Leave the decision to call/implement bn_sqr_mont to assembler developer.
[openssl.git] / crypto / bn / bn_asm.c
1 /* crypto/bn/bn_asm.c */
2 /* Copyright (C) 1995-1998 Eric Young (eay@cryptsoft.com)
3  * All rights reserved.
4  *
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.
8  * 
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).
15  * 
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.
22  * 
23  * Redistribution and use in source and binary forms, with or without
24  * modification, are permitted provided that the following conditions
25  * are met:
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)"
40  * 
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
51  * SUCH DAMAGE.
52  * 
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.]
57  */
58
59 #ifndef BN_DEBUG
60 # undef NDEBUG /* avoid conflicting definitions */
61 # define NDEBUG
62 #endif
63
64 #include <stdio.h>
65 #include <assert.h>
66 #include "cryptlib.h"
67 #include "bn_lcl.h"
68
69 #if defined(BN_LLONG) || defined(BN_UMULT_HIGH)
70
71 BN_ULONG bn_mul_add_words(BN_ULONG *rp, const BN_ULONG *ap, int num, BN_ULONG w)
72         {
73         BN_ULONG c1=0;
74
75         assert(num >= 0);
76         if (num <= 0) return(c1);
77
78 #ifndef OPENSSL_SMALL_FOOTPRINT
79         while (num&~3)
80                 {
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);
85                 ap+=4; rp+=4; num-=4;
86                 }
87 #endif
88         while (num)
89                 {
90                 mul_add(rp[0],ap[0],w,c1);
91                 ap++; rp++; num--;
92                 }
93         
94         return(c1);
95         } 
96
97 BN_ULONG bn_mul_words(BN_ULONG *rp, const BN_ULONG *ap, int num, BN_ULONG w)
98         {
99         BN_ULONG c1=0;
100
101         assert(num >= 0);
102         if (num <= 0) return(c1);
103
104 #ifndef OPENSSL_SMALL_FOOTPRINT
105         while (num&~3)
106                 {
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;
112                 }
113 #endif
114         while (num)
115                 {
116                 mul(rp[0],ap[0],w,c1);
117                 ap++; rp++; num--;
118                 }
119         return(c1);
120         } 
121
122 void bn_sqr_words(BN_ULONG *r, const BN_ULONG *a, int n)
123         {
124         assert(n >= 0);
125         if (n <= 0) return;
126
127 #ifndef OPENSSL_SMALL_FOOTPRINT
128         while (n&~3)
129                 {
130                 sqr(r[0],r[1],a[0]);
131                 sqr(r[2],r[3],a[1]);
132                 sqr(r[4],r[5],a[2]);
133                 sqr(r[6],r[7],a[3]);
134                 a+=4; r+=8; n-=4;
135                 }
136 #endif
137         while (n)
138                 {
139                 sqr(r[0],r[1],a[0]);
140                 a++; r+=2; n--;
141                 }
142         }
143
144 #else /* !(defined(BN_LLONG) || defined(BN_UMULT_HIGH)) */
145
146 BN_ULONG bn_mul_add_words(BN_ULONG *rp, const BN_ULONG *ap, int num, BN_ULONG w)
147         {
148         BN_ULONG c=0;
149         BN_ULONG bl,bh;
150
151         assert(num >= 0);
152         if (num <= 0) return((BN_ULONG)0);
153
154         bl=LBITS(w);
155         bh=HBITS(w);
156
157 #ifndef OPENSSL_SMALL_FOOTPRINT
158         while (num&~3)
159                 {
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;
165                 }
166 #endif
167         while (num)
168                 {
169                 mul_add(rp[0],ap[0],bl,bh,c);
170                 ap++; rp++; num--;
171                 }
172         return(c);
173         } 
174
175 BN_ULONG bn_mul_words(BN_ULONG *rp, const BN_ULONG *ap, int num, BN_ULONG w)
176         {
177         BN_ULONG carry=0;
178         BN_ULONG bl,bh;
179
180         assert(num >= 0);
181         if (num <= 0) return((BN_ULONG)0);
