]> git.ipfire.org Git - thirdparty/openssl.git/blob - crypto/bn/asm/x86_64-gcc.c
projects / thirdparty / openssl.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
mark all block comments that need format preserving so that
[thirdparty/openssl.git] / crypto / bn / asm / x86_64-gcc.c
1 #include "../bn_lcl.h"
2 #if !(defined(__GNUC__) && __GNUC__>=2)
3 # include "../bn_asm.c" /* kind of dirty hack for Sun Studio */
4 #else
5 /*-
6 * x86_64 BIGNUM accelerator version 0.1, December 2002.
7 *
8 * Implemented by Andy Polyakov <appro@fy.chalmers.se> for the OpenSSL
9 * project.
10 *
11 * Rights for redistribution and usage in source and binary forms are
12 * granted according to the OpenSSL license. Warranty of any kind is
13 * disclaimed.
14 *
15 * Q. Version 0.1? It doesn't sound like Andy, he used to assign real
16 * versions, like 1.0...
17 * A. Well, that's because this code is basically a quick-n-dirty
18 * proof-of-concept hack. As you can see it's implemented with
19 * inline assembler, which means that you're bound to GCC and that
20 * there might be enough room for further improvement.
21 *
22 * Q. Why inline assembler?
23 * A. x86_64 features own ABI which I'm not familiar with. This is
24 * why I decided to let the compiler take care of subroutine
25 * prologue/epilogue as well as register allocation. For reference.
26 * Win64 implements different ABI for AMD64, different from Linux.
27 *
28 * Q. How much faster does it get?
29 * A. 'apps/openssl speed rsa dsa' output with no-asm:
30 *
31 * sign verify sign/s verify/s
32 * rsa 512 bits 0.0006s 0.0001s 1683.8 18456.2
33 * rsa 1024 bits 0.0028s 0.0002s 356.0 6407.0
34 * rsa 2048 bits 0.0172s 0.0005s 58.0 1957.8
35 * rsa 4096 bits 0.1155s 0.0018s 8.7 555.6
36 * sign verify sign/s verify/s
37 * dsa 512 bits 0.0005s 0.0006s 2100.8 1768.3
38 * dsa 1024 bits 0.0014s 0.0018s 692.3 559.2
39 * dsa 2048 bits 0.0049s 0.0061s 204.7 165.0
40 *
41 * 'apps/openssl speed rsa dsa' output with this module:
42 *
43 * sign verify sign/s verify/s
44 * rsa 512 bits 0.0004s 0.0000s 2767.1 33297.9
45 * rsa 1024 bits 0.0012s 0.0001s 867.4 14674.7
46 * rsa 2048 bits 0.0061s 0.0002s 164.0 5270.0
47 * rsa 4096 bits 0.0384s 0.0006s 26.1 1650.8
48 * sign verify sign/s verify/s
49 * dsa 512 bits 0.0002s 0.0003s 4442.2 3786.3
50 * dsa 1024 bits 0.0005s 0.0007s 1835.1 1497.4
51 * dsa 2048 bits 0.0016s 0.0020s 620.4 504.6
52 *
53 * For the reference. IA-32 assembler implementation performs
54 * very much like 64-bit code compiled with no-asm on the same
55 * machine.
