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 */
66 #include <openssl/crypto.h>
70 #if defined(BN_LLONG) || defined(BN_UMULT_HIGH)
72 BN_ULONG bn_mul_add_words(BN_ULONG *rp, const BN_ULONG *ap, int num, BN_ULONG w)
77 if (num <= 0) return(c1);
79 #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);
91 mul_add(rp[0],ap[0],w,c1);
98 BN_ULONG bn_mul_words(BN_ULONG *rp, const BN_ULONG *ap, int num, BN_ULONG w)
103 if (num <= 0) return(c1);
105 #ifndef OPENSSL_SMALL_FOOTPRINT
108 mul(rp[0],ap[0],w,c1);
109 mul(rp[1],ap[1],w,c1);
110 mul(rp[2],ap[2],w,c1);
111 mul(rp[3],ap[3],w,c1);
112 ap+=4; rp+=4; num-=4;
117 mul(rp[0],ap[0],w,c1);
123 void bn_sqr_words(BN_ULONG *r, const BN_ULONG *a, int n)
128 #ifndef OPENSSL_SMALL_FOOTPRINT
145 #else /* !(defined(BN_LLONG) || defined(BN_UMULT_HIGH)) */
147 BN_ULONG bn_mul_add_words(BN_ULONG *rp, const BN_ULONG *ap, int num, BN_ULONG w)
153 if (num <= 0) return((BN_ULONG)0);
158 #ifndef OPENSSL_SMALL_FOOTPRINT
161 mul_add(rp[0],ap[0],bl,bh,c);
162 mul_add(rp[1],ap[1],bl,bh,c);
163 mul_add(rp[2],ap[2],bl,bh,c);
164 mul_add(rp[3],ap[3],bl,bh,c);
165 ap+=4; rp+=4; num-=4;
170 mul_add(rp[0],ap[0],bl,bh,c);
176 BN_ULONG bn_mul_words(BN_ULONG *rp, const BN_ULONG *ap, int num, BN_ULONG w)
182 if (num <= 0) return((BN_ULONG)0);
187 #ifndef OPENSSL_SMALL_FOOTPRINT
190 mul(rp[0],ap[0],bl,bh,carry);
191 mul(rp[1],ap[1],bl,bh,carry);
192 mul(rp[2],ap[2],bl,bh,carry);
193 mul(rp[3],ap[3],bl,bh,carry);
194 ap+=4; rp+=4; num-=4;
199 mul(rp[0],ap[0],bl,bh,carry);
205 void bn_sqr_words(BN_ULONG *r, const BN_ULONG *a, int n)
210 #ifndef OPENSSL_SMALL_FOOTPRINT
213 sqr64(r[0],r[1],a[0]);
214 sqr64(r[2],r[3],a[1]);
215 sqr64(r[4],r[5],a[2]);
216 sqr64(r[6],r[7],a[3]);
222 sqr64(r[0],r[1],a[0]);
227 #endif /* !(defined(BN_LLONG) || defined(BN_UMULT_HIGH)) */
229 #if defined(BN_LLONG) && defined(BN_DIV2W)
231 BN_ULONG bn_div_words(BN_ULONG h, BN_ULONG l, BN_ULONG d)
233 return((BN_ULONG)(((((BN_ULLONG)h)<<BN_BITS2)|l)/(BN_ULLONG)d));
238 /* Divide h,l by d and return the result. */
239 /* I need to test this some more :-( */
240 BN_ULONG bn_div_words(BN_ULONG h, BN_ULONG l, BN_ULONG d)
242 BN_ULONG dh,dl,q,ret=0,th,tl,t;
245 if (d == 0) return(BN_MASK2);
