• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /* x86_64 BIGNUM accelerator version 0.1, December 2002.
2  *
3  * Implemented by Andy Polyakov <appro@fy.chalmers.se> for the OpenSSL
4  * project.
5  *
6  * Rights for redistribution and usage in source and binary forms are
7  * granted according to the OpenSSL license. Warranty of any kind is
8  * disclaimed.
9  *
10  * Q. Version 0.1? It doesn't sound like Andy, he used to assign real
11  *    versions, like 1.0...
12  * A. Well, that's because this code is basically a quick-n-dirty
13  *    proof-of-concept hack. As you can see it's implemented with
14  *    inline assembler, which means that you're bound to GCC and that
15  *    there might be enough room for further improvement.
16  *
17  * Q. Why inline assembler?
18  * A. x86_64 features own ABI which I'm not familiar with. This is
19  *    why I decided to let the compiler take care of subroutine
20  *    prologue/epilogue as well as register allocation. For reference.
21  *    Win64 implements different ABI for AMD64, different from Linux.
22  *
23  * Q. How much faster does it get?
24  * A. 'apps/openssl speed rsa dsa' output with no-asm:
25  *
26  *	                  sign    verify    sign/s verify/s
27  *	rsa  512 bits   0.0006s   0.0001s   1683.8  18456.2
28  *	rsa 1024 bits   0.0028s   0.0002s    356.0   6407.0
29  *	rsa 2048 bits   0.0172s   0.0005s     58.0   1957.8
30  *	rsa 4096 bits   0.1155s   0.0018s      8.7    555.6
31  *	                  sign    verify    sign/s verify/s
32  *	dsa  512 bits   0.0005s   0.0006s   2100.8   1768.3
33  *	dsa 1024 bits   0.0014s   0.0018s    692.3    559.2
34  *	dsa 2048 bits   0.0049s   0.0061s    204.7    165.0
35  *
36  *    'apps/openssl speed rsa dsa' output with this module:
37  *
38  *	                  sign    verify    sign/s verify/s
39  *	rsa  512 bits   0.0004s   0.0000s   2767.1  33297.9
40  *	rsa 1024 bits   0.0012s   0.0001s    867.4  14674.7
41  *	rsa 2048 bits   0.0061s   0.0002s    164.0   5270.0
42  *	rsa 4096 bits   0.0384s   0.0006s     26.1   1650.8
43  *	                  sign    verify    sign/s verify/s
44  *	dsa  512 bits   0.0002s   0.0003s   4442.2   3786.3
45  *	dsa 1024 bits   0.0005s   0.0007s   1835.1   1497.4
46  *	dsa 2048 bits   0.0016s   0.0020s    620.4    504.6
47  *
48  *    For the reference. IA-32 assembler implementation performs
49  *    very much like 64-bit code compiled with no-asm on the same
50  *    machine.
51  */
52 
53 #include <openssl/bn.h>
54 
55 // TODO(davidben): Get this file working on MSVC x64.