56 */
57
58 #if defined(_WIN64) || !defined(__LP64__)
59 #define BN_ULONG unsigned long long
60 #else
61 #define BN_ULONG unsigned long
62 #endif
63
64 #undef mul
65 #undef mul_add
66
67 /*-
68 * "m"(a), "+m"(r) is the way to favor DirectPath ยต-code;
69 * "g"(0) let the compiler to decide where does it
70 * want to keep the value of zero;
71 */
72 #define mul_add(r,a,word,carry) do { \
73 register BN_ULONG high,low; \
74 asm ("mulq %3" \
75 : "=a"(low),"=d"(high) \
76 : "a"(word),"m"(a) \
77 : "cc"); \
78 asm ("addq %2,%0; adcq %3,%1" \
79 : "+r"(carry),"+d"(high)\
80 : "a"(low),"g"(0) \
81 : "cc"); \
82 asm ("addq %2,%0; adcq %3,%1" \
83 : "+m"(r),"+d"(high) \
84 : "r"(carry),"g"(0) \
85 : "cc"); \
86 carry=high; \
87 } while (0)
88
89 #define mul(r,a,word,carry) do { \
90 register BN_ULONG high,low; \
91 asm ("mulq %3" \
92 : "=a"(low),"=d"(high) \
93 : "a"(word),"g"(a) \
94 : "cc"); \
95 asm ("addq %2,%0; adcq %3,%1" \
96 : "+r"(carry),"+d"(high)\
97 : "a"(low),"g"(0) \
98 : "cc"); \
99 (r)=carry, carry=high; \
100 } while (0)
101 #undef sqr
102 #define sqr(r0,r1,a) \
103 asm ("mulq %2" \
104 : "=a"(r0),"=d"(r1) \
105 : "a"(a) \
106 : "cc");
107
108 BN_ULONG bn_mul_add_words(BN_ULONG *rp, const BN_ULONG *ap, int num, BN_ULONG w)
109 {
110 BN_ULONG c1=0;
111
112 if (num <= 0) return(c1);
113
114 while (num&~3)
115 {
116 mul_add(rp[0],ap[0],w,c1);
117 mul_add(rp[1],ap[1],w,c1);
118 mul_add(rp[2],ap[2],w,c1);
119 mul_add(rp[3],ap[3],w,c1);
120 ap+=4; rp+=4; num-=4;
121 }
122 if (num)
123 {
124 mul_add(rp[0],ap[0],w,c1); if (--num==0) return c1;
125 mul_add(rp[1],ap[1],w,c1); if (--num==0) return c1;
126 mul_add(rp[2],ap[2],w,c1); return c1;
127 }
128
129 return(c1);
130 }
131
132 BN_ULONG bn_mul_words(BN_ULONG *rp, const BN_ULONG *ap, int num, BN_ULONG w)
133 {
134 BN_ULONG c1=0;
135
136 if (num <= 0) return(c1);
137
138 while (num&~3)
139 {
140 mul(rp[0],ap[0],w,c1);
141 mul(rp[1],ap[1],w,c1);
142 mul(rp[2],ap[2],w,c1);
143 mul(rp[3],ap[3],w,c1);
144 ap+=4; rp+=4; num-=4;
145 }
146 if (num)
147 {
148 mul(rp[0],ap[0],w,c1); if (--num == 0) return c1;
149 mul(rp[1],ap[1],w,c1); if (--num == 0) return c1;
150 mul(rp[2],ap[2],w,c1);
151 }
152 return(c1);
153 }
154
155 void bn_sqr_words(BN_ULONG *r, const BN_ULONG *a, int n)
156 {
157 if (n <= 0) return;
158
159 while (n&~3)
160 {
161 sqr(r[0],r[1],a[0]);
162 sqr(r[2],r[3],a[1]);
163 sqr(r[4],r[5],a[2]);
164 sqr(r[6],r[7],a[3]);
165 a+=4; r+=8; n-=4;
166 }
167 if (n)
168 {
169 sqr(r[0],r[1],a[0]); if (--n == 0) return;
170 sqr(r[2],r[3],a[1]); if (--n == 0) return;
