xref: /openssl/crypto/bn/asm/x86_64-gcc.c (revision 706457b7)
1 /*
2  * Copyright 2002-2018 The OpenSSL Project Authors. All Rights Reserved.
3  *
4  * Licensed under the Apache License 2.0 (the "License").  You may not use
5  * this file except in compliance with the License.  You can obtain a copy
6  * in the file LICENSE in the source distribution or at
7  * https://www.openssl.org/source/license.html
8  */
9 
10 #include "../bn_local.h"
11 #if !(defined(__GNUC__) && __GNUC__>=2)
12 # include "../bn_asm.c"         /* kind of dirty hack for Sun Studio */
13 #else
14 /*-
15  * x86_64 BIGNUM accelerator version 0.1, December 2002.
16  *
17  * Implemented by Andy Polyakov <appro@openssl.org> for the OpenSSL
18  * project.
19  *
20  * Rights for redistribution and usage in source and binary forms are
21  * granted according to the License. Warranty of any kind is disclaimed.
22  *
23  * Q. Version 0.1? It doesn't sound like Andy, he used to assign real
24  *    versions, like 1.0...
25  * A. Well, that's because this code is basically a quick-n-dirty
26  *    proof-of-concept hack. As you can see it's implemented with
27  *    inline assembler, which means that you're bound to GCC and that
28  *    there might be enough room for further improvement.
29  *
30  * Q. Why inline assembler?
31  * A. x86_64 features own ABI which I'm not familiar with. This is
32  *    why I decided to let the compiler take care of subroutine
33  *    prologue/epilogue as well as register allocation. For reference.
34  *    Win64 implements different ABI for AMD64, different from Linux.
35  *
36  * Q. How much faster does it get?
37  * A. 'apps/openssl speed rsa dsa' output with no-asm:
38  *
39  *                        sign    verify    sign/s verify/s
40  *      rsa  512 bits   0.0006s   0.0001s   1683.8  18456.2
41  *      rsa 1024 bits   0.0028s   0.0002s    356.0   6407.0
42  *      rsa 2048 bits   0.0172s   0.0005s     58.0   1957.8
43  *      rsa 4096 bits   0.1155s   0.0018s      8.7    555.6
44  *                        sign    verify    sign/s verify/s
45  *      dsa  512 bits   0.0005s   0.0006s   2100.8   1768.3
46  *      dsa 1024 bits   0.0014s   0.0018s    692.3    559.2
47  *      dsa 2048 bits   0.0049s   0.0061s    204.7    165.0
48  *
49  *    'apps/openssl speed rsa dsa' output with this module:
50  *
51  *                        sign    verify    sign/s verify/s
52  *      rsa  512 bits   0.0004s   0.0000s   2767.1  33297.9
53  *      rsa 1024 bits   0.0012s   0.0001s    867.4  14674.7
54  *      rsa 2048 bits   0.0061s   0.0002s    164.0   5270.0
55  *      rsa 4096 bits   0.0384s   0.0006s     26.1   1650.8
56  *                        sign    verify    sign/s verify/s
57  *      dsa  512 bits   0.0002s   0.0003s   4442.2   3786.3
58  *      dsa 1024 bits   0.0005s   0.0007s   1835.1   1497.4
59  *      dsa 2048 bits   0.0016s   0.0020s    620.4    504.6
60  *
61  *    For the reference. IA-32 assembler implementation performs
62  *    very much like 64-bit code compiled with no-asm on the same
63  *    machine.
