eric Young should be given attribution
19 * as the author of the parts of the library used.
20 * This can be in the form of a textual message at program startup or
21 * in documentation (online or textual) provided with the package.
23 * Redistribution and use in source and binary forms, with or without
24 * modification, are permitted provided that the following conditions
26 * 1. Redistributions of source code must retain the copyright
27 * notice, this list of conditions and the following disclaimer.
28 * 2. Redistributions in binary form must reproduce the above copyright
29 * notice, this list of conditions and the following disclaimer in the
30 * documentation and/or other materials provided with the distribution.
31 * 3. All advertising materials mentioning features or use of this software
32 * must display the following acknowledgement:
33 * "This product includes cryptographic software written by
34 * Eric Young (eay@cryptsoft.com)"
35 * The word 'cryptographic' can be left out if the rouines from the library
36 * being used are not cryptographic related :-).
37 * 4. If you include any Windows specific code (or a derivative thereof) from
38 * the apps directory (application code) you must include an acknowledgement:
39 * "This product includes software written by Tim Hudson (tjh@cryptsoft.com)"
41 * THIS SOFTWARE IS PROVIDED BY ERIC YOUNG ``AS IS'' AND
42 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
43 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
44 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
45 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
46 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
47 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
48 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
49 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
50 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
53 * The licence and distribution terms for any publically available version or
54 * derivative of this code cannot be changed. i.e. this code cannot simply be
55 * copied and put under another distribution licence
56 * [including the GNU Public Licence.]
60 # undef NDEBUG /* avoid conflicting definitions */
69 #if defined(BN_LLONG) || defined(BN_UMULT_HIGH)
71 BN_ULONG bn_mul_add_words(BN_ULONG *rp, const BN_ULONG *ap, int num,
80 # ifndef OPENSSL_SMALL_FOOTPRINT
82 mul_add(rp[0], ap[0], w, c1);
83 mul_add(rp[1], ap[1], w, c1);
84 mul_add(rp[2], ap[2], w, c1);
85 mul_add(rp[3], ap[3], w, c1);
92 mul_add(rp[0], ap[0], w, c1);
101 BN_ULONG bn_mul_words(BN_ULONG *rp, const BN_ULONG *ap, int num, BN_ULONG w)
109 # ifndef OPENSSL_SMALL_FOOTPRINT
111 mul(rp[0], ap[0], w, c1);