171 sqr(r[4],r[5],a[2]);
172 }
173 }
174
175 BN_ULONG bn_div_words(BN_ULONG h, BN_ULONG l, BN_ULONG d)
176 { BN_ULONG ret,waste;
177
178 asm ("divq %4"
179 : "=a"(ret),"=d"(waste)
180 : "a"(l),"d"(h),"g"(d)
181 : "cc");
182
183 return ret;
184 }
185
186 BN_ULONG bn_add_words (BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp,int n)
187 { BN_ULONG ret;
188 size_t i=0;
189
190 if (n <= 0) return 0;
191
192 asm volatile (
193 " subq %0,%0 \n" /* clear carry */
194 " jmp 1f \n"
195 ".p2align 4 \n"
196 "1: movq (%4,%2,8),%0 \n"
197 " adcq (%5,%2,8),%0 \n"
198 " movq %0,(%3,%2,8) \n"
199 " lea 1(%2),%2 \n"
200 " loop 1b \n"
201 " sbbq %0,%0 \n"
202 : "=&r"(ret),"+c"(n),"+r"(i)
203 : "r"(rp),"r"(ap),"r"(bp)
204 : "cc", "memory"
205 );
206
207 return ret&1;
208 }
209
210 #ifndef SIMICS
211 BN_ULONG bn_sub_words (BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp,int n)
212 { BN_ULONG ret;
213 size_t i=0;
214
215 if (n <= 0) return 0;
216
217 asm volatile (
218 " subq %0,%0 \n" /* clear borrow */
219 " jmp 1f \n"
220 ".p2align 4 \n"
221 "1: movq (%4,%2,8),%0 \n"
222 " sbbq (%5,%2,8),%0 \n"
223 " movq %0,(%3,%2,8) \n"
224 " lea 1(%2),%2 \n"
225 " loop 1b \n"
226 " sbbq %0,%0 \n"
227 : "=&r"(ret),"+c"(n),"+r"(i)
228 : "r"(rp),"r"(ap),"r"(bp)
229 : "cc", "memory"
230 );
231
232 return ret&1;
233 }
234 #else
235 /* Simics 1.4<7 has buggy sbbq:-( */
236 #define BN_MASK2 0xffffffffffffffffL
237 BN_ULONG bn_sub_words(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b, int n)
238 {
239 BN_ULONG t1,t2;
240 int c=0;
241
242 if (n <= 0) return((BN_ULONG)0);
243
244 for (;;)
245 {
246 t1=a[0]; t2=b[0];
247 r[0]=(t1-t2-c)&BN_MASK2;
248 if (t1 != t2) c=(t1 < t2);
249 if (--n <= 0) break;
250
251 t1=a[1]; t2=b[1];
252 r[1]=(t1-t2-c)&BN_MASK2;
253 if (t1 != t2) c=(t1 < t2);
254 if (--n <= 0) break;
255
256 t1=a[2]; t2=b[2];
257 r[2]=(t1-t2-c)&BN_MASK2;
258 if (t1 != t2) c=(t1 < t2);
259 if (--n <= 0) break;
260
261 t1=a[3]; t2=b[3];
262 r[3]=(t1-t2-c)&BN_MASK2;
263 if (t1 != t2) c=(t1 < t2);
264 if (--n <= 0) break;
265
266 a+=4;
267 b+=4;
268 r+=4;
269 }
270 return(c);
271 }
272 #endif
273
274 /* mul_add_c(a,b,c0,c1,c2) -- c+=a*b for three word number c=(c2,c1,c0) */
275 /* mul_add_c2(a,b,c0,c1,c2) -- c+=2*a*b for three word number c=(c2,c1,c0) */
276 /* sqr_add_c(a,i,c0,c1,c2) -- c+=a[i]^2 for three word number c=(c2,c1,c0) */
277 /* sqr_add_c2(a,i,c0,c1,c2) -- c+=2*a[i]*a[j] for three word number c=(c2,c1,c0) */
278
279 #if 0
280 /* original macros are kept for reference purposes */
281 #define mul_add_c(a,b,c0,c1,c2) { \
282 BN_ULONG ta=(a),tb=(b); \