247 i=BN_num_bits_word(d);
248 assert((i == BN_BITS2) || (h <= (BN_ULONG)1<<i));
256 h=(h<<i)|(l>>(BN_BITS2-i));
259 dh=(d&BN_MASK2h)>>BN_BITS4;
263 if ((h>>BN_BITS4) == dh)
276 ((l&BN_MASK2h)>>BN_BITS4))))
283 tl=(tl<<BN_BITS4)&BN_MASK2h;
295 if (--count == 0) break;
298 h=((h<<BN_BITS4)|(l>>BN_BITS4))&BN_MASK2;
299 l=(l&BN_MASK2l)<<BN_BITS4;
304 #endif /* !defined(BN_LLONG) && defined(BN_DIV2W) */
307 BN_ULONG bn_add_words(BN_ULONG *r, const BN_ULONG *a, const BN_ULONG *b, int n)
312 if (n <= 0) return((BN_ULONG)0);
314 #ifndef OPENSSL_SMALL_FOOTPRINT
317 ll+=(BN_ULLONG)a[0]+b[0];
318 r[0]=(BN_ULONG)ll&BN_MASK2;
320 ll+=(BN_ULLONG)a[1]+b[1];
321 r[1]=(BN_ULONG)ll&BN_MASK2;
323 ll+=(BN_ULLONG)a[2]+b[2];
324 r[2]=(BN_ULONG)ll&BN_MASK2;
326 ll+=(BN_ULLONG)a[3]+b[3];
327 r[3]=(BN_ULONG)ll&BN_MASK2;
329 a+=4; b+=4; r+=4; n-=4;
334 ll+=(BN_ULLONG)a[0]+b[0];
335 r[0]=(BN_ULONG)ll&BN_MASK2;
339 return((BN_ULONG)ll);
341 #else /* !BN_LLONG */
342 BN_ULONG bn_add_words(BN_ULONG *r, const BN_ULONG *a, const BN_ULONG *b, int n)
347 if (n <= 0) return((BN_ULONG)0);
350 #ifndef OPENSSL_SMALL_FOOTPRINT
377 a+=4; b+=4; r+=4; n-=4;
392 #endif /* !BN_LLONG */
394 BN_ULONG bn_sub_words(BN_ULONG *r, const BN_ULONG *a, const BN_ULONG *b, int n)
400 if (n <= 0) return((BN_ULONG)0);
402 #ifndef OPENSSL_SMALL_FOOTPRINT
406 r[0]=(t1-t2-c)&BN_MASK2;
407 if (t1 != t2) c=(t1 < t2);
409 r[1]=(t1-t2-c)&BN_MASK2;
410 if (t1 != t2) c=(t1 < t2);
412 r[2]=(t1-t2-c)&BN_MASK2;
413 if (t1 != t2) c=(t1 < t2);
415 r[3]=(t1-t2-c)&BN_MASK2;
416 if (t1 != t2) c=(t1 < t2);
417 a+=4; b+=4; r+=4; n-=4;
423 r[0]=(t1-t2-c)&BN_MASK2;
424 if (t1 != t2) c=(t1 < t2);
430 #if defined(BN_MUL_COMBA) && !defined(OPENSSL_SMALL_FOOTPRINT)
432 #ifndef OPENSSL_FIPSCANISTER
439 /* mul_add_c(a,b,c0,c1,c2) -- c+=a*b for three word number c=(c2,c1,c0) */
440 /* mul_add_c2(a,b,c0,c1,c2) -- c+=2*a*b for three word number c=(c2,c1,c0) */
441 /* sqr_add_c(a,i,c0,c1,c2) -- c+=a[i]^2 for three word number c=(c2,c1,c0) */
442 /* sqr_add_c2(a,i,c0,c1,c2) -- c+=2*a[i]*a[j] for three word number c=(c2,c1,c0) */
445 #define mul_add_c(a,b,c0,c1,c2) \
447 t1=(BN_ULONG)Lw(t); \
448 t2=(BN_ULONG)Hw(t); \
449 c0=(c0+t1)&BN_MASK2; if ((c0) < t1) t2++; \