112 mul(rp[1], ap[1], w, c1);
113 mul(rp[2], ap[2], w, c1);
114 mul(rp[3], ap[3], w, c1);
121 mul(rp[0], ap[0], w, c1);
129 void bn_sqr_words(BN_ULONG *r, const BN_ULONG *a, int n)
135 # ifndef OPENSSL_SMALL_FOOTPRINT
137 sqr(r[0], r[1], a[0]);
138 sqr(r[2], r[3], a[1]);
139 sqr(r[4], r[5], a[2]);
140 sqr(r[6], r[7], a[3]);
147 sqr(r[0], r[1], a[0]);
154 #else /* !(defined(BN_LLONG) ||
155 * defined(BN_UMULT_HIGH)) */
157 BN_ULONG bn_mul_add_words(BN_ULONG *rp, const BN_ULONG *ap, int num,
165 return ((BN_ULONG)0);
170 # ifndef OPENSSL_SMALL_FOOTPRINT
172 mul_add(rp[0], ap[0], bl, bh, c);
173 mul_add(rp[1], ap[1], bl, bh, c);
174 mul_add(rp[2], ap[2], bl, bh, c);
175 mul_add(rp[3], ap[3], bl, bh, c);
182 mul_add(rp[0], ap[0], bl, bh, c);
190 BN_ULONG bn_mul_words(BN_ULONG *rp, const BN_ULONG *ap, int num, BN_ULONG w)
197 return ((BN_ULONG)0);
202 # ifndef OPENSSL_SMALL_FOOTPRINT
204 mul(rp[0], ap[0], bl, bh, carry);
205 mul(rp[1], ap[1], bl, bh, carry);
206 mul(rp[2], ap[2], bl, bh, carry);
207 mul(rp[3], ap[3], bl, bh, carry);
214 mul(rp[0], ap[0], bl, bh, carry);
222 void bn_sqr_words(BN_ULONG *r, const BN_ULONG *a, int n)
228 # ifndef OPENSSL_SMALL_FOOTPRINT
230 sqr64(r[0], r[1], a[0]);
231 sqr64(r[2], r[3], a[1]);
232 sqr64(r[4], r[5], a[2]);
233 sqr64(r[6], r[7], a[3]);
240 sqr64(r[0], r[1], a[0]);
247 #endif /* !(defined(BN_LLONG) ||
248 * defined(BN_UMULT_HIGH)) */
250 #if defined(BN_LLONG) && defined(BN_DIV2W)
252 BN_ULONG bn_div_words(BN_ULONG h, BN_ULONG l, BN_ULONG d)
254 return ((BN_ULONG)(((((BN_ULLONG) h) << BN_BITS2) | l) / (BN_ULLONG) d));
259 /* Divide h,l by d and return the result. */
260 /* I need to test this some more :-( */
261 BN_ULONG bn_div_words(BN_ULONG h, BN_ULONG l, BN_ULONG d)
263 BN_ULONG dh, dl, q, ret = 0, th, tl, t;
269 i = BN_num_bits_word(d);
270 assert((i == BN_BITS2) || (h <= (BN_ULONG)1 << i));
278 h = (h << i) | (l >> (BN_BITS2 - i));
281 dh = (d & BN_MASK2h) >> BN_BITS4;
282 dl = (d & BN_MASK2l);
284 if ((h >> BN_BITS4) == dh)
293 if ((t & BN_MASK2h) ||
294 ((tl) <= ((t << BN_BITS4) | ((l & BN_MASK2h) >> BN_BITS4))))
300 t = (tl >> BN_BITS4);
301 tl = (tl << BN_BITS4) & BN_MASK2h;
317 h = ((h << BN_BITS4) | (l >> BN_BITS4)) & BN_MASK2;
318 l = (l & BN_MASK2l) << BN_BITS4;
323 #endif /* !defined(BN_LLONG) && defined(BN_DIV2W) */
326 BN_ULONG bn_add_words(BN_ULONG *r, const BN_ULONG *a, const BN_ULONG *b,
333 return ((BN_ULONG)0);
335 # ifndef OPENSSL_SMALL_FOOTPRINT
337 ll += (BN_ULLONG) a[0] + b[0];
338 r[0] = (BN_ULONG)ll & BN_MASK2;
340 ll += (BN_ULLONG) a[1] + b[1];
341 r[1] = (BN_ULONG)ll & BN_MASK2;
343 ll += (BN_ULLONG) a[2] + b[2];
344 r[2] = (BN_ULONG)ll & BN_MASK2;