182
183         bl=LBITS(w);
184         bh=HBITS(w);
185
186 #ifndef OPENSSL_SMALL_FOOTPRINT
187         while (num&~3)
188                 {
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;
194                 }
195 #endif
196         while (num)
197                 {
198                 mul(rp[0],ap[0],bl,bh,carry);
199                 ap++; rp++; num--;
200                 }
201         return(carry);
202         } 
203
204 void bn_sqr_words(BN_ULONG *r, const BN_ULONG *a, int n)
205         {
206         assert(n >= 0);
207         if (n <= 0) return;
208
209 #ifndef OPENSSL_SMALL_FOOTPRINT
210         while (n&~3)
211                 {
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]);
216                 a+=4; r+=8; n-=4;
217                 }
218 #endif
219         while (n)
220                 {
221                 sqr64(r[0],r[1],a[0]);
222                 a++; r+=2; n--;
223                 }
224         }
225
226 #endif /* !(defined(BN_LLONG) || defined(BN_UMULT_HIGH)) */
227
228 #if defined(BN_LLONG) && defined(BN_DIV2W)
229
230 BN_ULONG bn_div_words(BN_ULONG h, BN_ULONG l, BN_ULONG d)
231         {
232         return((BN_ULONG)(((((BN_ULLONG)h)<<BN_BITS2)|l)/(BN_ULLONG)d));
233         }
234
235 #else
236
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)
240         {
241         BN_ULONG dh,dl,q,ret=0,th,tl,t;
242         int i,count=2;
243
244         if (d == 0) return(BN_MASK2);
245
246         i=BN_num_bits_word(d);
247         assert((i == BN_BITS2) || (h <= (BN_ULONG)1<<i));
248
249         i=BN_BITS2-i;
250         if (h >= d) h-=d;
251
252         if (i)
253                 {
254                 d<<=i;
255                 h=(h<<i)|(l>>(BN_BITS2-i));
256                 l<<=i;
257                 }
258         dh=(d&BN_MASK2h)>>BN_BITS4;
259         dl=(d&BN_MASK2l);
260         for (;;)
261                 {
262                 if ((h>>BN_BITS4) == dh)
263                         q=BN_MASK2l;
264                 else
265                         q=h/dh;
266
267                 th=q*dh;
268                 tl=dl*q;
269                 for (;;)
270                         {
271                         t=h-th;
272                         if ((t&BN_MASK2h) ||
273                                 ((tl) <= (
274                                         (t<<BN_BITS4)|
275                                         ((l&BN_MASK2h)>>BN_BITS4))))
276                                 break;
277                         q--;
278                         th-=dh;
279                         tl-=dl;
280                         }
281                 t=(tl>>BN_BITS4);
282                 tl=(tl<<BN_BITS4)&BN_MASK2h;
283                 th+=t;
284
285                 if (l < tl) th++;
286                 l-=tl;
287                 if (h < th)
288                         {
289                         h+=d;
290                         q--;
291                         }
292                 h-=th;
293
294                 if (--count == 0) break;
295
296                 ret=q<<BN_BITS4;
297                 h=((h<<BN_BITS4)|(l>>BN_BITS4))&BN_MASK2;
298                 l=(l&BN_MASK2l)<<BN_BITS4;
299                 }
300         ret|=q;
301         return(ret);
302         }
303 #endif /* !defined(BN_LLONG) && defined(BN_DIV2W) */
304
305 #ifdef BN_LLONG
306 BN_ULONG bn_add_words(BN_ULONG *r, const BN_ULONG *a, const BN_ULONG *b, int n)
307         {
308         BN_ULLONG ll=0;
309
310         assert(n >= 0);
311         if (n <= 0) return((BN_ULONG)0);
312
313 #ifndef OPENSSL_SMALL_FOOTPRINT
314         while (n&~3)
315                 {
316                 ll+=(BN_ULLONG)a[0]+b[0];
317                 r[0]=(BN_ULONG)ll&BN_MASK2;
318                 ll>>=BN_BITS2;
319                 ll+=(BN_ULLONG)a[1]+b[1];