64  */
65 
66 # undef mul
67 # undef mul_add
68 
69 /*-
70  * "m"(a), "+m"(r)      is the way to favor DirectPath µ-code;
71  * "g"(0)               let the compiler to decide where does it
72  *                      want to keep the value of zero;
73  */
74 # define mul_add(r,a,word,carry) do {   \
75         register BN_ULONG high,low;     \
76         asm ("mulq %3"                  \
77                 : "=a"(low),"=d"(high)  \
78                 : "a"(word),"m"(a)      \
79                 : "cc");                \
80         asm ("addq %2,%0; adcq %3,%1"   \
81                 : "+r"(carry),"+d"(high)\
82                 : "a"(low),"g"(0)       \
83                 : "cc");                \
84         asm ("addq %2,%0; adcq %3,%1"   \
85                 : "+m"(r),"+d"(high)    \
86                 : "r"(carry),"g"(0)     \
87                 : "cc");                \
88         carry=high;                     \
89         } while (0)
90 
91 # define mul(r,a,word,carry) do {       \
92         register BN_ULONG high,low;     \
93         asm ("mulq %3"                  \
94                 : "=a"(low),"=d"(high)  \
95                 : "a"(word),"g"(a)      \
96                 : "cc");                \
97         asm ("addq %2,%0; adcq %3,%1"   \
98                 : "+r"(carry),"+d"(high)\
99                 : "a"(low),"g"(0)       \
100                 : "cc");                \
101         (r)=carry, carry=high;          \
102         } while (0)
103 # undef sqr
104 # define sqr(r0,r1,a)                   \
105         asm ("mulq %2"                  \
106                 : "=a"(r0),"=d"(r1)     \
107                 : "a"(a)                \
108                 : "cc");
109 
bn_mul_add_words(BN_ULONG * rp,const BN_ULONG * ap,int num,BN_ULONG w)110 BN_ULONG bn_mul_add_words(BN_ULONG *rp, const BN_ULONG *ap, int num,
111                           BN_ULONG w)
112 {
113     BN_ULONG c1 = 0;
114 
115     if (num <= 0)
116         return c1;
117 
118     while (num & ~3) {
119         mul_add(rp[0], ap[0], w, c1);
120         mul_add(rp[1], ap[1], w, c1);
121         mul_add(rp[2], ap[2], w, c1);
122         mul_add(rp[3], ap[3], w, c1);
123         ap += 4;
124         rp += 4;
125         num -= 4;
126     }
127     if (num) {
128         mul_add(rp[0], ap[0], w, c1);
129         if (--num == 0)
130             return c1;
131         mul_add(rp[1], ap[1], w, c1);
132         if (--num == 0)
133             return c1;
134         mul_add(rp[2], ap[2], w, c1);
135         return c1;
136     }
137 
138     return c1;
139 }
140 
bn_mul_words(BN_ULONG * rp,const BN_ULONG * ap,int num,BN_ULONG w)141 BN_ULONG bn_mul_words(BN_ULONG *rp, const BN_ULONG *ap, int num, BN_ULONG w)
142 {
143     BN_ULONG c1 = 0;
144 
145     if (num <= 0)
146         return c1;
147 
148     while (num & ~3) {
149         mul(rp[0], ap[0], w, c1);
150         mul(rp[1], ap[1], w, c1);
151         mul(rp[2], ap[2], w, c1);
152         mul(rp[3], ap[3], w, c1);
153         ap += 4;
154         rp += 4;
155         num -= 4;
156     }
157     if (num) {
158         mul(rp[0], ap[0], w, c1);
159         if (--num == 0)
160             return c1;
161         mul(rp[1], ap[1], w, c1);
162         if (--num == 0)
163             return c1;
164         mul(rp[2], ap[2], w, c1);
165     }
166     return c1;
167 }
168 
bn_sqr_words(BN_ULONG * r,const BN_ULONG * a,int n)169 void bn_sqr_words(BN_ULONG *r, const BN_ULONG *a, int n)
170 {
171     if (n <= 0)
172         return;
173 
174     while (n & ~3) {
175         sqr(r[0], r[1], a[0]);
176         sqr(r[2], r[3], a[1]);
177         sqr(r[4], r[5], a[2]);
178         sqr(r[6], r[7], a[3]);
179         a += 4;
180         r += 8;
181         n -= 4;
182     }
183     if (n) {
184         sqr(r[0], r[1], a[0]);
185         if (--n == 0)
186             return;
187         sqr(r[2], r[3], a[1]);
188         if (--n == 0)
189             return;
190         sqr(r[4], r[5], a[2]);
191     }
192 }
193 
bn_div_words(BN_ULONG h,BN_ULONG l,BN_ULONG d)194 BN_ULONG bn_div_words(BN_ULONG h, BN_ULONG l, BN_ULONG d)
195 {
