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 */
65 #include <openssl/crypto.h>
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, BN_ULONG w)
76 if (num <= 0) return(c1);
78 #ifndef OPENSSL_SMALL_FOOTPRINT
81 mul_add(rp[0],ap[0],w,c1);
82 mul_add(rp[1],ap[1],w,c1);
83 mul_add(rp[2],ap[2],w,c1);
84 mul_add(rp[3],ap[3],w,c1);
90 mul_add(rp[0],ap[0],w,c1);
97 BN_ULONG bn_mul_words(BN_ULONG *rp, const BN_ULONG *ap, int num, BN_ULONG w)
102 if (num <= 0) return(c1);
104 #ifndef OPENSSL_SMALL_FOOTPRINT
107 mul(rp[0],ap[0],w,c1);
108 mul(rp[1],ap[1],w,c1);
109 mul(rp[2],ap[2],w,c1);
110 mul(rp[3],ap[3],w,c1);
111 ap+=4; rp+=4; num-=4;
116 mul(rp[0],ap[0],w,c1);
122 void bn_sqr_words(BN_ULONG *r, const BN_ULONG *a, int n)
127 #ifndef OPENSSL_SMALL_FOOTPRINT
144 #else /* !(defined(BN_LLONG) || defined(BN_UMULT_HIGH)) */
146 BN_ULONG bn_mul_add_words(BN_ULONG *rp, const BN_ULONG *ap, int num, BN_ULONG w)
152 if (num <= 0) return((BN_ULONG)0);
157 #ifndef OPENSSL_SMALL_FOOTPRINT
160 mul_add(rp[0],ap[0],bl,bh,c);
161 mul_add(rp[1],ap[1],bl,bh,c);
162 mul_add(rp[2],ap[2],bl,bh,c);
163 mul_add(rp[3],ap[3],bl,bh,c);
164 ap+=4; rp+=4; num-=4;
169 mul_add(rp[0],ap[0],bl,bh,c);
175 BN_ULONG bn_mul_words(BN_ULONG *rp, const BN_ULONG *ap, int num, BN_ULONG w)
181 if (num <= 0) return((BN_ULONG)0);
186 #ifndef OPENSSL_SMALL_FOOTPRINT
189 mul(rp[0],ap[0],bl,bh,carry);
190 mul(rp[1],ap[1],bl,bh,carry);
191 mul(rp[2],ap[2],bl,bh,carry);
192 mul(rp[3],ap[3],bl,bh,carry);
193 ap+=4; rp+=4; num-=4;
198 mul(rp[0],ap[0],bl,bh,carry);
204 void bn_sqr_words(BN_ULONG *r, const BN_ULONG *a, int n)
209 #ifndef OPENSSL_SMALL_FOOTPRINT
212 sqr64(r[0],r[1],a[0]);
213 sqr64(r[2],r[3],a[1]);
214 sqr64(r[4],r[5],a[2]);
215 sqr64(r[6],r[7],a[3]);
221 sqr64(r[0],r[1],a[0]);
226 #endif /* !(defined(BN_LLONG) || defined(BN_UMULT_HIGH)) */
228 #if defined(BN_LLONG) && defined(BN_DIV2W)
230 BN_ULONG bn_div_words(BN_ULONG h, BN_ULONG l, BN_ULONG d)
232 return((BN_ULONG)(((((BN_ULLONG)h)<<BN_BITS2)|l)/(BN_ULLONG)d));
237 /* Divide h,l by d and return the result. */
238 /* I need to test this some more :-( */
239 BN_ULONG bn_div_words(BN_ULONG h, BN_ULONG l, BN_ULONG d)
241 BN_ULONG dh,dl,q,ret=0,th,tl,t;