56 #if !defined(OPENSSL_NO_ASM) && defined(OPENSSL_X86_64) && \
57     (defined(__GNUC__) || defined(__clang__))
58 
59 #include "../internal.h"
60 
61 
62 #undef mul
63 #undef mul_add
64 
65 // "m"(a), "+m"(r)	is the way to favor DirectPath µ-code;
66 // "g"(0)		let the compiler to decide where does it
67 //			want to keep the value of zero;
68 #define mul_add(r, a, word, carry)                                         \
69   do {                                                                     \
70     register BN_ULONG high, low;                                           \
71     __asm__("mulq %3" : "=a"(low), "=d"(high) : "a"(word), "m"(a) : "cc"); \
72     __asm__("addq %2,%0; adcq %3,%1"                                       \
73             : "+r"(carry), "+d"(high)                                      \
74             : "a"(low), "g"(0)                                             \
75             : "cc");                                                       \
76     __asm__("addq %2,%0; adcq %3,%1"                                       \
77             : "+m"(r), "+d"(high)                                          \
78             : "r"(carry), "g"(0)                                           \
79             : "cc");                                                       \
80     (carry) = high;                                                        \
81   } while (0)
82 
83 #define mul(r, a, word, carry)                                             \
84   do {                                                                     \
85     register BN_ULONG high, low;                                           \
86     __asm__("mulq %3" : "=a"(low), "=d"(high) : "a"(word), "g"(a) : "cc"); \
87     __asm__("addq %2,%0; adcq %3,%1"                                       \
88             : "+r"(carry), "+d"(high)                                      \
89             : "a"(low), "g"(0)                                             \
90             : "cc");                                                       \
91     (r) = (carry);                                                         \
92     (carry) = high;                                                        \
93   } while (0)
94 #undef sqr
95 #define sqr(r0, r1, a) __asm__("mulq %2" : "=a"(r0), "=d"(r1) : "a"(a) : "cc");
96 
bn_mul_add_words(BN_ULONG * rp,const BN_ULONG * ap,size_t num,BN_ULONG w)97 BN_ULONG bn_mul_add_words(BN_ULONG *rp, const BN_ULONG *ap, size_t num,
98                           BN_ULONG w) {
99   BN_ULONG c1 = 0;
100 
101   if (num == 0) {
102     return (c1);
103   }
104 
105   while (num & ~3) {
106     mul_add(rp[0], ap[0], w, c1);
107     mul_add(rp[1], ap[1], w, c1);
108     mul_add(rp[2], ap[2], w, c1);
109     mul_add(rp[3], ap[3], w, c1);
110     ap += 4;
111     rp += 4;
112     num -= 4;
113   }
114   if (num) {
115     mul_add(rp[0], ap[0], w, c1);
116     if (--num == 0) {
117       return c1;
118     }
119     mul_add(rp[1], ap[1], w, c1);
120     if (--num == 0) {
121       return c1;
122     }
123     mul_add(rp[2], ap[2], w, c1);
124     return c1;
125   }
126 
127   return c1;
128 }
129 