196     BN_ULONG ret, waste;
197 
198  asm("divq      %4":"=a"(ret), "=d"(waste)
199  :     "a"(l), "d"(h), "r"(d)
200  :     "cc");
201 
202     return ret;
203 }
204 
bn_add_words(BN_ULONG * rp,const BN_ULONG * ap,const BN_ULONG * bp,int n)205 BN_ULONG bn_add_words(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp,
206                       int n)
207 {
208     BN_ULONG ret;
209     size_t i = 0;
210 
211     if (n <= 0)
212         return 0;
213 
214     asm volatile ("       subq    %0,%0           \n" /* clear carry */
215                   "       jmp     1f              \n"
216                   ".p2align 4                     \n"
217                   "1:     movq    (%4,%2,8),%0    \n"
218                   "       adcq    (%5,%2,8),%0    \n"
219                   "       movq    %0,(%3,%2,8)    \n"
220                   "       lea     1(%2),%2        \n"
221                   "       dec     %1              \n"
222                   "       jnz     1b              \n"
223                   "       sbbq    %0,%0           \n"
224                   :"=&r" (ret), "+c"(n), "+r"(i)
225                   :"r"(rp), "r"(ap), "r"(bp)
226                   :"cc", "memory");
227 
228     return ret & 1;
229 }
230 
231 # ifndef SIMICS
bn_sub_words(BN_ULONG * rp,const BN_ULONG * ap,const BN_ULONG * bp,int n)232 BN_ULONG bn_sub_words(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp,
233                       int n)
234 {
235     BN_ULONG ret;
236     size_t i = 0;
237 
238     if (n <= 0)
239         return 0;
240 
241     asm volatile ("       subq    %0,%0           \n" /* clear borrow */
242                   "       jmp     1f              \n"
243                   ".p2align 4                     \n"
244                   "1:     movq    (%4,%2,8),%0    \n"
245                   "       sbbq    (%5,%2,8),%0    \n"
246                   "       movq    %0,(%3,%2,8)    \n"
247                   "       lea     1(%2),%2        \n"
248                   "       dec     %1              \n"
249                   "       jnz     1b              \n"
250                   "       sbbq    %0,%0           \n"
251                   :"=&r" (ret), "+c"(n), "+r"(i)
252                   :"r"(rp), "r"(ap), "r"(bp)
253                   :"cc", "memory");
254 
255     return ret & 1;
256 }
257 # else
258 /* Simics 1.4<7 has buggy sbbq:-( */
259 #  define BN_MASK2 0xffffffffffffffffL
bn_sub_words(BN_ULONG * r,BN_ULONG * a,BN_ULONG * b,int n)260 BN_ULONG bn_sub_words(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b, int n)
261 {
262     BN_ULONG t1, t2;
263     int c = 0;
264 
265     if (n <= 0)
266         return (BN_ULONG)0;
267 
268     for (;;) {
269         t1 = a[0];
270         t2 = b[0];
271         r[0] = (t1 - t2 - c) & BN_MASK2;
272         if (t1 != t2)
273             c = (t1 < t2);
274         if (--n <= 0)
275             break;
276 
277         t1 = a[1];
278         t2 = b[1];
279         r[1] = (t1 - t2 - c) & BN_MASK2;
280         if (t1 != t2)
281             c = (t1 < t2);
282         if (--n <= 0)
283             break;
284 
285         t1 = a[2];
286         t2 = b[2];
287         r[2] = (t1 - t2 - c) & BN_MASK2;
288         if (t1 != t2)
289             c = (t1 < t2);
290         if (--n <= 0)
291             break;
292 
293         t1 = a[3];
294         t2 = b[3];
295         r[3] = (t1 - t2 - c) & BN_MASK2;
296         if (t1 != t2)
297             c = (t1 < t2);
298         if (--n <= 0)
299             break;
300 
301         a += 4;
302         b += 4;
303         r += 4;
304     }
305     return c;
306 }
307 # endif
308 
309 /* mul_add_c(a,b,c0,c1,c2)  -- c+=a*b for three word number c=(c2,c1,c0) */
310 /* mul_add_c2(a,b,c0,c1,c2) -- c+=2*a*b for three word number c=(c2,c1,c0) */
311 /* sqr_add_c(a,i,c0,c1,c2)  -- c+=a[i]^2 for three word number c=(c2,c1,c0) */
312 /*
313  * sqr_add_c2(a,i,c0,c1,c2) -- c+=2*a[i]*a[j] for three word number
314  * c=(c2,c1,c0)
315  */
316 
317 /*
318  * Keep in mind that carrying into high part of multiplication result
319  * can not overflow, because it cannot be all-ones.