244 if (d == 0) return(BN_MASK2);
246 i=BN_num_bits_word(d);
247 assert((i == BN_BITS2) || (h <= (BN_ULONG)1<<i));
255 h=(h<<i)|(l>>(BN_BITS2-i));
258 dh=(d&BN_MASK2h)>>BN_BITS4;
262 if ((h>>BN_BITS4) == dh)
275 ((l&BN_MASK2h)>>BN_BITS4))))
282 tl=(tl<<BN_BITS4)&BN_MASK2h;
294 if (--count == 0) break;
297 h=((h<<BN_BITS4)|(l>>BN_BITS4))&BN_MASK2;
298 l=(l&BN_MASK2l)<<BN_BITS4;
303 #endif /* !defined(BN_LLONG) && defined(BN_DIV2W) */
306 BN_ULONG bn_add_words(BN_ULONG *r, const BN_ULONG *a, const BN_ULONG *b, int n)
311 if (n <= 0) return((BN_ULONG)0);
313 #ifndef OPENSSL_SMALL_FOOTPRINT
316 ll+=(BN_ULLONG)a[0]+b[0];
317 r[0]=(BN_ULONG)ll&BN_MASK2;
319 ll+=(BN_ULLONG)a[1]+b[1];
320 r[1]=(BN_ULONG)ll&BN_MASK2;
322 ll+=(BN_ULLONG)a[2]+b[2];
323 r[2]=(BN_ULONG)ll&BN_MASK2;
325 ll+=(BN_ULLONG)a[3]+b[3];
326 r[3]=(BN_ULONG)ll&BN_MASK2;
328 a+=4; b+=4; r+=4; n-=4;
333 ll+=(BN_ULLONG)a[0]+b[0];
334 r[0]=(BN_ULONG)ll&BN_MASK2;
338 return((BN_ULONG)ll);
340 #else /* !BN_LLONG */
341 BN_ULONG bn_add_words(BN_ULONG *r, const BN_ULONG *a, const BN_ULONG *b, int n)
346 if (n <= 0) return((BN_ULONG)0);
349 #ifndef OPENSSL_SMALL_FOOTPRINT
376 a+=4; b+=4; r+=4; n-=4;
391 #endif /* !BN_LLONG */
393 BN_ULONG bn_sub_words(BN_ULONG *r, const BN_ULONG *a, const BN_ULONG *b, int n)
399 if (n <= 0) return((BN_ULONG)0);
401 #ifndef OPENSSL_SMALL_FOOTPRINT
405 r[0]=(t1-t2-c)&BN_MASK2;
406 if (t1 != t2) c=(t1 < t2);
408 r[1]=(t1-t2-c)&BN_MASK2;
409 if (t1 != t2) c=(t1 < t2);
411 r[2]=(t1-t2-c)&BN_MASK2;
412 if (t1 != t2) c=(t1 < t2);
414 r[3]=(t1-t2-c)&BN_MASK2;
415 if (t1 != t2) c=(t1 < t2);
416 a+=4; b+=4; r+=4; n-=4;
422 r[0]=(t1-t2-c)&BN_MASK2;
423 if (t1 != t2) c=(t1 < t2);
429 #if defined(BN_MUL_COMBA) && !defined(OPENSSL_SMALL_FOOTPRINT)
436 /* mul_add_c(a,b,c0,c1,c2) -- c+=a*b for three word number c=(c2,c1,c0) */
437 /* mul_add_c2(a,b,c0,c1,c2) -- c+=2*a*b for three word number c=(c2,c1,c0) */
438 /* sqr_add_c(a,i,c0,c1,c2) -- c+=a[i]^2 for three word number c=(c2,c1,c0) */
439 /* sqr_add_c2(a,i,c0,c1,c2) -- c+=2*a[i]*a[j] for three word number c=(c2,c1,c0) */
443 * Keep in mind that additions to multiplication result can not
444 * overflow, because its high half cannot be all-ones.