346 ll += (BN_ULLONG) a[3] + b[3];
347 r[3] = (BN_ULONG)ll & BN_MASK2;
356 ll += (BN_ULLONG) a[0] + b[0];
357 r[0] = (BN_ULONG)ll & BN_MASK2;
364 return ((BN_ULONG)ll);
366 #else /* !BN_LLONG */
367 BN_ULONG bn_add_words(BN_ULONG *r, const BN_ULONG *a, const BN_ULONG *b,
374 return ((BN_ULONG)0);
377 # ifndef OPENSSL_SMALL_FOOTPRINT
380 t = (t + c) & BN_MASK2;
382 l = (t + b[0]) & BN_MASK2;
386 t = (t + c) & BN_MASK2;
388 l = (t + b[1]) & BN_MASK2;
392 t = (t + c) & BN_MASK2;
394 l = (t + b[2]) & BN_MASK2;
398 t = (t + c) & BN_MASK2;
400 l = (t + b[3]) & BN_MASK2;
411 t = (t + c) & BN_MASK2;
413 l = (t + b[0]) & BN_MASK2;
421 return ((BN_ULONG)c);
423 #endif /* !BN_LLONG */
425 BN_ULONG bn_sub_words(BN_ULONG *r, const BN_ULONG *a, const BN_ULONG *b,
433 return ((BN_ULONG)0);
435 #ifndef OPENSSL_SMALL_FOOTPRINT
439 r[0] = (t1 - t2 - c) & BN_MASK2;
444 r[1] = (t1 - t2 - c) & BN_MASK2;
449 r[2] = (t1 - t2 - c) & BN_MASK2;
454 r[3] = (t1 - t2 - c) & BN_MASK2;
466 r[0] = (t1 - t2 - c) & BN_MASK2;
477 #if defined(BN_MUL_COMBA) && !defined(OPENSSL_SMALL_FOOTPRINT)
479 # undef bn_mul_comba8
480 # undef bn_mul_comba4
481 # undef bn_sqr_comba8
482 # undef bn_sqr_comba4
484 /* mul_add_c(a,b,c0,c1,c2) -- c+=a*b for three word number c=(c2,c1,c0) */
485 /* mul_add_c2(a,b,c0,c1,c2) -- c+=2*a*b for three word number c=(c2,c1,c0) */
486 /* sqr_add_c(a,i,c0,c1,c2) -- c+=a[i]^2 for three word number c=(c2,c1,c0) */
488 * sqr_add_c2(a,i,c0,c1,c2) -- c+=2*a[i]*a[j] for three word number
494 * Keep in mind that additions to multiplication result can not
495 * overflow, because its high half cannot be all-ones.
497 # define mul_add_c(a,b,c0,c1,c2) do { \
499 BN_ULLONG t = (BN_ULLONG)(a)*(b); \
500 t += c0; /* no carry */ \
501 c0 = (BN_ULONG)Lw(t); \
502 hi = (BN_ULONG)Hw(t); \
503 c1 = (c1+hi)&BN_MASK2; if (c1<hi) c2++; \
506 # define mul_add_c2(a,b,c0,c1,c2) do { \
508 BN_ULLONG t = (BN_ULLONG)(a)*(b); \
509 BN_ULLONG tt = t+c0; /* no carry */ \
510 c0 = (BN_ULONG)Lw(tt); \
511 hi = (BN_ULONG)Hw(tt); \
512 c1 = (c1+hi)&BN_MASK2; if (c1<hi) c2++; \
513 t += c0; /* no carry */ \
514 c0 = (BN_ULONG)Lw(t); \
515 hi = (BN_ULONG)Hw(t); \
516 c1 = (c1+hi)&BN_MASK2; if (c1<hi) c2++; \
519 # define sqr_add_c(a,i,c0,c1,c2) do { \
521 BN_ULLONG t = (BN_ULLONG)a[i]*a[i]; \
522 t += c0; /* no carry */ \
523 c0 = (BN_ULONG)Lw(t); \
524 hi = (BN_ULONG)Hw(t); \
525 c1 = (c1+hi)&BN_MASK2; if (c1<hi) c2++; \
528 # define sqr_add_c2(a,i,j,c0,c1,c2) \
529 mul_add_c2((a)[i],(a)[j],c0,c1,c2)
531 # elif defined(BN_UMULT_LOHI)
533 * Keep in mind that additions to hi can not overflow, because
534 * the high word of a multiplication result cannot be all-ones.
