]> git.ipfire.org Git - thirdparty/openssl.git/blob - crypto/bn/bn_asm.c
projects / thirdparty / openssl.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
RT2163: Remove some unneeded #include's
[thirdparty/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 <assert.h>
65 #include <openssl/crypto.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 #ifndef OPENSSL_FIPSCANISTER
432 #undef bn_mul_comba8
433 #undef bn_mul_comba4
434 #undef bn_sqr_comba8
435 #undef bn_sqr_comba4
436 #endif
437
438 /* mul_add_c(a,b,c0,c1,c2) -- c+=a*b for three word number c=(c2,c1,c0) */
439 /* mul_add_c2(a,b,c0,c1,c2) -- c+=2*a*b for three word number c=(c2,c1,c0) */
440 /* sqr_add_c(a,i,c0,c1,c2) -- c+=a[i]^2 for three word number c=(c2,c1,c0) */
441 /* sqr_add_c2(a,i,c0,c1,c2) -- c+=2*a[i]*a[j] for three word number c=(c2,c1,c0) */
442
443 #ifdef BN_LLONG
444 #define mul_add_c(a,b,c0,c1,c2) \
445 t=(BN_ULLONG)a*b; \
446 t1=(BN_ULONG)Lw(t); \
447 t2=(BN_ULONG)Hw(t); \
448 c0=(c0+t1)&BN_MASK2; if ((c0) < t1) t2++; \
449 c1=(c1+t2)&BN_MASK2; if ((c1) < t2) c2++;
450
451 #define mul_add_c2(a,b,c0,c1,c2) \
452 t=(BN_ULLONG)a*b; \
453 tt=(t+t)&BN_MASK; \
454 if (tt < t) c2++; \
455 t1=(BN_ULONG)Lw(tt); \
456 t2=(BN_ULONG)Hw(tt); \
457 c0=(c0+t1)&BN_MASK2; \
458 if ((c0 < t1) && (((++t2)&BN_MASK2) == 0)) c2++; \
459 c1=(c1+t2)&BN_MASK2; if ((c1) < t2) c2++;
460
461 #define sqr_add_c(a,i,c0,c1,c2) \
462 t=(BN_ULLONG)a[i]*a[i]; \
463 t1=(BN_ULONG)Lw(t); \
464 t2=(BN_ULONG)Hw(t); \
465 c0=(c0+t1)&BN_MASK2; if ((c0) < t1) t2++; \
466 c1=(c1+t2)&BN_MASK2; if ((c1) < t2) c2++;
467
468 #define sqr_add_c2(a,i,j,c0,c1,c2) \
469 mul_add_c2((a)[i],(a)[j],c0,c1,c2)
470
471 #elif defined(BN_UMULT_LOHI)
472
473 #define mul_add_c(a,b,c0,c1,c2) { \
474 BN_ULONG ta=(a),tb=(b); \
475 BN_UMULT_LOHI(t1,t2,ta,tb); \
476 c0 += t1; t2 += (c0<t1)?1:0; \
477 c1 += t2; c2 += (c1<t2)?1:0; \
478 }
479
480 #define mul_add_c2(a,b,c0,c1,c2) { \
481 BN_ULONG ta=(a),tb=(b),t0; \
482 BN_UMULT_LOHI(t0,t1,ta,tb); \
483 t2 = t1+t1; c2 += (t2<t1)?1:0; \
484 t1 = t0+t0; t2 += (t1<t0)?1:0; \
485 c0 += t1; t2 += (c0<t1)?1:0; \
486 c1 += t2; c2 += (c1<t2)?1:0; \
487 }
488
489 #define sqr_add_c(a,i,c0,c1,c2) { \
490 BN_ULONG ta=(a)[i]; \
491 BN_UMULT_LOHI(t1,t2,ta,ta); \