320                 r[1]=(BN_ULONG)ll&BN_MASK2;
321                 ll>>=BN_BITS2;
322                 ll+=(BN_ULLONG)a[2]+b[2];
323                 r[2]=(BN_ULONG)ll&BN_MASK2;
324                 ll>>=BN_BITS2;
325                 ll+=(BN_ULLONG)a[3]+b[3];
326                 r[3]=(BN_ULONG)ll&BN_MASK2;
327                 ll>>=BN_BITS2;
328                 a+=4; b+=4; r+=4; n-=4;
329                 }
330 #endif
331         while (n)
332                 {
333                 ll+=(BN_ULLONG)a[0]+b[0];
334                 r[0]=(BN_ULONG)ll&BN_MASK2;
335                 ll>>=BN_BITS2;
336                 a++; b++; r++; n--;
337                 }
338         return((BN_ULONG)ll);
339         }
340 #else /* !BN_LLONG */
341 BN_ULONG bn_add_words(BN_ULONG *r, const BN_ULONG *a, const BN_ULONG *b, int n)
342         {
343         BN_ULONG c,l,t;
344
345         assert(n >= 0);
346         if (n <= 0) return((BN_ULONG)0);
347
348         c=0;
349 #ifndef OPENSSL_SMALL_FOOTPRINT
350         while (n&~3)
351                 {
352                 t=a[0];
353                 t=(t+c)&BN_MASK2;
354                 c=(t < c);
355                 l=(t+b[0])&BN_MASK2;
356                 c+=(l < t);
357                 r[0]=l;
358                 t=a[1];
359                 t=(t+c)&BN_MASK2;
360                 c=(t < c);
361                 l=(t+b[1])&BN_MASK2;
362                 c+=(l < t);
363                 r[1]=l;
364                 t=a[2];
365                 t=(t+c)&BN_MASK2;
366                 c=(t < c);
367                 l=(t+b[2])&BN_MASK2;
368                 c+=(l < t);
369                 r[2]=l;
370                 t=a[3];
371                 t=(t+c)&BN_MASK2;
372                 c=(t < c);
373                 l=(t+b[3])&BN_MASK2;
374                 c+=(l < t);
375                 r[3]=l;
376                 a+=4; b+=4; r+=4; n-=4;
377                 }
378 #endif
379         while(n)
380                 {
381                 t=a[0];
382                 t=(t+c)&BN_MASK2;
383                 c=(t < c);
384                 l=(t+b[0])&BN_MASK2;
385                 c+=(l < t);
386                 r[0]=l;
387                 a++; b++; r++; n--;
388                 }
389         return((BN_ULONG)c);
390         }
391 #endif /* !BN_LLONG */
392
393 BN_ULONG bn_sub_words(BN_ULONG *r, const BN_ULONG *a, const BN_ULONG *b, int n)
394         {
395         BN_ULONG t1,t2;
396         int c=0;
397
398         assert(n >= 0);
399         if (n <= 0) return((BN_ULONG)0);
400
401 #ifndef OPENSSL_SMALL_FOOTPRINT
402         while (n&~3)
403                 {
404                 t1=a[0]; t2=b[0];
405                 r[0]=(t1-t2-c)&BN_MASK2;
406                 if (t1 != t2) c=(t1 < t2);
407                 t1=a[1]; t2=b[1];
408                 r[1]=(t1-t2-c)&BN_MASK2;
409                 if (t1 != t2) c=(t1 < t2);
410                 t1=a[2]; t2=b[2];
411                 r[2]=(t1-t2-c)&BN_MASK2;
412                 if (t1 != t2) c=(t1 < t2);
413                 t1=a[3]; t2=b[3];
414                 r[3]=(t1-t2-c)&BN_MASK2;
415                 if (t1 != t2) c=(t1 < t2);
416                 a+=4; b+=4; r+=4; n-=4;
417                 }
418 #endif
419         while (n)
420                 {
421                 t1=a[0]; t2=b[0];
422                 r[0]=(t1-t2-c)&BN_MASK2;
423                 if (t1 != t2) c=(t1 < t2);
424                 a++; b++; r++; n--;
425                 }
426         return(c);
427         }
428
429 #if defined(BN_MUL_COMBA) && !defined(OPENSSL_SMALL_FOOTPRINT)
430
431 #undef bn_mul_comba8
432 #undef bn_mul_comba4
433 #undef bn_sqr_comba8
434 #undef bn_sqr_comba4
435
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) */
440
441 #ifdef BN_LLONG
442 #define mul_add_c(a,b,c0,c1,c2) \
443         t=(BN_ULLONG)a*b; \