bn_mul_words(BN_ULONG * rp,const BN_ULONG * ap,size_t num,BN_ULONG w)130 BN_ULONG bn_mul_words(BN_ULONG *rp, const BN_ULONG *ap, size_t num,
131                       BN_ULONG w) {
132   BN_ULONG c1 = 0;
133 
134   if (num == 0) {
135     return c1;
136   }
137 
138   while (num & ~3) {
139     mul(rp[0], ap[0], w, c1);
140     mul(rp[1], ap[1], w, c1);
141     mul(rp[2], ap[2], w, c1);
142     mul(rp[3], ap[3], w, c1);
143     ap += 4;
144     rp += 4;
145     num -= 4;
146   }
147   if (num) {
148     mul(rp[0], ap[0], w, c1);
149     if (--num == 0) {
150       return c1;
151     }
152     mul(rp[1], ap[1], w, c1);
153     if (--num == 0) {
154       return c1;
155     }
156     mul(rp[2], ap[2], w, c1);
157   }
158   return c1;
159 }
160 
bn_sqr_words(BN_ULONG * r,const BN_ULONG * a,size_t n)161 void bn_sqr_words(BN_ULONG *r, const BN_ULONG *a, size_t n) {
162   if (n == 0) {
163     return;
164   }
165 
166   while (n & ~3) {
167     sqr(r[0], r[1], a[0]);
168     sqr(r[2], r[3], a[1]);
169     sqr(r[4], r[5], a[2]);
170     sqr(r[6], r[7], a[3]);
171     a += 4;
172     r += 8;
173     n -= 4;
174   }
175   if (n) {
176     sqr(r[0], r[1], a[0]);
177     if (--n == 0) {
178       return;
179     }
180     sqr(r[2], r[3], a[1]);
181     if (--n == 0) {
182       return;
183     }
184     sqr(r[4], r[5], a[2]);
185   }
186 }
187 
bn_add_words(BN_ULONG * rp,const BN_ULONG * ap,const BN_ULONG * bp,size_t n)188 BN_ULONG bn_add_words(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp,
189                       size_t n) {
190   BN_ULONG ret;
191   size_t i = 0;
192 
193   if (n == 0) {
194     return 0;
195   }
196 
197   __asm__ volatile (
198       "	subq	%0,%0		\n"  // clear carry
199       "	jmp	1f		\n"
200       ".p2align 4			\n"
201       "1:"
202       "	movq	(%4,%2,8),%0	\n"
203       "	adcq	(%5,%2,8),%0	\n"
204       "	movq	%0,(%3,%2,8)	\n"
205       "	lea	1(%2),%2	\n"
206       "	dec	%1		\n"
207       "	jnz	1b		\n"
208       "	sbbq	%0,%0		\n"
209       : "=&r"(ret), "+c"(n), "+r"(i)
210       : "r"(rp), "r"(ap), "r"(bp)
211       : "cc", "memory");
212 
213   return ret & 1;
214 }
215 
bn_sub_words(BN_ULONG * rp,const BN_ULONG * ap,const BN_ULONG * bp,size_t n)216 BN_ULONG bn_sub_words(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp,
217                       size_t n) {
218   BN_ULONG ret;
219   size_t i = 0;
220 
221   if (n == 0) {
222     return 0;
223   }
224 
225   __asm__ volatile (
226       "	subq	%0,%0		\n"  // clear borrow
227       "	jmp	1f		\n"
228       ".p2align 4			\n"
229       "1:"
230       "	movq	(%4,%2,8),%0	\n"
231       "	sbbq	(%5,%2,8),%0	\n"
232       "	movq	%0,(%3,%2,8)	\n"
233       "	lea	1(%2),%2	\n"
234       "	dec	%1		\n"
235       "	jnz	1b		\n"
236       "	sbbq	%0,%0		\n"
237       : "=&r"(ret), "+c"(n), "+r"(i)
238       : "r"(rp), "r"(ap), "r"(bp)
239       : "cc", "memory");
240 
241   return ret & 1;
242 }
243 
244 // mul_add_c(a,b,c0,c1,c2)  -- c+=a*b for three word number c=(c2,c1,c0)
245 // mul_add_c2(a,b,c0,c1,c2) -- c+=2*a*b for three word number c=(c2,c1,c0)
246 // sqr_add_c(a,i,c0,c1,c2)  -- c+=a[i]^2 for three word number c=(c2,c1,c0)
247 // sqr_add_c2(a,i,c0,c1,c2) -- c+=2*a[i]*a[j] for three word number c=(c2,c1,c0)