536 # define mul_add_c(a,b,c0,c1,c2) do { \
537 BN_ULONG ta = (a), tb = (b); \
539 BN_UMULT_LOHI(lo,hi,ta,tb); \
540 c0 += lo; hi += (c0<lo)?1:0; \
541 c1 += hi; c2 += (c1<hi)?1:0; \
544 # define mul_add_c2(a,b,c0,c1,c2) do { \
545 BN_ULONG ta = (a), tb = (b); \
546 BN_ULONG lo, hi, tt; \
547 BN_UMULT_LOHI(lo,hi,ta,tb); \
548 c0 += lo; tt = hi+((c0<lo)?1:0); \
549 c1 += tt; c2 += (c1<tt)?1:0; \
550 c0 += lo; hi += (c0<lo)?1:0; \
551 c1 += hi; c2 += (c1<hi)?1:0; \
554 # define sqr_add_c(a,i,c0,c1,c2) do { \
555 BN_ULONG ta = (a)[i]; \
557 BN_UMULT_LOHI(lo,hi,ta,ta); \
558 c0 += lo; hi += (c0<lo)?1:0; \
559 c1 += hi; c2 += (c1<hi)?1:0; \
562 # define sqr_add_c2(a,i,j,c0,c1,c2) \
563 mul_add_c2((a)[i],(a)[j],c0,c1,c2)
565 # elif defined(BN_UMULT_HIGH)
567 * Keep in mind that additions to hi can not overflow, because
568 * the high word of a multiplication result cannot be all-ones.
570 # define mul_add_c(a,b,c0,c1,c2) do { \
571 BN_ULONG ta = (a), tb = (b); \
572 BN_ULONG lo = ta * tb; \
573 BN_ULONG hi = BN_UMULT_HIGH(ta,tb); \
574 c0 += lo; hi += (c0<lo)?1:0; \
575 c1 += hi; c2 += (c1<hi)?1:0; \
578 # define mul_add_c2(a,b,c0,c1,c2) do { \
579 BN_ULONG ta = (a), tb = (b), tt; \
580 BN_ULONG lo = ta * tb; \
581 BN_ULONG hi = BN_UMULT_HIGH(ta,tb); \
582 c0 += lo; tt = hi + ((c0<lo)?1:0); \
583 c1 += tt; c2 += (c1<tt)?1:0; \
584 c0 += lo; hi += (c0<lo)?1:0; \
585 c1 += hi; c2 += (c1<hi)?1:0; \
588 # define sqr_add_c(a,i,c0,c1,c2) do { \
589 BN_ULONG ta = (a)[i]; \
590 BN_ULONG lo = ta * ta; \
591 BN_ULONG hi = BN_UMULT_HIGH(ta,ta); \
592 c0 += lo; hi += (c0<lo)?1:0; \
593 c1 += hi; c2 += (c1<hi)?1:0; \
596 # define sqr_add_c2(a,i,j,c0,c1,c2) \
597 mul_add_c2((a)[i],(a)[j],c0,c1,c2)
599 # else /* !BN_LLONG */
601 * Keep in mind that additions to hi can not overflow, because
602 * the high word of a multiplication result cannot be all-ones.
604 # define mul_add_c(a,b,c0,c1,c2) do { \
605 BN_ULONG lo = LBITS(a), hi = HBITS(a); \
606 BN_ULONG bl = LBITS(b), bh = HBITS(b); \
607 mul64(lo,hi,bl,bh); \
608 c0 = (c0+lo)&BN_MASK2; if (c0<lo) hi++; \
609 c1 = (c1+hi)&BN_MASK2; if (c1<hi) c2++; \
612 # define mul_add_c2(a,b,c0,c1,c2) do { \
614 BN_ULONG lo = LBITS(a), hi = HBITS(a); \
615 BN_ULONG bl = LBITS(b), bh = HBITS(b); \
616 mul64(lo,hi,bl,bh); \
618 c0 = (c0+lo)&BN_MASK2; if (c0<lo) tt++; \
619 c1 = (c1+tt)&BN_MASK2; if (c1<tt) c2++; \
620 c0 = (c0+lo)&BN_MASK2; if (c0<lo) hi++; \
621 c1 = (c1+hi)&BN_MASK2; if (c1<hi) c2++; \