492 c0 += t1; t2 += (c0<t1)?1:0; \
493 c1 += t2; c2 += (c1<t2)?1:0; \
494 }
495
496 #define sqr_add_c2(a,i,j,c0,c1,c2) \
497 mul_add_c2((a)[i],(a)[j],c0,c1,c2)
498
499 #elif defined(BN_UMULT_HIGH)
500
501 #define mul_add_c(a,b,c0,c1,c2) { \
502 BN_ULONG ta=(a),tb=(b); \
503 t1 = ta * tb; \
504 t2 = BN_UMULT_HIGH(ta,tb); \
505 c0 += t1; t2 += (c0<t1)?1:0; \
506 c1 += t2; c2 += (c1<t2)?1:0; \
507 }
508
509 #define mul_add_c2(a,b,c0,c1,c2) { \
510 BN_ULONG ta=(a),tb=(b),t0; \
511 t1 = BN_UMULT_HIGH(ta,tb); \
512 t0 = ta * tb; \
513 t2 = t1+t1; c2 += (t2<t1)?1:0; \
514 t1 = t0+t0; t2 += (t1<t0)?1:0; \
515 c0 += t1; t2 += (c0<t1)?1:0; \
516 c1 += t2; c2 += (c1<t2)?1:0; \
517 }
518
519 #define sqr_add_c(a,i,c0,c1,c2) { \
520 BN_ULONG ta=(a)[i]; \
521 t1 = ta * ta; \
522 t2 = BN_UMULT_HIGH(ta,ta); \
523 c0 += t1; t2 += (c0<t1)?1:0; \
524 c1 += t2; c2 += (c1<t2)?1:0; \
525 }
526
527 #define sqr_add_c2(a,i,j,c0,c1,c2) \
528 mul_add_c2((a)[i],(a)[j],c0,c1,c2)
529
530 #else /* !BN_LLONG */
531 #define mul_add_c(a,b,c0,c1,c2) \
532 t1=LBITS(a); t2=HBITS(a); \
533 bl=LBITS(b); bh=HBITS(b); \
534 mul64(t1,t2,bl,bh); \
535 c0=(c0+t1)&BN_MASK2; if ((c0) < t1) t2++; \
536 c1=(c1+t2)&BN_MASK2; if ((c1) < t2) c2++;
537
538 #define mul_add_c2(a,b,c0,c1,c2) \
539 t1=LBITS(a); t2=HBITS(a); \
540 bl=LBITS(b); bh=HBITS(b); \
541 mul64(t1,t2,bl,bh); \
542 if (t2 & BN_TBIT) c2++; \
543 t2=(t2+t2)&BN_MASK2; \
544 if (t1 & BN_TBIT) t2++; \
545 t1=(t1+t1)&BN_MASK2; \
546 c0=(c0+t1)&BN_MASK2; \
547 if ((c0 < t1) && (((++t2)&BN_MASK2) == 0)) c2++; \
548 c1=(c1+t2)&BN_MASK2; if ((c1) < t2) c2++;
549
550 #define sqr_add_c(a,i,c0,c1,c2) \
551 sqr64(t1,t2,(a)[i]); \
552 c0=(c0+t1)&BN_MASK2; if ((c0) < t1) t2++; \
553 c1=(c1+t2)&BN_MASK2; if ((c1) < t2) c2++;
554
555 #define sqr_add_c2(a,i,j,c0,c1,c2) \
556 mul_add_c2((a)[i],(a)[j],c0,c1,c2)
557 #endif /* !BN_LLONG */
558
559 void bn_mul_comba8(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b)
560 {
561 #ifdef BN_LLONG
562 BN_ULLONG t;
563 #else
564 BN_ULONG bl,bh;
565 #endif
566 BN_ULONG t1,t2;
567 BN_ULONG c1,c2,c3;
568
569 c1=0;
570 c2=0;
571 c3=0;
572 mul_add_c(a[0],b[0],c1,c2,c3);
573 r[0]=c1;
574 c1=0;
575 mul_add_c(a[0],b[1],c2,c3,c1);
576 mul_add_c(a[1],b[0],c2,c3,c1);