320  */
321 # if 0
322 /* original macros are kept for reference purposes */
323 #  define mul_add_c(a,b,c0,c1,c2)       do {    \
324         BN_ULONG ta = (a), tb = (b);            \
325         BN_ULONG lo, hi;                        \
326         BN_UMULT_LOHI(lo,hi,ta,tb);             \
327         c0 += lo; hi += (c0<lo)?1:0;            \
328         c1 += hi; c2 += (c1<hi)?1:0;            \
329         } while(0)
330 
331 #  define mul_add_c2(a,b,c0,c1,c2)      do {    \
332         BN_ULONG ta = (a), tb = (b);            \
333         BN_ULONG lo, hi, tt;                    \
334         BN_UMULT_LOHI(lo,hi,ta,tb);             \
335         c0 += lo; tt = hi+((c0<lo)?1:0);        \
336         c1 += tt; c2 += (c1<tt)?1:0;            \
337         c0 += lo; hi += (c0<lo)?1:0;            \
338         c1 += hi; c2 += (c1<hi)?1:0;            \
339         } while(0)
340 
341 #  define sqr_add_c(a,i,c0,c1,c2)       do {    \
342         BN_ULONG ta = (a)[i];                   \
343         BN_ULONG lo, hi;                        \
344         BN_UMULT_LOHI(lo,hi,ta,ta);             \
345         c0 += lo; hi += (c0<lo)?1:0;            \
346         c1 += hi; c2 += (c1<hi)?1:0;            \
347         } while(0)
348 # else
349 #  define mul_add_c(a,b,c0,c1,c2) do {  \
350         BN_ULONG t1,t2;                 \
351         asm ("mulq %3"                  \
352                 : "=a"(t1),"=d"(t2)     \
353                 : "a"(a),"m"(b)         \
354                 : "cc");                \
355         asm ("addq %3,%0; adcq %4,%1; adcq %5,%2"       \
356                 : "+r"(c0),"+r"(c1),"+r"(c2)            \
357                 : "r"(t1),"r"(t2),"g"(0)                \
358                 : "cc");                                \
359         } while (0)
360 
361 #  define sqr_add_c(a,i,c0,c1,c2) do {  \
362         BN_ULONG t1,t2;                 \
363         asm ("mulq %2"                  \
364                 : "=a"(t1),"=d"(t2)     \
365                 : "a"(a[i])             \
366                 : "cc");                \
367         asm ("addq %3,%0; adcq %4,%1; adcq %5,%2"       \
368                 : "+r"(c0),"+r"(c1),"+r"(c2)            \
369                 : "r"(t1),"r"(t2),"g"(0)                \
370                 : "cc");                                \
371         } while (0)
372 
373 #  define mul_add_c2(a,b,c0,c1,c2) do { \
374         BN_ULONG t1,t2;                 \
375         asm ("mulq %3"                  \
376                 : "=a"(t1),"=d"(t2)     \
377                 : "a"(a),"m"(b)         \
378                 : "cc");                \
379         asm ("addq %3,%0; adcq %4,%1; adcq %5,%2"       \
380                 : "+r"(c0),"+r"(c1),"+r"(c2)            \