283 t1 = ta * tb; \
284 t2 = BN_UMULT_HIGH(ta,tb); \
285 c0 += t1; t2 += (c0<t1)?1:0; \
286 c1 += t2; c2 += (c1<t2)?1:0; \
287 }
288
289 #define mul_add_c2(a,b,c0,c1,c2) { \
290 BN_ULONG ta=(a),tb=(b),t0; \
291 t1 = BN_UMULT_HIGH(ta,tb); \
292 t0 = ta * tb; \
293 t2 = t1+t1; c2 += (t2<t1)?1:0; \
294 t1 = t0+t0; t2 += (t1<t0)?1:0; \
295 c0 += t1; t2 += (c0<t1)?1:0; \
296 c1 += t2; c2 += (c1<t2)?1:0; \
297 }
298 #else
299 #define mul_add_c(a,b,c0,c1,c2) do { \
300 asm ("mulq %3" \
301 : "=a"(t1),"=d"(t2) \
302 : "a"(a),"m"(b) \
303 : "cc"); \
304 asm ("addq %2,%0; adcq %3,%1" \
305 : "+r"(c0),"+d"(t2) \
306 : "a"(t1),"g"(0) \
307 : "cc"); \
308 asm ("addq %2,%0; adcq %3,%1" \
309 : "+r"(c1),"+r"(c2) \
310 : "d"(t2),"g"(0) \
311 : "cc"); \
312 } while (0)
313
314 #define sqr_add_c(a,i,c0,c1,c2) do { \
315 asm ("mulq %2" \
316 : "=a"(t1),"=d"(t2) \
317 : "a"(a[i]) \
318 : "cc"); \
319 asm ("addq %2,%0; adcq %3,%1" \
320 : "+r"(c0),"+d"(t2) \
321 : "a"(t1),"g"(0) \
322 : "cc"); \
323 asm ("addq %2,%0; adcq %3,%1" \
324 : "+r"(c1),"+r"(c2) \
325 : "d"(t2),"g"(0) \
326 : "cc"); \
327 } while (0)
328
329 #define mul_add_c2(a,b,c0,c1,c2) do { \
330 asm ("mulq %3" \
331 : "=a"(t1),"=d"(t2) \
332 : "a"(a),"m"(b) \
333 : "cc"); \
334 asm ("addq %0,%0; adcq %2,%1" \
335 : "+d"(t2),"+r"(c2) \
336 : "g"(0) \
337 : "cc"); \
338 asm ("addq %0,%0; adcq %2,%1" \
339 : "+a"(t1),"+d"(t2) \
340 : "g"(0) \
341 : "cc"); \
342 asm ("addq %2,%0; adcq %3,%1" \
343 : "+r"(c0),"+d"(t2) \
344 : "a"(t1),"g"(0) \
345 : "cc"); \
346 asm ("addq %2,%0; adcq %3,%1" \
347 : "+r"(c1),"+r"(c2) \
348 : "d"(t2),"g"(0) \
349 : "cc"); \
350 } while (0)
351 #endif
352
353 #define sqr_add_c2(a,i,j,c0,c1,c2) \
354 mul_add_c2((a)[i],(a)[j],c0,c1,c2)
355
356 void bn_mul_comba8(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b)
357 {
358 BN_ULONG t1,t2;
359 BN_ULONG c1,c2,c3;
360
361 c1=0;
362 c2=0;
363 c3=0;
364 mul_add_c(a[0],b[0],c1,c2,c3);
365 r[0]=c1;
366 c1=0;
367 mul_add_c(a[0],b[1],c2,c3,c1);
368 mul_add_c(a[1],b[0],c2,c3,c1);
369 r[1]=c2;
370 c2=0;
371 mul_add_c(a[2],b[0],c3,c1,c2);
372 mul_add_c(a[1],b[1],c3,c1,c2);
373 mul_add_c(a[0],b[2],c3,c1,c2);
374 r[2]=c3;
375 c3=0;
376 mul_add_c(a[0],b[3],c1,c2,c3);
377 mul_add_c(a[1],b[2],c1,c2,c3);
378 mul_add_c(a[2],b[1],c1,c2,c3);
379 mul_add_c(a[3],b[0],c1,c2,c3);
380 r[3]=c1;
381 c1=0;
382 mul_add_c(a[4],b[0],c2,c3,c1);
383 mul_add_c(a[3],b[1],c2,c3,c1);
384 mul_add_c(a[2],b[2],c2,c3,c1);
385 mul_add_c(a[1],b[3],c2,c3,c1);
386 mul_add_c(a[0],b[4],c2,c3,c1);
387 r[4]=c2;
388 c2=0;
389 mul_add_c(a[0],b[5],c3,c1,c2);