450 c1=(c1+t2)&BN_MASK2; if ((c1) < t2) c2++;
452 #define mul_add_c2(a,b,c0,c1,c2) \
456 t1=(BN_ULONG)Lw(tt); \
457 t2=(BN_ULONG)Hw(tt); \
458 c0=(c0+t1)&BN_MASK2; \
459 if ((c0 < t1) && (((++t2)&BN_MASK2) == 0)) c2++; \
460 c1=(c1+t2)&BN_MASK2; if ((c1) < t2) c2++;
462 #define sqr_add_c(a,i,c0,c1,c2) \
463 t=(BN_ULLONG)a[i]*a[i]; \
464 t1=(BN_ULONG)Lw(t); \
465 t2=(BN_ULONG)Hw(t); \
466 c0=(c0+t1)&BN_MASK2; if ((c0) < t1) t2++; \
467 c1=(c1+t2)&BN_MASK2; if ((c1) < t2) c2++;
469 #define sqr_add_c2(a,i,j,c0,c1,c2) \
470 mul_add_c2((a)[i],(a)[j],c0,c1,c2)
472 #elif defined(BN_UMULT_LOHI)
474 #define mul_add_c(a,b,c0,c1,c2) { \
475 BN_ULONG ta=(a),tb=(b); \
476 BN_UMULT_LOHI(t1,t2,ta,tb); \
477 c0 += t1; t2 += (c0<t1)?1:0; \
478 c1 += t2; c2 += (c1<t2)?1:0; \
481 #define mul_add_c2(a,b,c0,c1,c2) { \
482 BN_ULONG ta=(a),tb=(b),t0; \
483 BN_UMULT_LOHI(t0,t1,ta,tb); \
484 t2 = t1+t1; c2 += (t2<t1)?1:0; \
485 t1 = t0+t0; t2 += (t1<t0)?1:0; \
486 c0 += t1; t2 += (c0<t1)?1:0; \
487 c1 += t2; c2 += (c1<t2)?1:0; \
490 #define sqr_add_c(a,i,c0,c1,c2) { \
491 BN_ULONG ta=(a)[i]; \
492 BN_UMULT_LOHI(t1,t2,ta,ta); \
493 c0 += t1; t2 += (c0<t1)?1:0; \
494 c1 += t2; c2 += (c1<t2)?1:0; \
497 #define sqr_add_c2(a,i,j,c0,c1,c2) \
498 mul_add_c2((a)[i],(a)[j],c0,c1,c2)
500 #elif defined(BN_UMULT_HIGH)
502 #define mul_add_c(a,b,c0,c1,c2) { \
503 BN_ULONG ta=(a),tb=(b); \
505 t2 = BN_UMULT_HIGH(ta,tb); \
506 c0 += t1; t2 += (c0<t1)?1:0; \
507 c1 += t2; c2 += (c1<t2)?1:0; \
510 #define mul_add_c2(a,b,c0,c1,c2) { \
511 BN_ULONG ta=(a),tb=(b),t0; \
512 t1 = BN_UMULT_HIGH(ta,tb); \
514 t2 = t1+t1; c2 += (t2<t1)?1:0; \
515 t1 = t0+t0; t2 += (t1<t0)?1:0; \
516 c0 += t1; t2 += (c0<t1)?1:0; \
517 c1 += t2; c2 += (c1<t2)?1:0; \
520 #define sqr_add_c(a,i,c0,c1,c2) { \
521 BN_ULONG ta=(a)[i]; \
523 t2 = BN_UMULT_HIGH(ta,ta); \
524 c0 += t1; t2 += (c0<t1)?1:0; \
525 c1 += t2; c2 += (c1<t2)?1:0; \
528 #define sqr_add_c2(a,i,j,c0,c1,c2) \
529 mul_add_c2((a)[i],(a)[j],c0,c1,c2)
531 #else /* !BN_LLONG */
532 #define mul_add_c(a,b,c0,c1,c2) \
533 t1=LBITS(a); t2=HBITS(a); \
534 bl=LBITS(b); bh=HBITS(b); \
535 mul64(t1,t2,bl,bh); \
536 c0=(c0+t1)&BN_MASK2; if ((c0) < t1) t2++; \