381                 : "r"(t1),"r"(t2),"g"(0)                \
382                 : "cc");                                \
383         asm ("addq %3,%0; adcq %4,%1; adcq %5,%2"       \
384                 : "+r"(c0),"+r"(c1),"+r"(c2)            \
385                 : "r"(t1),"r"(t2),"g"(0)                \
386                 : "cc");                                \
387         } while (0)
388 # endif
389 
390 # define sqr_add_c2(a,i,j,c0,c1,c2)      \
391         mul_add_c2((a)[i],(a)[j],c0,c1,c2)
392 
bn_mul_comba8(BN_ULONG * r,BN_ULONG * a,BN_ULONG * b)393 void bn_mul_comba8(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b)
394 {
395     BN_ULONG c1, c2, c3;
396 
397     c1 = 0;
398     c2 = 0;
399     c3 = 0;
400     mul_add_c(a[0], b[0], c1, c2, c3);
401     r[0] = c1;
402     c1 = 0;
403     mul_add_c(a[0], b[1], c2, c3, c1);
404     mul_add_c(a[1], b[0], c2, c3, c1);
405     r[1] = c2;
406     c2 = 0;
407     mul_add_c(a[2], b[0], c3, c1, c2);
408     mul_add_c(a[1], b[1], c3, c1, c2);
409     mul_add_c(a[0], b[2], c3, c1, c2);
410     r[2] = c3;
411     c3 = 0;
412     mul_add_c(a[0], b[3], c1, c2, c3);
413     mul_add_c(a[1], b[2], c1, c2, c3);
414     mul_add_c(a[2], b[1], c1, c2, c3);
415     mul_add_c(a[3], b[0], c1, c2, c3);
416     r[3] = c1;
417     c1 = 0;
418     mul_add_c(a[4], b[0], c2, c3, c1);
419     mul_add_c(a[3], b[1], c2, c3, c1);
420     mul_add_c(a[2], b[2], c2, c3, c1);
421     mul_add_c(a[1], b[3], c2, c3, c1);
422     mul_add_c(a[0], b[4], c2, c3, c1);
423     r[4] = c2;
424     c2 = 0;
425     mul_add_c(a[0], b[5], c3, c1, c2);
426     mul_add_c(a[1], b[4], c3, c1, c2);
427     mul_add_c(a[2], b[3], c3, c1, c2);
428     mul_add_c(a[3], b[2], c3, c1, c2);
429     mul_add_c(a[4], b[1], c3, c1, c2);
430     mul_add_c(a[5], b[0], c3, c1, c2);
431     r[5] = c3;
432     c3 = 0;
433     mul_add_c(a[6], b[0], c1, c2, c3);
434     mul_add_c(a[5], b[1], c1, c2, c3);
435     mul_add_c(a[4], b[2], c1, c2, c3);
436     mul_add_c(a[3], b[3], c1, c2, c3);
437     mul_add_c(a[2], b[4], c1, c2, c3);
438     mul_add_c(a[1], b[5], c1, c2, c3);
439     mul_add_c(a[0], b[6], c1, c2, c3);
440     r[6] = c1;
441     c1 = 0;
442     mul_add_c(a[0], b[7], c2, c3, c1);
443     mul_add_c(a[1], b[6], c2, c3, c1);
444     mul_add_c(a[2], b[5], c2, c3, c1);
445     mul_add_c(a[3], b[4], c2, c3, c1);
446     mul_add_c(a[4], b[3], c2, c3, c1);
447     mul_add_c(a[5], b[2], c2, c3, c1);
448     mul_add_c(a[6], b[1], c2, c3, c1);
449     mul_add_c(a[7], b[0], c2, c3, c1);
450     r[7] = c2;
451     c2 = 0;
452     mul_add_c(a[7], b[1], c3, c1, c2);
453     mul_add_c(a[6], b[2], c3, c1, c2);
454     mul_add_c(a[5], b[3], c3, c1, c2);
455     mul_add_c(a[4], b[4], c3, c1, c2);