577 r[1]=c2;
578 c2=0;
579 mul_add_c(a[2],b[0],c3,c1,c2);
580 mul_add_c(a[1],b[1],c3,c1,c2);
581 mul_add_c(a[0],b[2],c3,c1,c2);
582 r[2]=c3;
583 c3=0;
584 mul_add_c(a[0],b[3],c1,c2,c3);
585 mul_add_c(a[1],b[2],c1,c2,c3);
586 mul_add_c(a[2],b[1],c1,c2,c3);
587 mul_add_c(a[3],b[0],c1,c2,c3);
588 r[3]=c1;
589 c1=0;
590 mul_add_c(a[4],b[0],c2,c3,c1);
591 mul_add_c(a[3],b[1],c2,c3,c1);
592 mul_add_c(a[2],b[2],c2,c3,c1);
593 mul_add_c(a[1],b[3],c2,c3,c1);
594 mul_add_c(a[0],b[4],c2,c3,c1);
595 r[4]=c2;
596 c2=0;
597 mul_add_c(a[0],b[5],c3,c1,c2);
598 mul_add_c(a[1],b[4],c3,c1,c2);
599 mul_add_c(a[2],b[3],c3,c1,c2);
600 mul_add_c(a[3],b[2],c3,c1,c2);
601 mul_add_c(a[4],b[1],c3,c1,c2);
602 mul_add_c(a[5],b[0],c3,c1,c2);
603 r[5]=c3;
604 c3=0;
605 mul_add_c(a[6],b[0],c1,c2,c3);
606 mul_add_c(a[5],b[1],c1,c2,c3);
607 mul_add_c(a[4],b[2],c1,c2,c3);
608 mul_add_c(a[3],b[3],c1,c2,c3);
609 mul_add_c(a[2],b[4],c1,c2,c3);
610 mul_add_c(a[1],b[5],c1,c2,c3);
611 mul_add_c(a[0],b[6],c1,c2,c3);
612 r[6]=c1;
613 c1=0;
614 mul_add_c(a[0],b[7],c2,c3,c1);
615 mul_add_c(a[1],b[6],c2,c3,c1);
616 mul_add_c(a[2],b[5],c2,c3,c1);
617 mul_add_c(a[3],b[4],c2,c3,c1);
618 mul_add_c(a[4],b[3],c2,c3,c1);
619 mul_add_c(a[5],b[2],c2,c3,c1);
620 mul_add_c(a[6],b[1],c2,c3,c1);
621 mul_add_c(a[7],b[0],c2,c3,c1);
622 r[7]=c2;
623 c2=0;
624 mul_add_c(a[7],b[1],c3,c1,c2);
625 mul_add_c(a[6],b[2],c3,c1,c2);
626 mul_add_c(a[5],b[3],c3,c1,c2);
627 mul_add_c(a[4],b[4],c3,c1,c2);
628 mul_add_c(a[3],b[5],c3,c1,c2);
629 mul_add_c(a[2],b[6],c3,c1,c2);
630 mul_add_c(a[1],b[7],c3,c1,c2);
631 r[8]=c3;
632 c3=0;
633 mul_add_c(a[2],b[7],c1,c2,c3);
634 mul_add_c(a[3],b[6],c1,c2,c3);
635 mul_add_c(a[4],b[5],c1,c2,c3);
636 mul_add_c(a[5],b[4],c1,c2,c3);
637 mul_add_c(a[6],b[3],c1,c2,c3);
638 mul_add_c(a[7],b[2],c1,c2,c3);
639 r[9]=c1;
640 c1=0;
641 mul_add_c(a[7],b[3],c2,c3,c1);
642 mul_add_c(a[6],b[4],c2,c3,c1);
643 mul_add_c(a[5],b[5],c2,c3,c1);
644 mul_add_c(a[4],b[6],c2,c3,c1);
645 mul_add_c(a[3],b[7],c2,c3,c1);
646 r[10]=c2;
647 c2=0;
648 mul_add_c(a[4],b[7],c3,c1,c2);
649 mul_add_c(a[5],b[6],c3,c1,c2);
650 mul_add_c(a[6],b[5],c3,c1,c2);
651 mul_add_c(a[7],b[4],c3,c1,c2);
652 r[11]=c3;
653 c3=0;
654 mul_add_c(a[7],b[5],c1,c2,c3);
655 mul_add_c(a[6],b[6],c1,c2,c3);