537 c1=(c1+t2)&BN_MASK2; if ((c1) < t2) c2++;
539 #define mul_add_c2(a,b,c0,c1,c2) \
540 t1=LBITS(a); t2=HBITS(a); \
541 bl=LBITS(b); bh=HBITS(b); \
542 mul64(t1,t2,bl,bh); \
543 if (t2 & BN_TBIT) c2++; \
544 t2=(t2+t2)&BN_MASK2; \
545 if (t1 & BN_TBIT) t2++; \
546 t1=(t1+t1)&BN_MASK2; \
547 c0=(c0+t1)&BN_MASK2; \
548 if ((c0 < t1) && (((++t2)&BN_MASK2) == 0)) c2++; \
549 c1=(c1+t2)&BN_MASK2; if ((c1) < t2) c2++;
551 #define sqr_add_c(a,i,c0,c1,c2) \
552 sqr64(t1,t2,(a)[i]); \
553 c0=(c0+t1)&BN_MASK2; if ((c0) < t1) t2++; \
554 c1=(c1+t2)&BN_MASK2; if ((c1) < t2) c2++;
556 #define sqr_add_c2(a,i,j,c0,c1,c2) \
557 mul_add_c2((a)[i],(a)[j],c0,c1,c2)
558 #endif /* !BN_LLONG */
560 void bn_mul_comba8(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b)
573 mul_add_c(a[0],b[0],c1,c2,c3);
576 mul_add_c(a[0],b[1],c2,c3,c1);
577 mul_add_c(a[1],b[0],c2,c3,c1);
580 mul_add_c(a[2],b[0],c3,c1,c2);
581 mul_add_c(a[1],b[1],c3,c1,c2);
582 mul_add_c(a[0],b[2],c3,c1,c2);
585 mul_add_c(a[0],b[3],c1,c2,c3);
586 mul_add_c(a[1],b[2],c1,c2,c3);
587 mul_add_c(a[2],b[1],c1,c2,c3);
588 mul_add_c(a[3],b[0],c1,c2,c3);
591 mul_add_c(a[4],b[0],c2,c3,c1);
592 mul_add_c(a[3],b[1],c2,c3,c1);
593 mul_add_c(a[2],b[2],c2,c3,c1);
594 mul_add_c(a[1],b[3],c2,c3,c1);
595 mul_add_c(a[0],b[4],c2,c3,c1);
598 mul_add_c(a[0],b[5],c3,c1,c2);
599 mul_add_c(a[1],b[4],c3,c1,c2);
600 mul_add_c(a[2],b[3],c3,c1,c2);
601 mul_add_c(a[3],b[2],c3,c1,c2);
602 mul_add_c(a[4],b[1],c3,c1,c2);
603 mul_add_c(a[5],b[0],c3,c1,c2);
606 mul_add_c(a[6],b[0],c1,c2,c3);
607 mul_add_c(a[5],b[1],c1,c2,c3);
608 mul_add_c(a[4],b[2],c1,c2,c3);
609 mul_add_c(a[3],b[3],c1,c2,c3);
610 mul_add_c(a[2],b[4],c1,c2,c3);
611 mul_add_c(a[1],b[5],c1,c2,c3);
612 mul_add_c(a[0],b[6],c1,c2,c3);
615 mul_add_c(a[0],b[7],c2,c3,c1);
616 mul_add_c(a[1],b[6],c2,c3,c1);
617 mul_add_c(a[2],b[5],c2,c3,c1);
618 mul_add_c(a[3],b[4],c2,c3,c1);
619 mul_add_c(a[4],b[3],c2,c3,c1);
620 mul_add_c(a[5],b[2],c2,c3,c1);
621 mul_add_c(a[6],b[1],c2,c3,c1);
622 mul_add_c(a[7],b[0],c2,c3,c1);
625 mul_add_c(a[7],b[1],c3,c1,c2);
626 mul_add_c(a[6],b[2],c3,c1,c2);
627 mul_add_c(a[5],b[3],c3,c1,c2);
628 mul_add_c(a[4],b[4],c3,c1,c2);
629 mul_add_c(a[3],b[5],c3,c1,c2);
630 mul_add_c(a[2],b[6],c3,c1,c2);