624 # define sqr_add_c(a,i,c0,c1,c2) do { \
626 sqr64(lo,hi,(a)[i]); \
627 c0 = (c0+lo)&BN_MASK2; if (c0<lo) hi++; \
628 c1 = (c1+hi)&BN_MASK2; if (c1<hi) c2++; \
631 # define sqr_add_c2(a,i,j,c0,c1,c2) \
632 mul_add_c2((a)[i],(a)[j],c0,c1,c2)
633 # endif /* !BN_LLONG */
635 void bn_mul_comba8(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b)
642 mul_add_c(a[0], b[0], c1, c2, c3);
645 mul_add_c(a[0], b[1], c2, c3, c1);
646 mul_add_c(a[1], b[0], c2, c3, c1);
649 mul_add_c(a[2], b[0], c3, c1, c2);
650 mul_add_c(a[1], b[1], c3, c1, c2);
651 mul_add_c(a[0], b[2], c3, c1, c2);
654 mul_add_c(a[0], b[3], c1, c2, c3);
655 mul_add_c(a[1], b[2], c1, c2, c3);
656 mul_add_c(a[2], b[1], c1, c2, c3);
657 mul_add_c(a[3], b[0], c1, c2, c3);
660 mul_add_c(a[4], b[0], c2, c3, c1);
661 mul_add_c(a[3], b[1], c2, c3, c1);
662 mul_add_c(a[2], b[2], c2, c3, c1);
663 mul_add_c(a[1], b[3], c2, c3, c1);
664 mul_add_c(a[0], b[4], c2, c3, c1);
667 mul_add_c(a[0], b[5], c3, c1, c2);
668 mul_add_c(a[1], b[4], c3, c1, c2);
669 mul_add_c(a[2], b[3], c3, c1, c2);
670 mul_add_c(a[3], b[2], c3, c1, c2);
671 mul_add_c(a[4], b[1], c3, c1, c2);
672 mul_add_c(a[5], b[0], c3, c1, c2);
675 mul_add_c(a[6], b[0], c1, c2, c3);
676 mul_add_c(a[5], b[1], c1, c2, c3);
677 mul_add_c(a[4], b[2], c1, c2, c3);
678 mul_add_c(a[3], b[3], c1, c2, c3);
679 mul_add_c(a[2], b[4], c1, c2, c3);
680 mul_add_c(a[1], b[5], c1, c2, c3);
681 mul_add_c(a[0], b[6], c1, c2, c3);
684 mul_add_c(a[0], b[7], c2, c3, c1);
685 mul_add_c(a[1], b[6], c2, c3, c1);
686 mul_add_c(a[2], b[5], c2, c3, c1);
687 mul_add_c(a[3], b[4], c2, c3, c1);
688 mul_add_c(a[4], b[3], c2, c3, c1);
689 mul_add_c(a[5], b[2], c2, c3, c1);
690 mul_add_c(a[6], b[1], c2, c3, c1);
691 mul_add_c(a[7], b[0], c2, c3, c1);
694 mul_add_c(a[7], b[1], c3, c1, c2);
695 mul_add_c(a[6], b[2], c3, c1, c2);
696 mul_add_c(a[5], b[3], c3, c1, c2);
697 mul_add_c(a[4], b[4], c3, c1, c2);
698 mul_add_c(a[3], b[5], c3, c1, c2);
699 mul_add_c(a[2], b[6], c3, c1, c2);
700 mul_add_c(a[1], b[7], c3, c1, c2);
703 mul_add_c(a[2], b[7], c1, c2, c3);
704 mul_add_c(a[3], b[6], c1, c2, c3);
705 mul_add_c(a[4], b[5], c1, c2, c3);
706 mul_add_c(a[5], b[4], c1, c2, c3);
707 mul_add_c(a[6], b[3], c1, c2, c3);
708 mul_add_c(a[7], b[2], c1, c2, c3);
711 mul_add_c(a[7], b[3], c2, c3, c1);
712 mul_add_c(a[6], b[4], c2, c3, c1);
713 mul_add_c(a[5], b[5], c2, c3, c1);
714 mul_add_c(a[4], b[6], c2, c3, c1);
715 mul_add_c(a[3], b[7], c2, c3, c1);
718 mul_add_c(a[4], b[7], c3, c1, c2);