390 mul_add_c(a[1],b[4],c3,c1,c2);
391 mul_add_c(a[2],b[3],c3,c1,c2);
392 mul_add_c(a[3],b[2],c3,c1,c2);
393 mul_add_c(a[4],b[1],c3,c1,c2);
394 mul_add_c(a[5],b[0],c3,c1,c2);
395 r[5]=c3;
396 c3=0;
397 mul_add_c(a[6],b[0],c1,c2,c3);
398 mul_add_c(a[5],b[1],c1,c2,c3);
399 mul_add_c(a[4],b[2],c1,c2,c3);
400 mul_add_c(a[3],b[3],c1,c2,c3);
401 mul_add_c(a[2],b[4],c1,c2,c3);
402 mul_add_c(a[1],b[5],c1,c2,c3);
403 mul_add_c(a[0],b[6],c1,c2,c3);
404 r[6]=c1;
405 c1=0;
406 mul_add_c(a[0],b[7],c2,c3,c1);
407 mul_add_c(a[1],b[6],c2,c3,c1);
408 mul_add_c(a[2],b[5],c2,c3,c1);
409 mul_add_c(a[3],b[4],c2,c3,c1);
410 mul_add_c(a[4],b[3],c2,c3,c1);
411 mul_add_c(a[5],b[2],c2,c3,c1);
412 mul_add_c(a[6],b[1],c2,c3,c1);
413 mul_add_c(a[7],b[0],c2,c3,c1);
414 r[7]=c2;
415 c2=0;
416 mul_add_c(a[7],b[1],c3,c1,c2);
417 mul_add_c(a[6],b[2],c3,c1,c2);
418 mul_add_c(a[5],b[3],c3,c1,c2);
419 mul_add_c(a[4],b[4],c3,c1,c2);
420 mul_add_c(a[3],b[5],c3,c1,c2);
421 mul_add_c(a[2],b[6],c3,c1,c2);
422 mul_add_c(a[1],b[7],c3,c1,c2);
423 r[8]=c3;
424 c3=0;
425 mul_add_c(a[2],b[7],c1,c2,c3);
426 mul_add_c(a[3],b[6],c1,c2,c3);
427 mul_add_c(a[4],b[5],c1,c2,c3);
428 mul_add_c(a[5],b[4],c1,c2,c3);
429 mul_add_c(a[6],b[3],c1,c2,c3);
430 mul_add_c(a[7],b[2],c1,c2,c3);
431 r[9]=c1;
432 c1=0;
433 mul_add_c(a[7],b[3],c2,c3,c1);
434 mul_add_c(a[6],b[4],c2,c3,c1);
435 mul_add_c(a[5],b[5],c2,c3,c1);
436 mul_add_c(a[4],b[6],c2,c3,c1);
437 mul_add_c(a[3],b[7],c2,c3,c1);
438 r[10]=c2;
439 c2=0;
440 mul_add_c(a[4],b[7],c3,c1,c2);
441 mul_add_c(a[5],b[6],c3,c1,c2);
442 mul_add_c(a[6],b[5],c3,c1,c2);
443 mul_add_c(a[7],b[4],c3,c1,c2);
444 r[11]=c3;
445 c3=0;
446 mul_add_c(a[7],b[5],c1,c2,c3);
447 mul_add_c(a[6],b[6],c1,c2,c3);
448 mul_add_c(a[5],b[7],c1,c2,c3);
449 r[12]=c1;
450 c1=0;
451 mul_add_c(a[6],b[7],c2,c3,c1);
452 mul_add_c(a[7],b[6],c2,c3,c1);
453 r[13]=c2;
454 c2=0;
455 mul_add_c(a[7],b[7],c3,c1,c2);
456 r[14]=c3;
457 r[15]=c1;
458 }
459
460 void bn_mul_comba4(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b)
461 {
462 BN_ULONG t1,t2;
463 BN_ULONG c1,c2,c3;
464
465 c1=0;
466 c2=0;
467 c3=0;
468 mul_add_c(a[0],b[0],c1,c2,c3);
469 r[0]=c1;
470 c1=0;
471 mul_add_c(a[0],b[1],c2,c3,c1);
472 mul_add_c(a[1],b[0],c2,c3,c1);
473 r[1]=c2;
474 c2=0;
475 mul_add_c(a[2],b[0],c3,c1,c2);
476 mul_add_c(a[1],b[1],c3,c1,c2);
477 mul_add_c(a[0],b[2],c3,c1,c2);
478 r[2]=c3;
479 c3=0;
480 mul_add_c(a[0],b[3],c1,c2,c3);
481 mul_add_c(a[1],b[2],c1,c2,c3);
482 mul_add_c(a[2],b[1],c1,c2,c3);
483 mul_add_c(a[3],b[0],c1,c2,c3);
484 r[3]=c1;
485 c1=0;
486 mul_add_c(a[3],b[1],c2,c3,c1);