656 mul_add_c(a[5],b[7],c1,c2,c3);
657 r[12]=c1;
658 c1=0;
659 mul_add_c(a[6],b[7],c2,c3,c1);
660 mul_add_c(a[7],b[6],c2,c3,c1);
661 r[13]=c2;
662 c2=0;
663 mul_add_c(a[7],b[7],c3,c1,c2);
664 r[14]=c3;
665 r[15]=c1;
666 }
667
668 void bn_mul_comba4(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b)
669 {
670 #ifdef BN_LLONG
671 BN_ULLONG t;
672 #else
673 BN_ULONG bl,bh;
674 #endif
675 BN_ULONG t1,t2;
676 BN_ULONG c1,c2,c3;
677
678 c1=0;
679 c2=0;
680 c3=0;
681 mul_add_c(a[0],b[0],c1,c2,c3);
682 r[0]=c1;
683 c1=0;
684 mul_add_c(a[0],b[1],c2,c3,c1);
685 mul_add_c(a[1],b[0],c2,c3,c1);
686 r[1]=c2;
687 c2=0;
688 mul_add_c(a[2],b[0],c3,c1,c2);
689 mul_add_c(a[1],b[1],c3,c1,c2);
690 mul_add_c(a[0],b[2],c3,c1,c2);
691 r[2]=c3;
692 c3=0;
693 mul_add_c(a[0],b[3],c1,c2,c3);
694 mul_add_c(a[1],b[2],c1,c2,c3);
695 mul_add_c(a[2],b[1],c1,c2,c3);
696 mul_add_c(a[3],b[0],c1,c2,c3);
697 r[3]=c1;
698 c1=0;
699 mul_add_c(a[3],b[1],c2,c3,c1);
700 mul_add_c(a[2],b[2],c2,c3,c1);
701 mul_add_c(a[1],b[3],c2,c3,c1);
702 r[4]=c2;
703 c2=0;
704 mul_add_c(a[2],b[3],c3,c1,c2);
705 mul_add_c(a[3],b[2],c3,c1,c2);
706 r[5]=c3;
707 c3=0;
708 mul_add_c(a[3],b[3],c1,c2,c3);
709 r[6]=c1;
710 r[7]=c2;
711 }
712
713 void bn_sqr_comba8(BN_ULONG *r, const BN_ULONG *a)
714 {
715 #ifdef BN_LLONG
716 BN_ULLONG t,tt;
717 #else
718 BN_ULONG bl,bh;
719 #endif
720 BN_ULONG t1,t2;
721 BN_ULONG c1,c2,c3;
722
723 c1=0;
724 c2=0;
725 c3=0;
726 sqr_add_c(a,0,c1,c2,c3);
727 r[0]=c1;
728 c1=0;
729 sqr_add_c2(a,1,0,c2,c3,c1);
730 r[1]=c2;
731 c2=0;
732 sqr_add_c(a,1,c3,c1,c2);
733 sqr_add_c2(a,2,0,c3,c1,c2);
734 r[2]=c3;
735 c3=0;
736 sqr_add_c2(a,3,0,c1,c2,c3);
737 sqr_add_c2(a,2,1,c1,c2,c3);
738 r[3]=c1;
739 c1=0;
740 sqr_add_c(a,2,c2,c3,c1);
741 sqr_add_c2(a,3,1,c2,c3,c1);
742 sqr_add_c2(a,4,0,c2,c3,c1);
743 r[4]=c2;
744 c2=0;
745 sqr_add_c2(a,5,0,c3,c1,c2);
746 sqr_add_c2(a,4,1,c3,c1,c2);
747 sqr_add_c2(a,3,2,c3,c1,c2);
748 r[5]=c3;
749 c3=0;
750 sqr_add_c(a,3,c1,c2,c3);
751 sqr_add_c2(a,4,2,c1,c2,c3);
752 sqr_add_c2(a,5,1,c1,c2,c3);
753 sqr_add_c2(a,6,0,c1,c2,c3);
754 r[6]=c1;
755 c1=0;
756 sqr_add_c2(a,7,0,c2,c3,c1);
757 sqr_add_c2(a,6,1,c2,c3,c1);
758 sqr_add_c2(a,5,2,c2,c3,c1);
759 sqr_add_c2(a,4,3,c2,c3,c1);
760 r[7]=c2;
761 c2=0;