719 mul_add_c(a[5], b[6], c3, c1, c2);
720 mul_add_c(a[6], b[5], c3, c1, c2);
721 mul_add_c(a[7], b[4], c3, c1, c2);
724 mul_add_c(a[7], b[5], c1, c2, c3);
725 mul_add_c(a[6], b[6], c1, c2, c3);
726 mul_add_c(a[5], b[7], c1, c2, c3);
729 mul_add_c(a[6], b[7], c2, c3, c1);
730 mul_add_c(a[7], b[6], c2, c3, c1);
733 mul_add_c(a[7], b[7], c3, c1, c2);
738 void bn_mul_comba4(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b)
745 mul_add_c(a[0], b[0], c1, c2, c3);
748 mul_add_c(a[0], b[1], c2, c3, c1);
749 mul_add_c(a[1], b[0], c2, c3, c1);
752 mul_add_c(a[2], b[0], c3, c1, c2);
753 mul_add_c(a[1], b[1], c3, c1, c2);
754 mul_add_c(a[0], b[2], c3, c1, c2);
757 mul_add_c(a[0], b[3], c1, c2, c3);
758 mul_add_c(a[1], b[2], c1, c2, c3);
759 mul_add_c(a[2], b[1], c1, c2, c3);
760 mul_add_c(a[3], b[0], c1, c2, c3);
763 mul_add_c(a[3], b[1], c2, c3, c1);
764 mul_add_c(a[2], b[2], c2, c3, c1);
765 mul_add_c(a[1], b[3], c2, c3, c1);
768 mul_add_c(a[2], b[3], c3, c1, c2);
769 mul_add_c(a[3], b[2], c3, c1, c2);
772 mul_add_c(a[3], b[3], c1, c2, c3);
777 void bn_sqr_comba8(BN_ULONG *r, const BN_ULONG *a)
784 sqr_add_c(a, 0, c1, c2, c3);
787 sqr_add_c2(a, 1, 0, c2, c3, c1);
790 sqr_add_c(a, 1, c3, c1, c2);
791 sqr_add_c2(a, 2, 0, c3, c1, c2);
794 sqr_add_c2(a, 3, 0, c1, c2, c3);
795 sqr_add_c2(a, 2, 1, c1, c2, c3);
798 sqr_add_c(a, 2, c2, c3, c1);
799 sqr_add_c2(a, 3, 1, c2, c3, c1);
800 sqr_add_c2(a, 4, 0, c2, c3, c1);
803 sqr_add_c2(a, 5, 0, c3, c1, c2);
804 sqr_add_c2(a, 4, 1, c3, c1, c2);
805 sqr_add_c2(a, 3, 2, c3, c1, c2);
808 sqr_add_c(a, 3, c1, c2, c3);
809 sqr_add_c2(a, 4, 2, c1, c2, c3);
810 sqr_add_c2(a, 5, 1, c1, c2, c3);
811 sqr_add_c2(a, 6, 0, c1, c2, c3);
814 sqr_add_c2(a, 7, 0, c2, c3, c1);
815 sqr_add_c2(a, 6, 1, c2, c3, c1);
816 sqr_add_c2(a, 5, 2, c2, c3, c1);
817 sqr_add_c2(a, 4, 3, c2, c3, c1);
820 sqr_add_c(a, 4, c3, c1, c2);
821 sqr_add_c2(a, 5, 3, c3, c1, c2);
822 sqr_add_c2(a, 6, 2, c3, c1, c2);
823 sqr_add_c2(a, 7, 1, c3, c1, c2);
826 sqr_add_c2(a, 7, 2, c1, c2, c3);
827 sqr_add_c2(a, 6, 3, c1, c2, c3);
828 sqr_add_c2(a, 5, 4, c1, c2, c3);
831 sqr_add_c(a, 5, c2, c3, c1);
832 sqr_add_c2(a, 6, 4, c2, c3, c1);
833 sqr_add_c2(a, 7, 3, c2, c3, c1);
836 sqr_add_c2(a, 7, 4, c3, c1, c2);
837 sqr_add_c2(a, 6, 5, c3, c1, c2);
840 sqr_add_c(a, 6, c1, c2, c3);
841 sqr_add_c2(a, 7, 5, c1, c2, c3);
844 sqr_add_c2(a, 7, 6, c2, c3, c1);
847 sqr_add_c(a, 7, c3, c1, c2);
852 void bn_sqr_comba4(BN_ULONG *r, const BN_ULONG *a)