446 # define mul_add_c(a,b,c0,c1,c2) do { \
448 BN_ULLONG t = (BN_ULLONG)(a)*(b); \
449 t += c0; /* no carry */ \
450 c0 = (BN_ULONG)Lw(t); \
451 hi = (BN_ULONG)Hw(t); \
452 c1 = (c1+hi)&BN_MASK2; if (c1<hi) c2++; \
455 # define mul_add_c2(a,b,c0,c1,c2) do { \
457 BN_ULLONG t = (BN_ULLONG)(a)*(b); \
458 BN_ULLONG tt = t+c0; /* no carry */ \
459 c0 = (BN_ULONG)Lw(tt); \
460 hi = (BN_ULONG)Hw(tt); \
461 c1 = (c1+hi)&BN_MASK2; if (c1<hi) c2++; \
462 t += c0; /* no carry */ \
463 c0 = (BN_ULONG)Lw(t); \
464 hi = (BN_ULONG)Hw(t); \
465 c1 = (c1+hi)&BN_MASK2; if (c1<hi) c2++; \
468 # define sqr_add_c(a,i,c0,c1,c2) do { \
470 BN_ULLONG t = (BN_ULLONG)a[i]*a[i]; \
471 t += c0; /* no carry */ \
472 c0 = (BN_ULONG)Lw(t); \
473 hi = (BN_ULONG)Hw(t); \
474 c1 = (c1+hi)&BN_MASK2; if (c1<hi) c2++; \
477 # define sqr_add_c2(a,i,j,c0,c1,c2) \
478 mul_add_c2((a)[i],(a)[j],c0,c1,c2)
480 #elif defined(BN_UMULT_LOHI)
482 * Keep in mind that additions to hi can not overflow, because
483 * the high word of a multiplication result cannot be all-ones.
485 # define mul_add_c(a,b,c0,c1,c2) do { \
486 BN_ULONG ta = (a), tb = (b); \
488 BN_UMULT_LOHI(lo,hi,ta,tb); \
489 c0 += lo; hi += (c0<lo)?1:0; \
490 c1 += hi; c2 += (c1<hi)?1:0; \
493 # define mul_add_c2(a,b,c0,c1,c2) do { \
494 BN_ULONG ta = (a), tb = (b); \
495 BN_ULONG lo, hi, tt; \
496 BN_UMULT_LOHI(lo,hi,ta,tb); \
497 c0 += lo; tt = hi+((c0<lo)?1:0); \
498 c1 += tt; c2 += (c1<tt)?1:0; \
499 c0 += lo; hi += (c0<lo)?1:0; \
500 c1 += hi; c2 += (c1<hi)?1:0; \
503 # define sqr_add_c(a,i,c0,c1,c2) do { \
504 BN_ULONG ta = (a)[i]; \
506 BN_UMULT_LOHI(lo,hi,ta,ta); \
507 c0 += lo; hi += (c0<lo)?1:0; \
508 c1 += hi; c2 += (c1<hi)?1:0; \
511 # define sqr_add_c2(a,i,j,c0,c1,c2) \
512 mul_add_c2((a)[i],(a)[j],c0,c1,c2)
514 #elif defined(BN_UMULT_HIGH)
516 * Keep in mind that additions to hi can not overflow, because
517 * the high word of a multiplication result cannot be all-ones.
519 # define mul_add_c(a,b,c0,c1,c2) do { \
520 BN_ULONG ta = (a), tb = (b); \
521 BN_ULONG lo = ta * tb; \
522 BN_ULONG hi = BN_UMULT_HIGH(ta,tb); \
523 c0 += lo; hi += (c0<lo)?1:0; \
524 c1 += hi; c2 += (c1<hi)?1:0; \
527 # define mul_add_c2(a,b,c0,c1,c2) do { \
528 BN_ULONG ta = (a), tb = (b), tt; \
529 BN_ULONG lo = ta * tb; \
530 BN_ULONG hi = BN_UMULT_HIGH(ta,tb); \
531 c0 += lo; tt = hi + ((c0<lo)?1:0); \
532 c1 += tt; c2 += (c1<tt)?1:0; \
533 c0 += lo; hi += (c0<lo)?1:0; \
534 c1 += hi; c2 += (c1<hi)?1:0; \
537 # define sqr_add_c(a,i,c0,c1,c2) do { \
538 BN_ULONG ta = (a)[i]; \
539 BN_ULONG lo = ta * ta; \
540 BN_ULONG hi = BN_UMULT_HIGH(ta,ta); \
541 c0 += lo; hi += (c0<lo)?1:0; \
542 c1 += hi; c2 += (c1<hi)?1:0; \
545 #define sqr_add_c2(a,i,j,c0,c1,c2) \
546 mul_add_c2((a)[i],(a)[j],c0,c1,c2)
548 #else /* !BN_LLONG */
550 * Keep in mind that additions to hi can not overflow, because
551 * the high word of a multiplication result cannot be all-ones.