444         t1=(BN_ULONG)Lw(t); \
445         t2=(BN_ULONG)Hw(t); \
446         c0=(c0+t1)&BN_MASK2; if ((c0) < t1) t2++; \
447         c1=(c1+t2)&BN_MASK2; if ((c1) < t2) c2++;
448
449 #define mul_add_c2(a,b,c0,c1,c2) \
450         t=(BN_ULLONG)a*b; \
451         tt=(t+t)&BN_MASK; \
452         if (tt < t) c2++; \
453         t1=(BN_ULONG)Lw(tt); \
454         t2=(BN_ULONG)Hw(tt); \
455         c0=(c0+t1)&BN_MASK2;  \
456         if ((c0 < t1) && (((++t2)&BN_MASK2) == 0)) c2++; \
457         c1=(c1+t2)&BN_MASK2; if ((c1) < t2) c2++;
458
459 #define sqr_add_c(a,i,c0,c1,c2) \
460         t=(BN_ULLONG)a[i]*a[i]; \
461         t1=(BN_ULONG)Lw(t); \
462         t2=(BN_ULONG)Hw(t); \
463         c0=(c0+t1)&BN_MASK2; if ((c0) < t1) t2++; \
464         c1=(c1+t2)&BN_MASK2; if ((c1) < t2) c2++;
465
466 #define sqr_add_c2(a,i,j,c0,c1,c2) \
467         mul_add_c2((a)[i],(a)[j],c0,c1,c2)
468
469 #elif defined(BN_UMULT_LOHI)
470
471 #define mul_add_c(a,b,c0,c1,c2) {       \
472         BN_ULONG ta=(a),tb=(b);         \
473         BN_UMULT_LOHI(t1,t2,ta,tb);     \
474         c0 += t1; t2 += (c0<t1)?1:0;    \
475         c1 += t2; c2 += (c1<t2)?1:0;    \
476         }
477
478 #define mul_add_c2(a,b,c0,c1,c2) {      \
479         BN_ULONG ta=(a),tb=(b),t0;      \
480         BN_UMULT_LOHI(t0,t1,ta,tb);     \
481         t2 = t1+t1; c2 += (t2<t1)?1:0;  \
482         t1 = t0+t0; t2 += (t1<t0)?1:0;  \
483         c0 += t1; t2 += (c0<t1)?1:0;    \
484         c1 += t2; c2 += (c1<t2)?1:0;    \
485         }
486
487 #define sqr_add_c(a,i,c0,c1,c2) {       \
488         BN_ULONG ta=(a)[i];             \
489         BN_UMULT_LOHI(t1,t2,ta,ta);     \
490         c0 += t1; t2 += (c0<t1)?1:0;    \
491         c1 += t2; c2 += (c1<t2)?1:0;    \
492         }
493
494 #define sqr_add_c2(a,i,j,c0,c1,c2)      \
495         mul_add_c2((a)[i],(a)[j],c0,c1,c2)
496
497 #elif defined(BN_UMULT_HIGH)
498
499 #define mul_add_c(a,b,c0,c1,c2) {       \
500         BN_ULONG ta=(a),tb=(b);         \
501         t1 = ta * tb;                   \
502         t2 = BN_UMULT_HIGH(ta,tb);      \
503         c0 += t1; t2 += (c0<t1)?1:0;    \
504         c1 += t2; c2 += (c1<t2)?1:0;    \
505         }
506
507 #define mul_add_c2(a,b,c0,c1,c2) {      \
508         BN_ULONG ta=(a),tb=(b),t0;      \
509         t1 = BN_UMULT_HIGH(ta,tb);      \
510         t0 = ta * tb;                   \
511         t2 = t1+t1; c2 += (t2<t1)?1:0;  \
512         t1 = t0+t0; t2 += (t1<t0)?1:0;  \
513         c0 += t1; t2 += (c0<t1)?1:0;    \
514         c1 += t2; c2 += (c1<t2)?1:0;    \
515         }
516
517 #define sqr_add_c(a,i,c0,c1,c2) {       \
518         BN_ULONG ta=(a)[i];             \
519         t1 = ta * ta;                   \
520         t2 = BN_UMULT_HIGH(ta,ta);      \
521         c0 += t1; t2 += (c0<t1)?1:0;    \
522         c1 += t2; c2 += (c1<t2)?1:0;    \
523         }
524
525 #define sqr_add_c2(a,i,j,c0,c1,c2)      \
526         mul_add_c2((a)[i],(a)[j],c0,c1,c2)
527
528 #else /* !BN_LLONG */
529 #define mul_add_c(a,b,c0,c1,c2) \
530         t1=LBITS(a); t2=HBITS(a); \
531         bl=LBITS(b); bh=HBITS(b); \
532         mul64(t1,t2,bl,bh); \
533         c0=(c0+t1)&BN_MASK2; if ((c0) < t1) t2++; \
534         c1=(c1+t2)&BN_MASK2; if ((c1) < t2) c2++;
535
536 #define mul_add_c2(a,b,c0,c1,c2) \
537         t1=LBITS(a); t2=HBITS(a); \
538         bl=LBITS(b); bh=HBITS(b); \
539         mul64(t1,t2,bl,bh); \
540         if (t2 & BN_TBIT) c2++; \
541         t2=(t2+t2)&BN_MASK2; \
542         if (t1 & BN_TBIT) t2++; \
543         t1=(t1+t1)&BN_MASK2; \
544         c0=(c0+t1)&BN_MASK2;  \
545         if ((c0 < t1) && (((++t2)&BN_MASK2) == 0)) c2++; \
546         c1=(c1+t2)&BN_MASK2; if ((c1) < t2) c2++;
547
548 #define sqr_add_c(a,i,c0,c1,c2) \