859 sqr_add_c(a, 0, c1, c2, c3);
862 sqr_add_c2(a, 1, 0, c2, c3, c1);
865 sqr_add_c(a, 1, c3, c1, c2);
866 sqr_add_c2(a, 2, 0, c3, c1, c2);
869 sqr_add_c2(a, 3, 0, c1, c2, c3);
870 sqr_add_c2(a, 2, 1, c1, c2, c3);
873 sqr_add_c(a, 2, c2, c3, c1);
874 sqr_add_c2(a, 3, 1, c2, c3, c1);
877 sqr_add_c2(a, 3, 2, c3, c1, c2);
880 sqr_add_c(a, 3, c1, c2, c3);
885 # ifdef OPENSSL_NO_ASM
886 # ifdef OPENSSL_BN_ASM_MONT
889 * This is essentially reference implementation, which may or may not
890 * result in performance improvement. E.g. on IA-32 this routine was
891 * observed to give 40% faster rsa1024 private key operations and 10%
892 * faster rsa4096 ones, while on AMD64 it improves rsa1024 sign only
893 * by 10% and *worsens* rsa4096 sign by 15%. Once again, it's a
894 * reference implementation, one to be used as starting point for
895 * platform-specific assembler. Mentioned numbers apply to compiler
896 * generated code compiled with and without -DOPENSSL_BN_ASM_MONT and
897 * can vary not only from platform to platform, but even for compiler
898 * versions. Assembler vs. assembler improvement coefficients can
899 * [and are known to] differ and are to be documented elsewhere.
901 int bn_mul_mont(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp,
902 const BN_ULONG *np, const BN_ULONG *n0p, int num)
904 BN_ULONG c0, c1, ml, *tp, n0;
908 volatile BN_ULONG *vp;
911 # if 0 /* template for platform-specific
914 return bn_sqr_mont(rp, ap, np, n0p, num);
916 vp = tp = alloca((num + 2) * sizeof(BN_ULONG));
925 for (j = 0; j < num; ++j)
926 mul(tp[j], ap[j], ml, mh, c0);
928 for (j = 0; j < num; ++j)
929 mul(tp[j], ap[j], ml, c0);
936 for (i = 0; i < num; i++) {
942 for (j = 0; j < num; ++j)
943 mul_add(tp[j], ap[j], ml, mh, c0);
945 for (j = 0; j < num; ++j)
946 mul_add(tp[j], ap[j], ml, c0);
948 c1 = (tp[num] + c0) & BN_MASK2;
950 tp[num + 1] = (c1 < c0 ? 1 : 0);
953 ml = (c1 * n0) & BN_MASK2;
958 mul_add(c1, np[0], ml, mh, c0);
960 mul_add(c1, ml, np[0], c0);
962 for (j = 1; j < num; j++) {
965 mul_add(c1, np[j], ml, mh, c0);
967 mul_add(c1, ml, np[j], c0);
969 tp[j - 1] = c1 & BN_MASK2;
971 c1 = (tp[num] + c0) & BN_MASK2;
973 tp[num] = tp[num + 1] + (c1 < c0 ? 1 : 0);
976 if (tp[num] != 0 || tp[num - 1] >= np[num - 1]) {
977 c0 = bn_sub_words(rp, tp, np, num);
978 if (tp[num] != 0 || c0 == 0) {
979 for (i = 0; i < num + 2; i++)
984 for (i = 0; i < num; i++)
985 rp[i] = tp[i], vp[i] = 0;
992 * Return value of 0 indicates that multiplication/convolution was not