553 # define mul_add_c(a,b,c0,c1,c2) do { \
554 BN_ULONG lo = LBITS(a), hi = HBITS(a); \
555 BN_ULONG bl = LBITS(b), bh = HBITS(b); \
556 mul64(lo,hi,bl,bh); \
557 c0 = (c0+lo)&BN_MASK2; if (c0<lo) hi++; \
558 c1 = (c1+hi)&BN_MASK2; if (c1<hi) c2++; \
561 # define mul_add_c2(a,b,c0,c1,c2) do { \
563 BN_ULONG lo = LBITS(a), hi = HBITS(a); \
564 BN_ULONG bl = LBITS(b), bh = HBITS(b); \
565 mul64(lo,hi,bl,bh); \
567 c0 = (c0+lo)&BN_MASK2; if (c0<lo) tt++; \
568 c1 = (c1+tt)&BN_MASK2; if (c1<tt) c2++; \
569 c0 = (c0+lo)&BN_MASK2; if (c0<lo) hi++; \
570 c1 = (c1+hi)&BN_MASK2; if (c1<hi) c2++; \
573 # define sqr_add_c(a,i,c0,c1,c2) do { \
575 sqr64(lo,hi,(a)[i]); \
576 c0 = (c0+lo)&BN_MASK2; if (c0<lo) hi++; \
577 c1 = (c1+hi)&BN_MASK2; if (c1<hi) c2++; \
580 # define sqr_add_c2(a,i,j,c0,c1,c2) \
581 mul_add_c2((a)[i],(a)[j],c0,c1,c2)
582 #endif /* !BN_LLONG */
584 void bn_mul_comba8(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b)
591 mul_add_c(a[0],b[0],c1,c2,c3);
594 mul_add_c(a[0],b[1],c2,c3,c1);
595 mul_add_c(a[1],b[0],c2,c3,c1);
598 mul_add_c(a[2],b[0],c3,c1,c2);
599 mul_add_c(a[1],b[1],c3,c1,c2);
600 mul_add_c(a[0],b[2],c3,c1,c2);
603 mul_add_c(a[0],b[3],c1,c2,c3);
604 mul_add_c(a[1],b[2],c1,c2,c3);
605 mul_add_c(a[2],b[1],c1,c2,c3);
606 mul_add_c(a[3],b[0],c1,c2,c3);
609 mul_add_c(a[4],b[0],c2,c3,c1);
610 mul_add_c(a[3],b[1],c2,c3,c1);
611 mul_add_c(a[2],b[2],c2,c3,c1);
612 mul_add_c(a[1],b[3],c2,c3,c1);
613 mul_add_c(a[0],b[4],c2,c3,c1);
616 mul_add_c(a[0],b[5],c3,c1,c2);
617 mul_add_c(a[1],b[4],c3,c1,c2);
618 mul_add_c(a[2],b[3],c3,c1,c2);
619 mul_add_c(a[3],b[2],c3,c1,c2);
620 mul_add_c(a[4],b[1],c3,c1,c2);
621 mul_add_c(a[5],b[0],c3,c1,c2);
624 mul_add_c(a[6],b[0],c1,c2,c3);
625 mul_add_c(a[5],b[1],c1,c2,c3);
626 mul_add_c(a[4],b[2],c1,c2,c3);
627 mul_add_c(a[3],b[3],c1,c2,c3);
628 mul_add_c(a[2],b[4],c1,c2,c3);
629 mul_add_c(a[1],b[5],c1,c2,c3);
630 mul_add_c(a[0],b[6],c1,c2,c3);
633 mul_add_c(a[0],b[7],c2,c3,c1);
634 mul_add_c(a[1],b[6],c2,c3,c1);
635 mul_add_c(a[2],b[5],c2,c3,c1);
636 mul_add_c(a[3],b[4],c2,c3,c1);