549         sqr64(t1,t2,(a)[i]); \
550         c0=(c0+t1)&BN_MASK2; if ((c0) < t1) t2++; \
551         c1=(c1+t2)&BN_MASK2; if ((c1) < t2) c2++;
552
553 #define sqr_add_c2(a,i,j,c0,c1,c2) \
554         mul_add_c2((a)[i],(a)[j],c0,c1,c2)
555 #endif /* !BN_LLONG */
556
557 void bn_mul_comba8(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b)
558         {
559 #ifdef BN_LLONG
560         BN_ULLONG t;
561 #else
562         BN_ULONG bl,bh;
563 #endif
564         BN_ULONG t1,t2;
565         BN_ULONG c1,c2,c3;
566
567         c1=0;
568         c2=0;
569         c3=0;
570         mul_add_c(a[0],b[0],c1,c2,c3);
571         r[0]=c1;
572         c1=0;
573         mul_add_c(a[0],b[1],c2,c3,c1);
574         mul_add_c(a[1],b[0],c2,c3,c1);
575         r[1]=c2;
576         c2=0;
577         mul_add_c(a[2],b[0],c3,c1,c2);
578         mul_add_c(a[1],b[1],c3,c1,c2);
579         mul_add_c(a[0],b[2],c3,c1,c2);
580         r[2]=c3;
581         c3=0;
582         mul_add_c(a[0],b[3],c1,c2,c3);
583         mul_add_c(a[1],b[2],c1,c2,c3);
584         mul_add_c(a[2],b[1],c1,c2,c3);
585         mul_add_c(a[3],b[0],c1,c2,c3);
586         r[3]=c1;
587         c1=0;
588         mul_add_c(a[4],b[0],c2,c3,c1);
589         mul_add_c(a[3],b[1],c2,c3,c1);
590         mul_add_c(a[2],b[2],c2,c3,c1);
591         mul_add_c(a[1],b[3],c2,c3,c1);
592         mul_add_c(a[0],b[4],c2,c3,c1);
593         r[4]=c2;
594         c2=0;
595         mul_add_c(a[0],b[5],c3,c1,c2);
596         mul_add_c(a[1],b[4],c3,c1,c2);
597         mul_add_c(a[2],b[3],c3,c1,c2);
598         mul_add_c(a[3],b[2],c3,c1,c2);
599         mul_add_c(a[4],b[1],c3,c1,c2);
600         mul_add_c(a[5],b[0],c3,c1,c2);
601         r[5]=c3;
602         c3=0;
603         mul_add_c(a[6],b[0],c1,c2,c3);
604         mul_add_c(a[5],b[1],c1,c2,c3);
605         mul_add_c(a[4],b[2],c1,c2,c3);
606         mul_add_c(a[3],b[3],c1,c2,c3);
607         mul_add_c(a[2],b[4],c1,c2,c3);
608         mul_add_c(a[1],b[5],c1,c2,c3);
609         mul_add_c(a[0],b[6],c1,c2,c3);
610         r[6]=c1;
611         c1=0;
612         mul_add_c(a[0],b[7],c2,c3,c1);
613         mul_add_c(a[1],b[6],c2,c3,c1);
614         mul_add_c(a[2],b[5],c2,c3,c1);
615         mul_add_c(a[3],b[4],c2,c3,c1);
616         mul_add_c(a[4],b[3],c2,c3,c1);
617         mul_add_c(a[5],b[2],c2,c3,c1);
618         mul_add_c(a[6],b[1],c2,c3,c1);
619         mul_add_c(a[7],b[0],c2,c3,c1);
620         r[7]=c2;
621         c2=0;
622         mul_add_c(a[7],b[1],c3,c1,c2);
623         mul_add_c(a[6],b[2],c3,c1,c2);
624         mul_add_c(a[5],b[3],c3,c1,c2);
625         mul_add_c(a[4],b[4],c3,c1,c2);
626         mul_add_c(a[3],b[5],c3,c1,c2);
627         mul_add_c(a[2],b[6],c3,c1,c2);
628         mul_add_c(a[1],b[7],c3,c1,c2);
629         r[8]=c3;
630         c3=0;
631         mul_add_c(a[2],b[7],c1,c2,c3);
632         mul_add_c(a[3],b[6],c1,c2,c3);
633         mul_add_c(a[4],b[5],c1,c2,c3);
634         mul_add_c(a[5],b[4],c1,c2,c3);
635         mul_add_c(a[6],b[3],c1,c2,c3);
636         mul_add_c(a[7],b[2],c1,c2,c3);
637         r[9]=c1;
638         c1=0;
639         mul_add_c(a[7],b[3],c2,c3,c1);
640         mul_add_c(a[6],b[4],c2,c3,c1);
641         mul_add_c(a[5],b[5],c2,c3,c1);
642         mul_add_c(a[4],b[6],c2,c3,c1);
643         mul_add_c(a[3],b[7],c2,c3,c1);
644         r[10]=c2;
645         c2=0;
646         mul_add_c(a[4],b[7],c3,c1,c2);
647         mul_add_c(a[5],b[6],c3,c1,c2);
648         mul_add_c(a[6],b[5],c3,c1,c2);
649         mul_add_c(a[7],b[4],c3,c1,c2);
650         r[11]=c3;
651         c3=0;
652         mul_add_c(a[7],b[5],c1,c2,c3);
653         mul_add_c(a[6],b[6],c1,c2,c3);
654         mul_add_c(a[5],b[7],c1,c2,c3);
655         r[12]=c1;
656         c1=0;
657         mul_add_c(a[6],b[7],c2,c3,c1);
658         mul_add_c(a[7],b[6],c2,c3,c1);
659         r[13]=c2;