762 sqr_add_c(a,4,c3,c1,c2);
763 sqr_add_c2(a,5,3,c3,c1,c2);
764 sqr_add_c2(a,6,2,c3,c1,c2);
765 sqr_add_c2(a,7,1,c3,c1,c2);
766 r[8]=c3;
767 c3=0;
768 sqr_add_c2(a,7,2,c1,c2,c3);
769 sqr_add_c2(a,6,3,c1,c2,c3);
770 sqr_add_c2(a,5,4,c1,c2,c3);
771 r[9]=c1;
772 c1=0;
773 sqr_add_c(a,5,c2,c3,c1);
774 sqr_add_c2(a,6,4,c2,c3,c1);
775 sqr_add_c2(a,7,3,c2,c3,c1);
776 r[10]=c2;
777 c2=0;
778 sqr_add_c2(a,7,4,c3,c1,c2);
779 sqr_add_c2(a,6,5,c3,c1,c2);
780 r[11]=c3;
781 c3=0;
782 sqr_add_c(a,6,c1,c2,c3);
783 sqr_add_c2(a,7,5,c1,c2,c3);
784 r[12]=c1;
785 c1=0;
786 sqr_add_c2(a,7,6,c2,c3,c1);
787 r[13]=c2;
788 c2=0;
789 sqr_add_c(a,7,c3,c1,c2);
790 r[14]=c3;
791 r[15]=c1;
792 }
793
794 void bn_sqr_comba4(BN_ULONG *r, const BN_ULONG *a)
795 {
796 #ifdef BN_LLONG
797 BN_ULLONG t,tt;
798 #else
799 BN_ULONG bl,bh;
800 #endif
801 BN_ULONG t1,t2;
802 BN_ULONG c1,c2,c3;
803
804 c1=0;
805 c2=0;
806 c3=0;
807 sqr_add_c(a,0,c1,c2,c3);
808 r[0]=c1;
809 c1=0;
810 sqr_add_c2(a,1,0,c2,c3,c1);
811 r[1]=c2;
812 c2=0;
813 sqr_add_c(a,1,c3,c1,c2);
814 sqr_add_c2(a,2,0,c3,c1,c2);
815 r[2]=c3;
816 c3=0;
817 sqr_add_c2(a,3,0,c1,c2,c3);
818 sqr_add_c2(a,2,1,c1,c2,c3);
819 r[3]=c1;
820 c1=0;
821 sqr_add_c(a,2,c2,c3,c1);
822 sqr_add_c2(a,3,1,c2,c3,c1);
823 r[4]=c2;
824 c2=0;
825 sqr_add_c2(a,3,2,c3,c1,c2);
826 r[5]=c3;
827 c3=0;
828 sqr_add_c(a,3,c1,c2,c3);
829 r[6]=c1;
830 r[7]=c2;
831 }
832
833 #ifdef OPENSSL_NO_ASM
834 #ifdef OPENSSL_BN_ASM_MONT
835 #include <alloca.h>
836 /*
837 * This is essentially reference implementation, which may or may not
838 * result in performance improvement. E.g. on IA-32 this routine was
839 * observed to give 40% faster rsa1024 private key operations and 10%
840 * faster rsa4096 ones, while on AMD64 it improves rsa1024 sign only
841 * by 10% and *worsens* rsa4096 sign by 15%. Once again, it's a
842 * reference implementation, one to be used as starting point for
843 * platform-specific assembler. Mentioned numbers apply to compiler
844 * generated code compiled with and without -DOPENSSL_BN_ASM_MONT and
845 * can vary not only from platform to platform, but even for compiler
846 * versions. Assembler vs. assembler improvement coefficients can
847 * [and are known to] differ and are to be documented elsewhere.