637 mul_add_c(a[4],b[3],c2,c3,c1);
638 mul_add_c(a[5],b[2],c2,c3,c1);
639 mul_add_c(a[6],b[1],c2,c3,c1);
640 mul_add_c(a[7],b[0],c2,c3,c1);
643 mul_add_c(a[7],b[1],c3,c1,c2);
644 mul_add_c(a[6],b[2],c3,c1,c2);
645 mul_add_c(a[5],b[3],c3,c1,c2);
646 mul_add_c(a[4],b[4],c3,c1,c2);
647 mul_add_c(a[3],b[5],c3,c1,c2);
648 mul_add_c(a[2],b[6],c3,c1,c2);
649 mul_add_c(a[1],b[7],c3,c1,c2);
652 mul_add_c(a[2],b[7],c1,c2,c3);
653 mul_add_c(a[3],b[6],c1,c2,c3);
654 mul_add_c(a[4],b[5],c1,c2,c3);
655 mul_add_c(a[5],b[4],c1,c2,c3);
656 mul_add_c(a[6],b[3],c1,c2,c3);
657 mul_add_c(a[7],b[2],c1,c2,c3);
660 mul_add_c(a[7],b[3],c2,c3,c1);
661 mul_add_c(a[6],b[4],c2,c3,c1);
662 mul_add_c(a[5],b[5],c2,c3,c1);
663 mul_add_c(a[4],b[6],c2,c3,c1);
664 mul_add_c(a[3],b[7],c2,c3,c1);
667 mul_add_c(a[4],b[7],c3,c1,c2);
668 mul_add_c(a[5],b[6],c3,c1,c2);
669 mul_add_c(a[6],b[5],c3,c1,c2);
670 mul_add_c(a[7],b[4],c3,c1,c2);
673 mul_add_c(a[7],b[5],c1,c2,c3);
674 mul_add_c(a[6],b[6],c1,c2,c3);
675 mul_add_c(a[5],b[7],c1,c2,c3);
678 mul_add_c(a[6],b[7],c2,c3,c1);
679 mul_add_c(a[7],b[6],c2,c3,c1);
682 mul_add_c(a[7],b[7],c3,c1,c2);
687 void bn_mul_comba4(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b)
694 mul_add_c(a[0],b[0],c1,c2,c3);
697 mul_add_c(a[0],b[1],c2,c3,c1);
698 mul_add_c(a[1],b[0],c2,c3,c1);
701 mul_add_c(a[2],b[0],c3,c1,c2);
702 mul_add_c(a[1],b[1],c3,c1,c2);
703 mul_add_c(a[0],b[2],c3,c1,c2);
706 mul_add_c(a[0],b[3],c1,c2,c3);
707 mul_add_c(a[1],b[2],c1,c2,c3);
708 mul_add_c(a[2],b[1],c1,c2,c3);
709 mul_add_c(a[3],b[0],c1,c2,c3);
712 mul_add_c(a[3],b[1],c2,c3,c1);
713 mul_add_c(a[2],b[2],c2,c3,c1);
714 mul_add_c(a[1],b[3],c2,c3,c1);
717 mul_add_c(a[2],b[3],c3,c1,c2);
718 mul_add_c(a[3],b[2],c3,c1,c2);
721 mul_add_c(a[3],b[3],c1,c2,c3);
726 void bn_sqr_comba8(BN_ULONG *r, const BN_ULONG *a)
733 sqr_add_c(a,0,c1,c2,c3);
736 sqr_add_c2(a,1,0,c2,c3,c1);
739 sqr_add_c(a,1,c3,c1,c2);
740 sqr_add_c2(a,2,0,c3,c1,c2);
743 sqr_add_c2(a,3,0,c1,c2,c3);
744 sqr_add_c2(a,2,1,c1,c2,c3);
747 sqr_add_c(a,2,c2,c3,c1);
748 sqr_add_c2(a,3,1,c2,c3,c1);