993 * performed to signal the caller to fall down to alternative/original
996 int bn_mul_mont(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp,
997 const BN_ULONG *np, const BN_ULONG *n0, int num)
1001 # endif /* OPENSSL_BN_ASM_MONT */
1004 #else /* !BN_MUL_COMBA */
1006 /* hmm... is it faster just to do a multiply? */
1007 # undef bn_sqr_comba4
1008 void bn_sqr_comba4(BN_ULONG *r, const BN_ULONG *a)
1011 bn_sqr_normal(r, a, 4, t);
1014 # undef bn_sqr_comba8
1015 void bn_sqr_comba8(BN_ULONG *r, const BN_ULONG *a)
1018 bn_sqr_normal(r, a, 8, t);
1021 void bn_mul_comba4(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b)
1023 r[4] = bn_mul_words(&(r[0]), a, 4, b[0]);
1024 r[5] = bn_mul_add_words(&(r[1]), a, 4, b[1]);
1025 r[6] = bn_mul_add_words(&(r[2]), a, 4, b[2]);
1026 r[7] = bn_mul_add_words(&(r[3]), a, 4, b[3]);
1029 void bn_mul_comba8(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b)
1031 r[8] = bn_mul_words(&(r[0]), a, 8, b[0]);
1032 r[9] = bn_mul_add_words(&(r[1]), a, 8, b[1]);
1033 r[10] = bn_mul_add_words(&(r[2]), a, 8, b[2]);
1034 r[11] = bn_mul_add_words(&(r[3]), a, 8, b[3]);
1035 r[12] = bn_mul_add_words(&(r[4]), a, 8, b[4]);
1036 r[13] = bn_mul_add_words(&(r[5]), a, 8, b[5]);
1037 r[14] = bn_mul_add_words(&(r[6]), a, 8, b[6]);
1038 r[15] = bn_mul_add_words(&(r[7]), a, 8, b[7]);
1041 # ifdef OPENSSL_NO_ASM
1042 # ifdef OPENSSL_BN_ASM_MONT
1043 # include <alloca.h>
1044 int bn_mul_mont(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp,
1045 const BN_ULONG *np, const BN_ULONG *n0p, int num)
1047 BN_ULONG c0, c1, *tp, n0 = *n0p;
1048 volatile BN_ULONG *vp;
1051 vp = tp = alloca((num + 2) * sizeof(BN_ULONG));
1053 for (i = 0; i <= num; i++)
1056 for (i = 0; i < num; i++) {
1057 c0 = bn_mul_add_words(tp, ap, num, bp[i]);
1058 c1 = (tp[num] + c0) & BN_MASK2;
1060 tp[num + 1] = (c1 < c0 ? 1 : 0);
1062 c0 = bn_mul_add_words(tp, np, num, tp[0] * n0);
1063 c1 = (tp[num] + c0) & BN_MASK2;
1065 tp[num + 1] += (c1 < c0 ? 1 : 0);
1066 for (j = 0; j <= num; j++)
1070 if (tp[num] != 0 || tp[num - 1] >= np[num - 1]) {
1071 c0 = bn_sub_words(rp, tp, np, num);
1072 if (tp[num] != 0 || c0 == 0) {
1073 for (i = 0; i < num + 2; i++)
1078 for (i = 0; i < num; i++)
1079 rp[i] = tp[i], vp[i] = 0;
1085 int bn_mul_mont(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp,
1086 const BN_ULONG *np, const BN_ULONG *n0, int num)
1090 # endif /* OPENSSL_BN_ASM_MONT */
1093 #endif /* !BN_MUL_COMBA */