660         c2=0;
661         mul_add_c(a[7],b[7],c3,c1,c2);
662         r[14]=c3;
663         r[15]=c1;
664         }
665
666 void bn_mul_comba4(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b)
667         {
668 #ifdef BN_LLONG
669         BN_ULLONG t;
670 #else
671         BN_ULONG bl,bh;
672 #endif
673         BN_ULONG t1,t2;
674         BN_ULONG c1,c2,c3;
675
676         c1=0;
677         c2=0;
678         c3=0;
679         mul_add_c(a[0],b[0],c1,c2,c3);
680         r[0]=c1;
681         c1=0;
682         mul_add_c(a[0],b[1],c2,c3,c1);
683         mul_add_c(a[1],b[0],c2,c3,c1);
684         r[1]=c2;
685         c2=0;
686         mul_add_c(a[2],b[0],c3,c1,c2);
687         mul_add_c(a[1],b[1],c3,c1,c2);
688         mul_add_c(a[0],b[2],c3,c1,c2);
689         r[2]=c3;
690         c3=0;
691         mul_add_c(a[0],b[3],c1,c2,c3);
692         mul_add_c(a[1],b[2],c1,c2,c3);
693         mul_add_c(a[2],b[1],c1,c2,c3);
694         mul_add_c(a[3],b[0],c1,c2,c3);
695         r[3]=c1;
696         c1=0;
697         mul_add_c(a[3],b[1],c2,c3,c1);
698         mul_add_c(a[2],b[2],c2,c3,c1);
699         mul_add_c(a[1],b[3],c2,c3,c1);
700         r[4]=c2;
701         c2=0;
702         mul_add_c(a[2],b[3],c3,c1,c2);
703         mul_add_c(a[3],b[2],c3,c1,c2);
704         r[5]=c3;
705         c3=0;
706         mul_add_c(a[3],b[3],c1,c2,c3);
707         r[6]=c1;
708         r[7]=c2;
709         }
710
711 void bn_sqr_comba8(BN_ULONG *r, const BN_ULONG *a)
712         {
713 #ifdef BN_LLONG
714         BN_ULLONG t,tt;
715 #else
716         BN_ULONG bl,bh;
717 #endif
718         BN_ULONG t1,t2;
719         BN_ULONG c1,c2,c3;
720
721         c1=0;
722         c2=0;
723         c3=0;
724         sqr_add_c(a,0,c1,c2,c3);
725         r[0]=c1;
726         c1=0;
727         sqr_add_c2(a,1,0,c2,c3,c1);
728         r[1]=c2;
729         c2=0;
730         sqr_add_c(a,1,c3,c1,c2);
731         sqr_add_c2(a,2,0,c3,c1,c2);
732         r[2]=c3;
733         c3=0;
734         sqr_add_c2(a,3,0,c1,c2,c3);
735         sqr_add_c2(a,2,1,c1,c2,c3);
736         r[3]=c1;
737         c1=0;
738         sqr_add_c(a,2,c2,c3,c1);
739         sqr_add_c2(a,3,1,c2,c3,c1);
740         sqr_add_c2(a,4,0,c2,c3,c1);
741         r[4]=c2;
742         c2=0;
743         sqr_add_c2(a,5,0,c3,c1,c2);
744         sqr_add_c2(a,4,1,c3,c1,c2);
745         sqr_add_c2(a,3,2,c3,c1,c2);
746         r[5]=c3;
747         c3=0;
748         sqr_add_c(a,3,c1,c2,c3);
749         sqr_add_c2(a,4,2,c1,c2,c3);
750         sqr_add_c2(a,5,1,c1,c2,c3);
751         sqr_add_c2(a,6,0,c1,c2,c3);
752         r[6]=c1;
753         c1=0;
754         sqr_add_c2(a,7,0,c2,c3,c1);
755         sqr_add_c2(a,6,1,c2,c3,c1);
756         sqr_add_c2(a,5,2,c2,c3,c1);
757         sqr_add_c2(a,4,3,c2,c3,c1);
758         r[7]=c2;
759         c2=0;
760         sqr_add_c(a,4,c3,c1,c2);
761         sqr_add_c2(a,5,3,c3,c1,c2);
762         sqr_add_c2(a,6,2,c3,c1,c2);
763         sqr_add_c2(a,7,1,c3,c1,c2);
764         r[8]=c3;
765         c3=0;
766         sqr_add_c2(a,7,2,c1,c2,c3);
767         sqr_add_c2(a,6,3,c1,c2,c3);
768         sqr_add_c2(a,5,4,c1,c2,c3);
769         r[9]=c1;
770         c1=0;
771         sqr_add_c(a,5,c2,c3,c1);
772         sqr_add_c2(a,6,4,c2,c3,c1);
773         sqr_add_c2(a,7,3,c2,c3,c1);
774         r[10]=c2;
775         c2=0;
776         sqr_add_c2(a,7,4,c3,c1,c2);
777         sqr_add_c2(a,6,5,c3,c1,c2);
778         r[11]=c3;
779         c3=0;
780         sqr_add_c(a,6,c1,c2,c3);
781         sqr_add_c2(a,7,5,c1,c2,c3);
782         r[12]=c1;
783         c1=0;
784         sqr_add_c2(a,7,6,c2,c3,c1);
785         r[13]=c2;
786         c2=0;
787         sqr_add_c(a,7,c3,c1,c2);
788         r[14]=c3;
789         r[15]=c1;
790         }
791
792 void bn_sqr_comba4(BN_ULONG *r, const BN_ULONG *a)