749 sqr_add_c2(a,4,0,c2,c3,c1);
752 sqr_add_c2(a,5,0,c3,c1,c2);
753 sqr_add_c2(a,4,1,c3,c1,c2);
754 sqr_add_c2(a,3,2,c3,c1,c2);
757 sqr_add_c(a,3,c1,c2,c3);
758 sqr_add_c2(a,4,2,c1,c2,c3);
759 sqr_add_c2(a,5,1,c1,c2,c3);
760 sqr_add_c2(a,6,0,c1,c2,c3);
763 sqr_add_c2(a,7,0,c2,c3,c1);
764 sqr_add_c2(a,6,1,c2,c3,c1);
765 sqr_add_c2(a,5,2,c2,c3,c1);
766 sqr_add_c2(a,4,3,c2,c3,c1);
769 sqr_add_c(a,4,c3,c1,c2);
770 sqr_add_c2(a,5,3,c3,c1,c2);
771 sqr_add_c2(a,6,2,c3,c1,c2);
772 sqr_add_c2(a,7,1,c3,c1,c2);
775 sqr_add_c2(a,7,2,c1,c2,c3);
776 sqr_add_c2(a,6,3,c1,c2,c3);
777 sqr_add_c2(a,5,4,c1,c2,c3);
780 sqr_add_c(a,5,c2,c3,c1);
781 sqr_add_c2(a,6,4,c2,c3,c1);
782 sqr_add_c2(a,7,3,c2,c3,c1);
785 sqr_add_c2(a,7,4,c3,c1,c2);
786 sqr_add_c2(a,6,5,c3,c1,c2);
789 sqr_add_c(a,6,c1,c2,c3);
790 sqr_add_c2(a,7,5,c1,c2,c3);
793 sqr_add_c2(a,7,6,c2,c3,c1);
796 sqr_add_c(a,7,c3,c1,c2);
801 void bn_sqr_comba4(BN_ULONG *r, const BN_ULONG *a)
808 sqr_add_c(a,0,c1,c2,c3);
811 sqr_add_c2(a,1,0,c2,c3,c1);
814 sqr_add_c(a,1,c3,c1,c2);
815 sqr_add_c2(a,2,0,c3,c1,c2);
818 sqr_add_c2(a,3,0,c1,c2,c3);
819 sqr_add_c2(a,2,1,c1,c2,c3);
822 sqr_add_c(a,2,c2,c3,c1);
823 sqr_add_c2(a,3,1,c2,c3,c1);
826 sqr_add_c2(a,3,2,c3,c1,c2);
829 sqr_add_c(a,3,c1,c2,c3);
834 #ifdef OPENSSL_NO_ASM
835 #ifdef OPENSSL_BN_ASM_MONT
838 * This is essentially reference implementation, which may or may not
839 * result in performance improvement. E.g. on IA-32 this routine was
840 * observed to give 40% faster rsa1024 private key operations and 10%
841 * faster rsa4096 ones, while on AMD64 it improves rsa1024 sign only
842 * by 10% and *worsens* rsa4096 sign by 15%. Once again, it's a
843 * reference implementation, one to be used as starting point for
844 * platform-specific assembler. Mentioned numbers apply to compiler
845 * generated code compiled with and without -DOPENSSL_BN_ASM_MONT and
846 * can vary not only from platform to platform, but even for compiler
847 * versions. Assembler vs. assembler improvement coefficients can
848 * [and are known to] differ and are to be documented elsewhere.