848 */
849 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)
850 {
851 BN_ULONG c0,c1,ml,*tp,n0;
852 #ifdef mul64
853 BN_ULONG mh;
854 #endif
855 volatile BN_ULONG *vp;
856 int i=0,j;
857
858 #if 0 /* template for platform-specific implementation */
859 if (ap==bp) return bn_sqr_mont(rp,ap,np,n0p,num);
860 #endif
861 vp = tp = alloca((num+2)*sizeof(BN_ULONG));
862
863 n0 = *n0p;
864
865 c0 = 0;
866 ml = bp[0];
867 #ifdef mul64
868 mh = HBITS(ml);
869 ml = LBITS(ml);
870 for (j=0;j<num;++j)
871 mul(tp[j],ap[j],ml,mh,c0);
872 #else
873 for (j=0;j<num;++j)
874 mul(tp[j],ap[j],ml,c0);
875 #endif
876
877 tp[num] = c0;
878 tp[num+1] = 0;
879 goto enter;
880
881 for(i=0;i<num;i++)
882 {
883 c0 = 0;
884 ml = bp[i];
885 #ifdef mul64
886 mh = HBITS(ml);
887 ml = LBITS(ml);
888 for (j=0;j<num;++j)
889 mul_add(tp[j],ap[j],ml,mh,c0);
890 #else
891 for (j=0;j<num;++j)
892 mul_add(tp[j],ap[j],ml,c0);
893 #endif
894 c1 = (tp[num] + c0)&BN_MASK2;
895 tp[num] = c1;
896 tp[num+1] = (c1<c0?1:0);
897 enter:
898 c1 = tp[0];
899 ml = (c1*n0)&BN_MASK2;
900 c0 = 0;
901 #ifdef mul64
902 mh = HBITS(ml);
903 ml = LBITS(ml);
904 mul_add(c1,np[0],ml,mh,c0);
905 #else
906 mul_add(c1,ml,np[0],c0);
907 #endif
908 for(j=1;j<num;j++)
909 {
910 c1 = tp[j];
911 #ifdef mul64
912 mul_add(c1,np[j],ml,mh,c0);
913 #else
914 mul_add(c1,ml,np[j],c0);
915 #endif
916 tp[j-1] = c1&BN_MASK2;
917 }
918 c1 = (tp[num] + c0)&BN_MASK2;
919 tp[num-1] = c1;
920 tp[num] = tp[num+1] + (c1<c0?1:0);
921 }
922
923 if (tp[num]!=0 || tp[num-1]>=np[num-1])
924 {
925 c0 = bn_sub_words(rp,tp,np,num);
926 if (tp[num]!=0 || c0==0)
927 {
928 for(i=0;i<num+2;i++) vp[i] = 0;
929 return 1;
930 }
931 }
932 for(i=0;i<num;i++) rp[i] = tp[i], vp[i] = 0;
933 vp[num] = 0;
934 vp[num+1] = 0;
935 return 1;
936 }
937 #else
938 /*
939 * Return value of 0 indicates that multiplication/convolution was not
940 * performed to signal the caller to fall down to alternative/original
941 * code-path.