631 mul_add_c(a[1],b[7],c3,c1,c2);
634 mul_add_c(a[2],b[7],c1,c2,c3);
635 mul_add_c(a[3],b[6],c1,c2,c3);
636 mul_add_c(a[4],b[5],c1,c2,c3);
637 mul_add_c(a[5],b[4],c1,c2,c3);
638 mul_add_c(a[6],b[3],c1,c2,c3);
639 mul_add_c(a[7],b[2],c1,c2,c3);
642 mul_add_c(a[7],b[3],c2,c3,c1);
643 mul_add_c(a[6],b[4],c2,c3,c1);
644 mul_add_c(a[5],b[5],c2,c3,c1);
645 mul_add_c(a[4],b[6],c2,c3,c1);
646 mul_add_c(a[3],b[7],c2,c3,c1);
649 mul_add_c(a[4],b[7],c3,c1,c2);
650 mul_add_c(a[5],b[6],c3,c1,c2);
651 mul_add_c(a[6],b[5],c3,c1,c2);
652 mul_add_c(a[7],b[4],c3,c1,c2);
655 mul_add_c(a[7],b[5],c1,c2,c3);
656 mul_add_c(a[6],b[6],c1,c2,c3);
657 mul_add_c(a[5],b[7],c1,c2,c3);
660 mul_add_c(a[6],b[7],c2,c3,c1);
661 mul_add_c(a[7],b[6],c2,c3,c1);
664 mul_add_c(a[7],b[7],c3,c1,c2);
669 void bn_mul_comba4(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b)
682 mul_add_c(a[0],b[0],c1,c2,c3);
685 mul_add_c(a[0],b[1],c2,c3,c1);
686 mul_add_c(a[1],b[0],c2,c3,c1);
689 mul_add_c(a[2],b[0],c3,c1,c2);
690 mul_add_c(a[1],b[1],c3,c1,c2);
691 mul_add_c(a[0],b[2],c3,c1,c2);
694 mul_add_c(a[0],b[3],c1,c2,c3);
695 mul_add_c(a[1],b[2],c1,c2,c3);
696 mul_add_c(a[2],b[1],c1,c2,c3);
697 mul_add_c(a[3],b[0],c1,c2,c3);
700 mul_add_c(a[3],b[1],c2,c3,c1);
701 mul_add_c(a[2],b[2],c2,c3,c1);
702 mul_add_c(a[1],b[3],c2,c3,c1);
705 mul_add_c(a[2],b[3],c3,c1,c2);
706 mul_add_c(a[3],b[2],c3,c1,c2);
709 mul_add_c(a[3],b[3],c1,c2,c3);
714 void bn_sqr_comba8(BN_ULONG *r, const BN_ULONG *a)
727 sqr_add_c(a,0,c1,c2,c3);
730 sqr_add_c2(a,1,0,c2,c3,c1);
733 sqr_add_c(a,1,c3,c1,c2);
734 sqr_add_c2(a,2,0,c3,c1,c2);
737 sqr_add_c2(a,3,0,c1,c2,c3);
738 sqr_add_c2(a,2,1,c1,c2,c3);
741 sqr_add_c(a,2,c2,c3,c1);
742 sqr_add_c2(a,3,1,c2,c3,c1);
743 sqr_add_c2(a,4,0,c2,c3,c1);
746 sqr_add_c2(a,5,0,c3,c1,c2);
747 sqr_add_c2(a,4,1,c3,c1,c2);
748 sqr_add_c2(a,3,2,c3,c1,c2);
751 sqr_add_c(a,3,c1,c2,c3);
752 sqr_add_c2(a,4,2,c1,c2,c3);
753 sqr_add_c2(a,5,1,c1,c2,c3);
754 sqr_add_c2(a,6,0,c1,c2,c3);
757 sqr_add_c2(a,7,0,c2,c3,c1);
758 sqr_add_c2(a,6,1,c2,c3,c1);
759 sqr_add_c2(a,5,2,c2,c3,c1);
760 sqr_add_c2(a,4,3,c2,c3,c1);
763 sqr_add_c(a,4,c3,c1,c2);
764 sqr_add_c2(a,5,3,c3,c1,c2);
765 sqr_add_c2(a,6,2,c3,c1,c2);