456     mul_add_c(a[3], b[5], c3, c1, c2);
457     mul_add_c(a[2], b[6], c3, c1, c2);
458     mul_add_c(a[1], b[7], c3, c1, c2);
459     r[8] = c3;
460     c3 = 0;
461     mul_add_c(a[2], b[7], c1, c2, c3);
462     mul_add_c(a[3], b[6], c1, c2, c3);
463     mul_add_c(a[4], b[5], c1, c2, c3);
464     mul_add_c(a[5], b[4], c1, c2, c3);
465     mul_add_c(a[6], b[3], c1, c2, c3);
466     mul_add_c(a[7], b[2], c1, c2, c3);
467     r[9] = c1;
468     c1 = 0;
469     mul_add_c(a[7], b[3], c2, c3, c1);
470     mul_add_c(a[6], b[4], c2, c3, c1);
471     mul_add_c(a[5], b[5], c2, c3, c1);
472     mul_add_c(a[4], b[6], c2, c3, c1);
473     mul_add_c(a[3], b[7], c2, c3, c1);
474     r[10] = c2;
475     c2 = 0;
476     mul_add_c(a[4], b[7], c3, c1, c2);
477     mul_add_c(a[5], b[6], c3, c1, c2);
478     mul_add_c(a[6], b[5], c3, c1, c2);
479     mul_add_c(a[7], b[4], c3, c1, c2);
480     r[11] = c3;
481     c3 = 0;
482     mul_add_c(a[7], b[5], c1, c2, c3);
483     mul_add_c(a[6], b[6], c1, c2, c3);
484     mul_add_c(a[5], b[7], c1, c2, c3);
485     r[12] = c1;
486     c1 = 0;
487     mul_add_c(a[6], b[7], c2, c3, c1);
488     mul_add_c(a[7], b[6], c2, c3, c1);
489     r[13] = c2;
490     c2 = 0;
491     mul_add_c(a[7], b[7], c3, c1, c2);
492     r[14] = c3;
493     r[15] = c1;
494 }
495 
bn_mul_comba4(BN_ULONG * r,BN_ULONG * a,BN_ULONG * b)496 void bn_mul_comba4(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b)
497 {
498     BN_ULONG c1, c2, c3;
499 
500     c1 = 0;
501     c2 = 0;
502     c3 = 0;
503     mul_add_c(a[0], b[0], c1, c2, c3);
504     r[0] = c1;
505     c1 = 0;
506     mul_add_c(a[0], b[1], c2, c3, c1);
507     mul_add_c(a[1], b[0], c2, c3, c1);
508     r[1] = c2;
509     c2 = 0;
510     mul_add_c(a[2], b[0], c3, c1, c2);
511     mul_add_c(a[1], b[1], c3, c1, c2);
512     mul_add_c(a[0], b[2], c3, c1, c2);
513     r[2] = c3;
514     c3 = 0;
515     mul_add_c(a[0], b[3], c1, c2, c3);
516     mul_add_c(a[1], b[2], c1, c2, c3);
517     mul_add_c(a[2], b[1], c1, c2, c3);
518     mul_add_c(a[3], b[0], c1, c2, c3);
519     r[3] = c1;
520     c1 = 0;
521     mul_add_c(a[3], b[1], c2, c3, c1);
522     mul_add_c(a[2], b[2], c2, c3, c1);
523     mul_add_c(a[1], b[3], c2, c3, c1);
524     r[4] = c2;
525     c2 = 0;
526     mul_add_c(a[2], b[3], c3, c1, c2);
527     mul_add_c(a[3], b[2], c3, c1, c2);
528     r[5] = c3;
529     c3 = 0;
530     mul_add_c(a[3], b[3], c1, c2, c3);
531     r[6] = c1;
532     r[7] = c2;
533 }
534 
bn_sqr_comba8(BN_ULONG * r,const BN_ULONG * a)535 void bn_sqr_comba8(BN_ULONG *r, const BN_ULONG *a)
536 {
537     BN_ULONG c1, c2, c3;
538 
539     c1 = 0;
540     c2 = 0;
541     c3 = 0;
542     sqr_add_c(a, 0, c1, c2, c3);
543     r[0] = c1;
544     c1 = 0;
545     sqr_add_c2(a, 1, 0, c2, c3, c1);