793         {
794 #ifdef BN_LLONG
795         BN_ULLONG t,tt;
796 #else
797         BN_ULONG bl,bh;
798 #endif
799         BN_ULONG t1,t2;
800         BN_ULONG c1,c2,c3;
801
802         c1=0;
803         c2=0;
804         c3=0;
805         sqr_add_c(a,0,c1,c2,c3);
806         r[0]=c1;
807         c1=0;
808         sqr_add_c2(a,1,0,c2,c3,c1);
809         r[1]=c2;
810         c2=0;
811         sqr_add_c(a,1,c3,c1,c2);
812         sqr_add_c2(a,2,0,c3,c1,c2);
813         r[2]=c3;
814         c3=0;
815         sqr_add_c2(a,3,0,c1,c2,c3);
816         sqr_add_c2(a,2,1,c1,c2,c3);
817         r[3]=c1;
818         c1=0;
819         sqr_add_c(a,2,c2,c3,c1);
820         sqr_add_c2(a,3,1,c2,c3,c1);
821         r[4]=c2;
822         c2=0;
823         sqr_add_c2(a,3,2,c3,c1,c2);
824         r[5]=c3;
825         c3=0;
826         sqr_add_c(a,3,c1,c2,c3);
827         r[6]=c1;
828         r[7]=c2;
829         }
830
831 #ifdef OPENSSL_BN_ASM_MONT
832 /*
833  * This is essentially reference implementation, which may or may not
834  * result in performance improvement. E.g. on IA-32 this routine was
835  * observed to give 40% faster rsa1024 private key operations and 10%
836  * faster rsa4096 ones, while on AMD64 it improves rsa1024 sign only
837  * by 10% and *worsens* rsa4096 sign by 15%. Once again, it's a
838  * reference implementation, one to be used as start-point for
839  * platform-specific assembler.
840  */
841 int bn_mul_mont(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp, const BN_ULONG *np,BN_ULONG n0, int num)
842         {
843         BN_ULONG c0,c1,ml,*tp;
844 #ifdef mul64
845         BN_ULONG mh;
846 #endif
847         volatile BN_ULONG *vp;
848         int i=0,j;
849
850 #if 0   /* template for platform-specific implementation */
851         if (ap==bp)     return bn_sqr_mont(rp,ap,np,n0,num);
852 #endif
853         vp = tp = alloca((num+2)*sizeof(BN_ULONG));
854
855         tp[num]   = bn_mul_words(tp,ap,num,bp[0]);
856         tp[num+1] = 0;
857         goto enter;
858
859         for(i=0;i<num;i++)
860                 {
861                 c0 = bn_mul_add_words(tp,ap,num,bp[i]);
862                 c1 = (tp[num] + c0)&BN_MASK2;
863                 tp[num]   = c1;
864                 tp[num+1] = (c1<c0?1:0);
865         enter:
866                 c1  = tp[0];
867                 ml = (c1*n0)&BN_MASK2;
868                 c0 = 0;
869 #ifdef mul64
870                 mh = HBITS(ml);
871                 ml = LBITS(ml);
872                 mul_add(c1,np[0],ml,mh,c0);
873 #else
874                 mul_add(c1,ml,np[0],c0);
875 #endif
876                 for(j=1;j<num;j++)
877                         {
878                         c1 = tp[j];
879 #ifdef mul64
880                         mul_add(c1,np[j],ml,mh,c0);
881 #else
882                         mul_add(c1,ml,np[j],c0);
883 #endif
884                         tp[j-1] = c1&BN_MASK2;
885                         }
886                 c1        = (tp[num] + c0)&BN_MASK2;
887                 tp[num-1] = c1;
888                 tp[num]   = tp[num+1] + (c1<c0?1:0);
889                 }
890
891         if (tp[num]!=0 || tp[num-1]>=np[num-1])
892                 {
893                 c0 = bn_sub_words(rp,tp,np,num);
894                 if (tp[num]!=0 || c0==0)
895                         {
896                         for(i=0;i<num+2;i++)    vp[i] = 0;
897                         return 1;
898                         }
899                 }
900         for(i=0;i<num;i++)      rp[i] = tp[i],  vp[i] = 0;
901         vp[num]   = 0;
902         vp[num+1] = 0;
903         return 1;
904         }
905 #else
906 /*
907  * Return value of 0 indicates that multiplication/convolution was not
908  * performed to signal the caller to fall down to alternative/original
909  * code-path.