248 
249 // Keep in mind that carrying into high part of multiplication result can not
250 // overflow, because it cannot be all-ones.
251 #define mul_add_c(a, b, c0, c1, c2)                                  \
252   do {                                                               \
253     BN_ULONG t1, t2;                                                 \
254     __asm__("mulq %3" : "=a"(t1), "=d"(t2) : "a"(a), "m"(b) : "cc"); \
255     __asm__("addq %3,%0; adcq %4,%1; adcq %5,%2"                     \
256             : "+r"(c0), "+r"(c1), "+r"(c2)                           \
257             : "r"(t1), "r"(t2), "g"(0)                               \
258             : "cc");                                                 \
259   } while (0)
260 
261 #define sqr_add_c(a, i, c0, c1, c2)                               \
262   do {                                                            \
263     BN_ULONG t1, t2;                                              \
264     __asm__("mulq %2" : "=a"(t1), "=d"(t2) : "a"((a)[i]) : "cc"); \
265     __asm__("addq %3,%0; adcq %4,%1; adcq %5,%2"                  \
266             : "+r"(c0), "+r"(c1), "+r"(c2)                        \
267             : "r"(t1), "r"(t2), "g"(0)                            \
268             : "cc");                                              \
269   } while (0)
270 
271 #define mul_add_c2(a, b, c0, c1, c2)                                 \
272   do {                                                               \
273     BN_ULONG t1, t2;                                                 \
274     __asm__("mulq %3" : "=a"(t1), "=d"(t2) : "a"(a), "m"(b) : "cc"); \
275     __asm__("addq %3,%0; adcq %4,%1; adcq %5,%2"                     \
276             : "+r"(c0), "+r"(c1), "+r"(c2)                           \
277             : "r"(t1), "r"(t2), "g"(0)                               \
278             : "cc");                                                 \
279     __asm__("addq %3,%0; adcq %4,%1; adcq %5,%2"                     \
280             : "+r"(c0), "+r"(c1), "+r"(c2)                           \
281             : "r"(t1), "r"(t2), "g"(0)                               \
282             : "cc");                                                 \
283   } while (0)
284 
285 #define sqr_add_c2(a, i, j, c0, c1, c2) mul_add_c2((a)[i], (a)[j], c0, c1, c2)
286 
bn_mul_comba8(BN_ULONG r[16],const BN_ULONG a[8],const BN_ULONG b[8])287 void bn_mul_comba8(BN_ULONG r[16], const BN_ULONG a[8], const BN_ULONG b[8]) {
288   BN_ULONG c1, c2, c3;
289 
290   c1 = 0;
291   c2 = 0;
292   c3 = 0;
293   mul_add_c(a[0], b[0], c1, c2, c3);
294   r[0] = c1;
295   c1 = 0;
296   mul_add_c(a[0], b[1], c2, c3, c1);
297   mul_add_c(a[1], b[0], c2, c3, c1);
298   r[1] = c2;
299   c2 = 0;
300   mul_add_c(a[2], b[0], c3, c1, c2);
301   mul_add_c(a[1], b[1], c3, c1, c2);
302   mul_add_c(a[0], b[2], c3, c1, c2);
303   r[2] = c3;
304   c3 = 0;
305   mul_add_c(a[0], b[3], c1, c2, c3);
306   mul_add_c(a[1], b[2], c1, c2, c3);
307   mul_add_c(a[2], b[1], c1, c2, c3);
308   mul_add_c(a[3], b[0], c1, c2, c3);
309   r[3] = c1;
310   c1 = 0;
311   mul_add_c(a[4], b[0], c2, c3, c1);