546     r[1] = c2;
547     c2 = 0;
548     sqr_add_c(a, 1, c3, c1, c2);
549     sqr_add_c2(a, 2, 0, c3, c1, c2);
550     r[2] = c3;
551     c3 = 0;
552     sqr_add_c2(a, 3, 0, c1, c2, c3);
553     sqr_add_c2(a, 2, 1, c1, c2, c3);
554     r[3] = c1;
555     c1 = 0;
556     sqr_add_c(a, 2, c2, c3, c1);
557     sqr_add_c2(a, 3, 1, c2, c3, c1);
558     sqr_add_c2(a, 4, 0, c2, c3, c1);
559     r[4] = c2;
560     c2 = 0;
561     sqr_add_c2(a, 5, 0, c3, c1, c2);
562     sqr_add_c2(a, 4, 1, c3, c1, c2);
563     sqr_add_c2(a, 3, 2, c3, c1, c2);
564     r[5] = c3;
565     c3 = 0;
566     sqr_add_c(a, 3, c1, c2, c3);
567     sqr_add_c2(a, 4, 2, c1, c2, c3);
568     sqr_add_c2(a, 5, 1, c1, c2, c3);
569     sqr_add_c2(a, 6, 0, c1, c2, c3);
570     r[6] = c1;
571     c1 = 0;
572     sqr_add_c2(a, 7, 0, c2, c3, c1);
573     sqr_add_c2(a, 6, 1, c2, c3, c1);
574     sqr_add_c2(a, 5, 2, c2, c3, c1);
575     sqr_add_c2(a, 4, 3, c2, c3, c1);
576     r[7] = c2;
577     c2 = 0;
578     sqr_add_c(a, 4, c3, c1, c2);
579     sqr_add_c2(a, 5, 3, c3, c1, c2);
580     sqr_add_c2(a, 6, 2, c3, c1, c2);
581     sqr_add_c2(a, 7, 1, c3, c1, c2);
582     r[8] = c3;
583     c3 = 0;
584     sqr_add_c2(a, 7, 2, c1, c2, c3);
585     sqr_add_c2(a, 6, 3, c1, c2, c3);
586     sqr_add_c2(a, 5, 4, c1, c2, c3);
587     r[9] = c1;
588     c1 = 0;
589     sqr_add_c(a, 5, c2, c3, c1);
590     sqr_add_c2(a, 6, 4, c2, c3, c1);
591     sqr_add_c2(a, 7, 3, c2, c3, c1);
592     r[10] = c2;
593     c2 = 0;
594     sqr_add_c2(a, 7, 4, c3, c1, c2);
595     sqr_add_c2(a, 6, 5, c3, c1, c2);
596     r[11] = c3;
597     c3 = 0;
598     sqr_add_c(a, 6, c1, c2, c3);
599     sqr_add_c2(a, 7, 5, c1, c2, c3);
600     r[12] = c1;
601     c1 = 0;
602     sqr_add_c2(a, 7, 6, c2, c3, c1);
603     r[13] = c2;
604     c2 = 0;
605     sqr_add_c(a, 7, c3, c1, c2);
606     r[14] = c3;
607     r[15] = c1;
608 }
609 
bn_sqr_comba4(BN_ULONG * r,const BN_ULONG * a)610 void bn_sqr_comba4(BN_ULONG *r, const BN_ULONG *a)
611 {
612     BN_ULONG c1, c2, c3;
613 
614     c1 = 0;
615     c2 = 0;
616     c3 = 0;
617     sqr_add_c(a, 0, c1, c2, c3);
618     r[0] = c1;
619     c1 = 0;
620     sqr_add_c2(a, 1, 0, c2, c3, c1);
621     r[1] = c2;
622     c2 = 0;
623     sqr_add_c(a, 1, c3, c1, c2);
624     sqr_add_c2(a, 2, 0, c3, c1, c2);
625     r[2] = c3;
626     c3 = 0;
627     sqr_add_c2(a, 3, 0, c1, c2, c3);
628     sqr_add_c2(a, 2, 1, c1, c2, c3);
629     r[3] = c1;
630     c1 = 0;
631     sqr_add_c(a, 2, c2, c3, c1);
632     sqr_add_c2(a, 3, 1, c2, c3, c1);
633     r[4] = c2;
634     c2 = 0;
635     sqr_add_c2(a, 3, 2, c3, c1, c2);
636     r[5] = c3;
637     c3 = 0;
638     sqr_add_c(a, 3, c1, c2, c3);
639     r[6] = c1;
640     r[7] = c2;
641 }
642 #endif
643