487 mul_add_c(a[2],b[2],c2,c3,c1);
488 mul_add_c(a[1],b[3],c2,c3,c1);
489 r[4]=c2;
490 c2=0;
491 mul_add_c(a[2],b[3],c3,c1,c2);
492 mul_add_c(a[3],b[2],c3,c1,c2);
493 r[5]=c3;
494 c3=0;
495 mul_add_c(a[3],b[3],c1,c2,c3);
496 r[6]=c1;
497 r[7]=c2;
498 }
499
500 void bn_sqr_comba8(BN_ULONG *r, const BN_ULONG *a)
501 {
502 BN_ULONG t1,t2;
503 BN_ULONG c1,c2,c3;
504
505 c1=0;
506 c2=0;
507 c3=0;
508 sqr_add_c(a,0,c1,c2,c3);
509 r[0]=c1;
510 c1=0;
511 sqr_add_c2(a,1,0,c2,c3,c1);
512 r[1]=c2;
513 c2=0;
514 sqr_add_c(a,1,c3,c1,c2);
515 sqr_add_c2(a,2,0,c3,c1,c2);
516 r[2]=c3;
517 c3=0;
518 sqr_add_c2(a,3,0,c1,c2,c3);
519 sqr_add_c2(a,2,1,c1,c2,c3);
520 r[3]=c1;
521 c1=0;
522 sqr_add_c(a,2,c2,c3,c1);
523 sqr_add_c2(a,3,1,c2,c3,c1);
524 sqr_add_c2(a,4,0,c2,c3,c1);
525 r[4]=c2;
526 c2=0;
527 sqr_add_c2(a,5,0,c3,c1,c2);
528 sqr_add_c2(a,4,1,c3,c1,c2);
529 sqr_add_c2(a,3,2,c3,c1,c2);
530 r[5]=c3;
531 c3=0;
532 sqr_add_c(a,3,c1,c2,c3);
533 sqr_add_c2(a,4,2,c1,c2,c3);
534 sqr_add_c2(a,5,1,c1,c2,c3);
535 sqr_add_c2(a,6,0,c1,c2,c3);
536 r[6]=c1;
537 c1=0;
538 sqr_add_c2(a,7,0,c2,c3,c1);
539 sqr_add_c2(a,6,1,c2,c3,c1);
540 sqr_add_c2(a,5,2,c2,c3,c1);
541 sqr_add_c2(a,4,3,c2,c3,c1);
542 r[7]=c2;
543 c2=0;
544 sqr_add_c(a,4,c3,c1,c2);
545 sqr_add_c2(a,5,3,c3,c1,c2);
546 sqr_add_c2(a,6,2,c3,c1,c2);
547 sqr_add_c2(a,7,1,c3,c1,c2);
548 r[8]=c3;
549 c3=0;
550 sqr_add_c2(a,7,2,c1,c2,c3);
551 sqr_add_c2(a,6,3,c1,c2,c3);
552 sqr_add_c2(a,5,4,c1,c2,c3);
553 r[9]=c1;
554 c1=0;
555 sqr_add_c(a,5,c2,c3,c1);
556 sqr_add_c2(a,6,4,c2,c3,c1);
557 sqr_add_c2(a,7,3,c2,c3,c1);
558 r[10]=c2;
559 c2=0;
560 sqr_add_c2(a,7,4,c3,c1,c2);
561 sqr_add_c2(a,6,5,c3,c1,c2);
562 r[11]=c3;
563 c3=0;
564 sqr_add_c(a,6,c1,c2,c3);
565 sqr_add_c2(a,7,5,c1,c2,c3);
566 r[12]=c1;
567 c1=0;
568 sqr_add_c2(a,7,6,c2,c3,c1);
569 r[13]=c2;
570 c2=0;
571 sqr_add_c(a,7,c3,c1,c2);
572 r[14]=c3;
573 r[15]=c1;
574 }
575
576 void bn_sqr_comba4(BN_ULONG *r, const BN_ULONG *a)
577 {
578 BN_ULONG t1,t2;
579 BN_ULONG c1,c2,c3;
580
581 c1=0;
582 c2=0;
583 c3=0;
584 sqr_add_c(a,0,c1,c2,c3);
585 r[0]=c1;
586 c1=0;
587 sqr_add_c2(a,1,0,c2,c3,c1);
588 r[1]=c2;
589 c2=0;
590 sqr_add_c(a,1,c3,c1,c2);
591 sqr_add_c2(a,2,0,c3,c1,c2);
592 r[2]=c3;
593 c3=0;
594 sqr_add_c2(a,3,0,c1,c2,c3);
595 sqr_add_c2(a,2,1,c1,c2,c3);
596 r[3]=c1;
597 c1=0;
598 sqr_add_c(a,2,c2,c3,c1);
599 sqr_add_c2(a,3,1,c2,c3,c1);
600 r[4]=c2;
601 c2=0;
602 sqr_add_c2(a,3,2,c3,c1,c2);
603 r[5]=c3;
604 c3=0;
605 sqr_add_c(a,3,c1,c2,c3);
606 r[6]=c1;
607 r[7]=c2;
608 }
609 #endif
A mirror of the official openssl repository