312   mul_add_c(a[3], b[1], c2, c3, c1);
313   mul_add_c(a[2], b[2], c2, c3, c1);
314   mul_add_c(a[1], b[3], c2, c3, c1);
315   mul_add_c(a[0], b[4], c2, c3, c1);
316   r[4] = c2;
317   c2 = 0;
318   mul_add_c(a[0], b[5], c3, c1, c2);
319   mul_add_c(a[1], b[4], c3, c1, c2);
320   mul_add_c(a[2], b[3], c3, c1, c2);
321   mul_add_c(a[3], b[2], c3, c1, c2);
322   mul_add_c(a[4], b[1], c3, c1, c2);
323   mul_add_c(a[5], b[0], c3, c1, c2);
324   r[5] = c3;
325   c3 = 0;
326   mul_add_c(a[6], b[0], c1, c2, c3);
327   mul_add_c(a[5], b[1], c1, c2, c3);
328   mul_add_c(a[4], b[2], c1, c2, c3);
329   mul_add_c(a[3], b[3], c1, c2, c3);
330   mul_add_c(a[2], b[4], c1, c2, c3);
331   mul_add_c(a[1], b[5], c1, c2, c3);
332   mul_add_c(a[0], b[6], c1, c2, c3);
333   r[6] = c1;
334   c1 = 0;
335   mul_add_c(a[0], b[7], c2, c3, c1);
336   mul_add_c(a[1], b[6], c2, c3, c1);
337   mul_add_c(a[2], b[5], c2, c3, c1);
338   mul_add_c(a[3], b[4], c2, c3, c1);
339   mul_add_c(a[4], b[3], c2, c3, c1);
340   mul_add_c(a[5], b[2], c2, c3, c1);
341   mul_add_c(a[6], b[1], c2, c3, c1);
342   mul_add_c(a[7], b[0], c2, c3, c1);
343   r[7] = c2;
344   c2 = 0;
345   mul_add_c(a[7], b[1], c3, c1, c2);
346   mul_add_c(a[6], b[2], c3, c1, c2);
347   mul_add_c(a[5], b[3], c3, c1, c2);
348   mul_add_c(a[4], b[4], c3, c1, c2);
349   mul_add_c(a[3], b[5], c3, c1, c2);
350   mul_add_c(a[2], b[6], c3, c1, c2);
351   mul_add_c(a[1], b[7], c3, c1, c2);
352   r[8] = c3;
353   c3 = 0;
354   mul_add_c(a[2], b[7], c1, c2, c3);
355   mul_add_c(a[3], b[6], c1, c2, c3);
356   mul_add_c(a[4], b[5], c1, c2, c3);
357   mul_add_c(a[5], b[4], c1, c2, c3);
358   mul_add_c(a[6], b[3], c1, c2, c3);
359   mul_add_c(a[7], b[2], c1, c2, c3);
360   r[9] = c1;
361   c1 = 0;
362   mul_add_c(a[7], b[3], c2, c3, c1);
363   mul_add_c(a[6], b[4], c2, c3, c1);
364   mul_add_c(a[5], b[5], c2, c3, c1);
365   mul_add_c(a[4], b[6], c2, c3, c1);
366   mul_add_c(a[3], b[7], c2, c3, c1);
367   r[10] = c2;
368   c2 = 0;
369   mul_add_c(a[4], b[7], c3, c1, c2);
370   mul_add_c(a[5], b[6], c3, c1, c2);
371   mul_add_c(a[6], b[5], c3, c1, c2);
372   mul_add_c(a[7], b[4], c3, c1, c2);
373   r[11] = c3;
374   c3 = 0;
375   mul_add_c(a[7], b[5], c1, c2, c3);
376   mul_add_c(a[6], b[6], c1, c2, c3);
377   mul_add_c(a[5], b[7], c1, c2, c3);
378   r[12] = c1;
379   c1 = 0;
380   mul_add_c(a[6], b[7], c2, c3, c1);
381   mul_add_c(a[7], b[6], c2, c3, c1);
382   r[13] = c2;
383   c2 = 0;
384   mul_add_c(a[7], b[7], c3, c1, c2);
385   r[14] = c3;
386   r[15] = c1;
387 }
388 
bn_mul_comba4(BN_ULONG r[8],const BN_ULONG a[4],const BN_ULONG b[4])389 void bn_mul_comba4(BN_ULONG r[8], const BN_ULONG a[4], const BN_ULONG b[4]) {
390   BN_ULONG c1, c2, c3;
391 
392   c1 = 0;
393   c2 = 0;
394   c3 = 0;
395   mul_add_c(a[0], b[0], c1, c2, c3);
396   r[0] = c1;
397   c1 = 0;
398   mul_add_c(a[0], b[1], c2, c3, c1);
399   mul_add_c(a[1], b[0], c2, c3, c1);
400   r[1] = c2;
401   c2 = 0;
402   mul_add_c(a[2], b[0], c3, c1, c2);
403   mul_add_c(a[1], b[1], c3, c1, c2);
404   mul_add_c(a[0], b[2], c3, c1, c2);
405   r[2] = c3;
406   c3 = 0;
407   mul_add_c(a[0], b[3], c1, c2, c3);
408   mul_add_c(a[1], b[2], c1, c2, c3);