942 */
943 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)
944 { return 0; }
945 #endif /* OPENSSL_BN_ASM_MONT */
946 #endif
947
948 #else /* !BN_MUL_COMBA */
949
950 /* hmm... is it faster just to do a multiply? */
951 #ifndef OPENSSL_FIPSCANISTER
952 #undef bn_sqr_comba4
953 #undef bn_sqr_comba8
954 #endif
955 void bn_sqr_comba4(BN_ULONG *r, const BN_ULONG *a)
956 {
957 BN_ULONG t[8];
958 bn_sqr_normal(r,a,4,t);
959 }
960
961 void bn_sqr_comba8(BN_ULONG *r, const BN_ULONG *a)
962 {
963 BN_ULONG t[16];
964 bn_sqr_normal(r,a,8,t);
965 }
966
967 void bn_mul_comba4(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b)
968 {
969 r[4]=bn_mul_words( &(r[0]),a,4,b[0]);
970 r[5]=bn_mul_add_words(&(r[1]),a,4,b[1]);
971 r[6]=bn_mul_add_words(&(r[2]),a,4,b[2]);
972 r[7]=bn_mul_add_words(&(r[3]),a,4,b[3]);
973 }
974
975 void bn_mul_comba8(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b)
976 {
977 r[ 8]=bn_mul_words( &(r[0]),a,8,b[0]);
978 r[ 9]=bn_mul_add_words(&(r[1]),a,8,b[1]);
979 r[10]=bn_mul_add_words(&(r[2]),a,8,b[2]);
980 r[11]=bn_mul_add_words(&(r[3]),a,8,b[3]);
981 r[12]=bn_mul_add_words(&(r[4]),a,8,b[4]);
982 r[13]=bn_mul_add_words(&(r[5]),a,8,b[5]);
983 r[14]=bn_mul_add_words(&(r[6]),a,8,b[6]);
984 r[15]=bn_mul_add_words(&(r[7]),a,8,b[7]);
985 }
986
987 #ifdef OPENSSL_NO_ASM
988 #ifdef OPENSSL_BN_ASM_MONT
989 #include <alloca.h>
990 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 {
992 BN_ULONG c0,c1,*tp,n0=*n0p;
993 volatile BN_ULONG *vp;
994 int i=0,j;
995
996 vp = tp = alloca((num+2)*sizeof(BN_ULONG));
997
998 for(i=0;i<=num;i++) tp[i]=0;
999
1000 for(i=0;i<num;i++)
1001 {
1002 c0 = bn_mul_add_words(tp,ap,num,bp[i]);
1003 c1 = (tp[num] + c0)&BN_MASK2;
1004 tp[num] = c1;
1005 tp[num+1] = (c1<c0?1:0);
1006
1007 c0 = bn_mul_add_words(tp,np,num,tp[0]*n0);
1008 c1 = (tp[num] + c0)&BN_MASK2;
1009 tp[num] = c1;
1010 tp[num+1] += (c1<c0?1:0);
1011 for(j=0;j<=num;j++) tp[j]=tp[j+1];
1012 }
1013
1014 if (tp[num]!=0 || tp[num-1]>=np[num-1])
1015 {
1016 c0 = bn_sub_words(rp,tp,np,num);
1017 if (tp[num]!=0 || c0==0)
1018 {
1019 for(i=0;i<num+2;i++) vp[i] = 0;
1020 return 1;
1021 }
1022 }
1023 for(i=0;i<num;i++) rp[i] = tp[i], vp[i] = 0;
1024 vp[num] = 0;
1025 vp[num+1] = 0;
1026 return 1;
1027 }
1028 #else
1029 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 { return 0; }
1031 #endif /* OPENSSL_BN_ASM_MONT */
1032 #endif
1033
1034 #endif /* !BN_MUL_COMBA */
A mirror of the official openssl repository