766 sqr_add_c2(a,7,1,c3,c1,c2);
769 sqr_add_c2(a,7,2,c1,c2,c3);
770 sqr_add_c2(a,6,3,c1,c2,c3);
771 sqr_add_c2(a,5,4,c1,c2,c3);
774 sqr_add_c(a,5,c2,c3,c1);
775 sqr_add_c2(a,6,4,c2,c3,c1);
776 sqr_add_c2(a,7,3,c2,c3,c1);
779 sqr_add_c2(a,7,4,c3,c1,c2);
780 sqr_add_c2(a,6,5,c3,c1,c2);
783 sqr_add_c(a,6,c1,c2,c3);
784 sqr_add_c2(a,7,5,c1,c2,c3);
787 sqr_add_c2(a,7,6,c2,c3,c1);
790 sqr_add_c(a,7,c3,c1,c2);
795 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? */
952 #ifndef OPENSSL_FIPSCANISTER
956 void bn_sqr_comba4(BN_ULONG *r, const BN_ULONG *a)
959 bn_sqr_normal(r,a,4,t);
962 void bn_sqr_comba8(BN_ULONG *r, const BN_ULONG *a)
965 bn_sqr_normal(r,a,8,t);
968 void bn_mul_comba4(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b)
970 r[4]=bn_mul_words( &(r[0]),a,4,b[0]);
971 r[5]=bn_mul_add_words(&(r[1]),a,4,b[1]);
972 r[6]=bn_mul_add_words(&(r[2]),a,4,b[2]);
973 r[7]=bn_mul_add_words(&(r[3]),a,4,b[3]);
976 void bn_mul_comba8(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b)
978 r[ 8]=bn_mul_words( &(r[0]),a,8,b[0]);
979 r[ 9]=bn_mul_add_words(&(r[1]),a,8,b[1]);
980 r[10]=bn_mul_add_words(&(r[2]),a,8,b[2]);
981 r[11]=bn_mul_add_words(&(r[3]),a,8,b[3]);
982 r[12]=bn_mul_add_words(&(r[4]),a,8,b[4]);
983 r[13]=bn_mul_add_words(&(r[5]),a,8,b[5]);
984 r[14]=bn_mul_add_words(&(r[6]),a,8,b[6]);
985 r[15]=bn_mul_add_words(&(r[7]),a,8,b[7]);
988 #ifdef OPENSSL_NO_ASM
989 #ifdef OPENSSL_BN_ASM_MONT
991 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)
993 BN_ULONG c0,c1,*tp,n0=*n0p;
994 volatile BN_ULONG *vp;
997 vp = tp = alloca((num+2)*sizeof(BN_ULONG));
999 for(i=0;i<=num;i++) tp[i]=0;
1003 c0 = bn_mul_add_words(tp,ap,num,bp[i]);
1004 c1 = (tp[num] + c0)&BN_MASK2;
1006 tp[num+1] = (c1<c0?1:0);
1008 c0 = bn_mul_add_words(tp,np,num,tp[0]*n0);
1009 c1 = (tp[num] + c0)&BN_MASK2;
1011 tp[num+1] += (c1<c0?1:0);
1012 for(j=0;j<=num;j++) tp[j]=tp[j+1];
1015 if (tp[num]!=0 || tp[num-1]>=np[num-1])
1017 c0 = bn_sub_words(rp,tp,np,num);
1018 if (tp[num]!=0 || c0==0)
1020 for(i=0;i<num+2;i++) vp[i] = 0;
1024 for(i=0;i<num;i++) rp[i] = tp[i], vp[i] = 0;
1030 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)
1032 #endif /* OPENSSL_BN_ASM_MONT */
1035 #endif /* !BN_MUL_COMBA */