850 int bn_mul_mont(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp, const BN_ULONG *np,const BN_ULONG *n0p, int num)
852 BN_ULONG c0,c1,ml,*tp,n0;
856 volatile BN_ULONG *vp;
859 #if 0 /* template for platform-specific implementation */
860 if (ap==bp) return bn_sqr_mont(rp,ap,np,n0p,num);
862 vp = tp = alloca((num+2)*sizeof(BN_ULONG));
872 mul(tp[j],ap[j],ml,mh,c0);
875 mul(tp[j],ap[j],ml,c0);
890 mul_add(tp[j],ap[j],ml,mh,c0);
893 mul_add(tp[j],ap[j],ml,c0);
895 c1 = (tp[num] + c0)&BN_MASK2;
897 tp[num+1] = (c1<c0?1:0);
900 ml = (c1*n0)&BN_MASK2;
905 mul_add(c1,np[0],ml,mh,c0);
907 mul_add(c1,ml,np[0],c0);
913 mul_add(c1,np[j],ml,mh,c0);
915 mul_add(c1,ml,np[j],c0);
917 tp[j-1] = c1&BN_MASK2;
919 c1 = (tp[num] + c0)&BN_MASK2;
921 tp[num] = tp[num+1] + (c1<c0?1:0);
924 if (tp[num]!=0 || tp[num-1]>=np[num-1])
926 c0 = bn_sub_words(rp,tp,np,num);
927 if (tp[num]!=0 || c0==0)
929 for(i=0;i<num+2;i++) vp[i] = 0;
933 for(i=0;i<num;i++) rp[i] = tp[i], vp[i] = 0;
940 * Return value of 0 indicates that multiplication/convolution was not
941 * performed to signal the caller to fall down to alternative/original
944 int bn_mul_mont(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp, const BN_ULONG *np,const BN_ULONG *n0, int num)
946 #endif /* OPENSSL_BN_ASM_MONT */
949 #else /* !BN_MUL_COMBA */
951 /* hmm... is it faster just to do a multiply? */
954 void bn_sqr_comba4(BN_ULONG *r, const BN_ULONG *a)
957 bn_sqr_normal(r,a,4,t);
960 void bn_sqr_comba8(BN_ULONG *r, const BN_ULONG *a)
963 bn_sqr_normal(r,a,8,t);
966 void bn_mul_comba4(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b)
968 r[4]=bn_mul_words( &(r[0]),a,4,b[0]);
969 r[5]=bn_mul_add_words(&(r[1]),a,4,b[1]);
970 r[6]=bn_mul_add_words(&(r[2]),a,4,b[2]);
971 r[7]=bn_mul_add_words(&(r[3]),a,4,b[3]);
974 void bn_mul_comba8(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b)
976 r[ 8]=bn_mul_words( &(r[0]),a,8,b[0]);
977 r[ 9]=bn_mul_add_words(&(r[1]),a,8,b[1]);
978 r[10]=bn_mul_add_words(&(r[2]),a,8,b[2]);
979 r[11]=bn_mul_add_words(&(r[3]),a,8,b[3]);
980 r[12]=bn_mul_add_words(&(r[4]),a,8,b[4]);
981 r[13]=bn_mul_add_words(&(r[5]),a,8,b[5]);
982 r[14]=bn_mul_add_words(&(r[6]),a,8,b[6]);
983 r[15]=bn_mul_add_words(&(r[7]),a,8,b[7]);
986 #ifdef OPENSSL_NO_ASM
987 #ifdef OPENSSL_BN_ASM_MONT
989 int bn_mul_mont(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp, const BN_ULONG *np,const BN_ULONG *n0p, int num)
991 BN_ULONG c0,c1,*tp,n0=*n0p;
992 volatile BN_ULONG *vp;
995 vp = tp = alloca((num+2)*sizeof(BN_ULONG));
997 for(i=0;i<=num;i++) tp[i]=0;
1001 c0 = bn_mul_add_words(tp,ap,num,bp[i]);
1002 c1 = (tp[num] + c0)&BN_MASK2;
1004 tp[num+1] = (c1<c0?1:0);
1006 c0 = bn_mul_add_words(tp,np,num,tp[0]*n0);
1007 c1 = (tp[num] + c0)&BN_MASK2;
1009 tp[num+1] += (c1<c0?1:0);
1010 for(j=0;j<=num;j++) tp[j]=tp[j+1];
1013 if (tp[num]!=0 || tp[num-1]>=np[num-1])
1015 c0 = bn_sub_words(rp,tp,np,num);
1016 if (tp[num]!=0 || c0==0)
1018 for(i=0;i<num+2;i++) vp[i] = 0;
1022 for(i=0;i<num;i++) rp[i] = tp[i], vp[i] = 0;
1028 int bn_mul_mont(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp, const BN_ULONG *np,const BN_ULONG *n0, int num)
1030 #endif /* OPENSSL_BN_ASM_MONT */
1033 #endif /* !BN_MUL_COMBA */