910  */
911 int bn_mul_mont(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp, const BN_ULONG *np,BN_ULONG n0, int num)
912 {       return 0;       }
913 #endif /* OPENSSL_BN_ASM_MONT */
914
915 #else /* !BN_MUL_COMBA */
916
917 /* hmm... is it faster just to do a multiply? */
918 #undef bn_sqr_comba4
919 void bn_sqr_comba4(BN_ULONG *r, const BN_ULONG *a)
920         {
921         BN_ULONG t[8];
922         bn_sqr_normal(r,a,4,t);
923         }
924
925 #undef bn_sqr_comba8
926 void bn_sqr_comba8(BN_ULONG *r, const BN_ULONG *a)
927         {
928         BN_ULONG t[16];
929         bn_sqr_normal(r,a,8,t);
930         }
931
932 void bn_mul_comba4(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b)
933         {
934         r[4]=bn_mul_words(    &(r[0]),a,4,b[0]);
935         r[5]=bn_mul_add_words(&(r[1]),a,4,b[1]);
936         r[6]=bn_mul_add_words(&(r[2]),a,4,b[2]);
937         r[7]=bn_mul_add_words(&(r[3]),a,4,b[3]);
938         }
939
940 void bn_mul_comba8(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b)
941         {
942         r[ 8]=bn_mul_words(    &(r[0]),a,8,b[0]);
943         r[ 9]=bn_mul_add_words(&(r[1]),a,8,b[1]);
944         r[10]=bn_mul_add_words(&(r[2]),a,8,b[2]);
945         r[11]=bn_mul_add_words(&(r[3]),a,8,b[3]);
946         r[12]=bn_mul_add_words(&(r[4]),a,8,b[4]);
947         r[13]=bn_mul_add_words(&(r[5]),a,8,b[5]);
948         r[14]=bn_mul_add_words(&(r[6]),a,8,b[6]);
949         r[15]=bn_mul_add_words(&(r[7]),a,8,b[7]);
950         }
951
952 #ifdef OPENSSL_BN_ASM_MONT
953 int bn_mul_mont(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp, const BN_ULONG *np,BN_ULONG n0, int num)
954         {
955         BN_ULONG c0,c1,*tp;
956         volatile BN_ULONG *vp;
957         int i=0,j;
958
959         vp = tp = alloca((num+2)*sizeof(BN_ULONG));
960
961         for(i=0;i<=num;i++)     tp[i]=0;
962
963         for(i=0;i<num;i++)
964                 {
965                 c0         = bn_mul_add_words(tp,ap,num,bp[i]);
966                 c1         = tp[num] + c0;
967                 tp[num]    = c1;
968                 tp[num+1]  = (c1<c0?1:0);
969
970                 c0         = bn_mul_add_words(tp,np,num,tp[0]*n0);
971                 c1         = tp[num] + c0;
972                 tp[num]    = c1;
973                 tp[num+1] += (c1<c0?1:0);
974                 for(j=0;j<=num;j++)     tp[j]=tp[j+1];
975                 }
976
977         if (tp[num]!=0 || tp[num-1]>=np[num-1])
978                 {
979                 c0 = bn_sub_words(rp,tp,np,num);
980                 if (tp[num]!=0 || c0==0)
981                         {
982                         for(i=0;i<num+2;i++)    vp[i] = 0;
983                         return 1;
984                         }
985                 }
986         for(i=0;i<num;i++)      rp[i] = tp[i],  vp[i] = 0;
987         vp[num]   = 0;
988         vp[num+1] = 0;
989         return 1;
990         }
991 #else
992 int bn_mul_mont(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp, const BN_ULONG *np,BN_ULONG n0, int num)
993 {       return 0;       }
994 #endif /* OPENSSL_BN_ASM_MONT */
995
996 #endif /* !BN_MUL_COMBA */