409   mul_add_c(a[2], b[1], c1, c2, c3);
410   mul_add_c(a[3], b[0], c1, c2, c3);
411   r[3] = c1;
412   c1 = 0;
413   mul_add_c(a[3], b[1], c2, c3, c1);
414   mul_add_c(a[2], b[2], c2, c3, c1);
415   mul_add_c(a[1], b[3], c2, c3, c1);
416   r[4] = c2;
417   c2 = 0;
418   mul_add_c(a[2], b[3], c3, c1, c2);
419   mul_add_c(a[3], b[2], c3, c1, c2);
420   r[5] = c3;
421   c3 = 0;
422   mul_add_c(a[3], b[3], c1, c2, c3);
423   r[6] = c1;
424   r[7] = c2;
425 }
426 
bn_sqr_comba8(BN_ULONG r[16],const BN_ULONG a[8])427 void bn_sqr_comba8(BN_ULONG r[16], const BN_ULONG a[8]) {
428   BN_ULONG c1, c2, c3;
429 
430   c1 = 0;
431   c2 = 0;
432   c3 = 0;
433   sqr_add_c(a, 0, c1, c2, c3);
434   r[0] = c1;
435   c1 = 0;
436   sqr_add_c2(a, 1, 0, c2, c3, c1);
437   r[1] = c2;
438   c2 = 0;
439   sqr_add_c(a, 1, c3, c1, c2);
440   sqr_add_c2(a, 2, 0, c3, c1, c2);
441   r[2] = c3;
442   c3 = 0;
443   sqr_add_c2(a, 3, 0, c1, c2, c3);
444   sqr_add_c2(a, 2, 1, c1, c2, c3);
445   r[3] = c1;
446   c1 = 0;
447   sqr_add_c(a, 2, c2, c3, c1);
448   sqr_add_c2(a, 3, 1, c2, c3, c1);
449   sqr_add_c2(a, 4, 0, c2, c3, c1);
450   r[4] = c2;
451   c2 = 0;
452   sqr_add_c2(a, 5, 0, c3, c1, c2);
453   sqr_add_c2(a, 4, 1, c3, c1, c2);
454   sqr_add_c2(a, 3, 2, c3, c1, c2);
455   r[5] = c3;
456   c3 = 0;
457   sqr_add_c(a, 3, c1, c2, c3);
458   sqr_add_c2(a, 4, 2, c1, c2, c3);
459   sqr_add_c2(a, 5, 1, c1, c2, c3);
460   sqr_add_c2(a, 6, 0, c1, c2, c3);
461   r[6] = c1;
462   c1 = 0;
463   sqr_add_c2(a, 7, 0, c2, c3, c1);
464   sqr_add_c2(a, 6, 1, c2, c3, c1);
465   sqr_add_c2(a, 5, 2, c2, c3, c1);
466   sqr_add_c2(a, 4, 3, c2, c3, c1);
467   r[7] = c2;
468   c2 = 0;
469   sqr_add_c(a, 4, c3, c1, c2);
470   sqr_add_c2(a, 5, 3, c3, c1, c2);
471   sqr_add_c2(a, 6, 2, c3, c1, c2);
472   sqr_add_c2(a, 7, 1, c3, c1, c2);
473   r[8] = c3;
474   c3 = 0;
475   sqr_add_c2(a, 7, 2, c1, c2, c3);
476   sqr_add_c2(a, 6, 3, c1, c2, c3);
477   sqr_add_c2(a, 5, 4, c1, c2, c3);
478   r[9] = c1;
479   c1 = 0;
480   sqr_add_c(a, 5, c2, c3, c1);
481   sqr_add_c2(a, 6, 4, c2, c3, c1);
482   sqr_add_c2(a, 7, 3, c2, c3, c1);
483   r[10] = c2;
484   c2 = 0;
485   sqr_add_c2(a, 7, 4, c3, c1, c2);
486   sqr_add_c2(a, 6, 5, c3, c1, c2);
487   r[11] = c3;
488   c3 = 0;
489   sqr_add_c(a, 6, c1, c2, c3);
490   sqr_add_c2(a, 7, 5, c1, c2, c3);
491   r[12] = c1;
492   c1 = 0;
493   sqr_add_c2(a, 7, 6, c2, c3, c1);
494   r[13] = c2;
495   c2 = 0;
496   sqr_add_c(a, 7, c3, c1, c2);
497   r[14] = c3;
498   r[15] = c1;
499 }
500 
bn_sqr_comba4(BN_ULONG r[8],const BN_ULONG a[4])501 void bn_sqr_comba4(BN_ULONG r[8], const BN_ULONG a[4]) {
502   BN_ULONG c1, c2, c3;
503 
504   c1 = 0;
505   c2 = 0;
506   c3 = 0;
507   sqr_add_c(a, 0, c1, c2, c3);
508   r[0] = c1;
509   c1 = 0;
510   sqr_add_c2(a, 1, 0, c2, c3, c1);
511   r[1] = c2;
512   c2 = 0;
513   sqr_add_c(a, 1, c3, c1, c2);
514   sqr_add_c2(a, 2, 0, c3, c1, c2);
515   r[2] = c3;
516   c3 = 0;
517   sqr_add_c2(a, 3, 0, c1, c2, c3);
518   sqr_add_c2(a, 2, 1, c1, c2, c3);
519   r[3] = c1;
520   c1 = 0;
521   sqr_add_c(a, 2, c2, c3, c1);
522   sqr_add_c2(a, 3, 1, c2, c3, c1);
523   r[4] = c2;
524   c2 = 0;
525   sqr_add_c2(a, 3, 2, c3, c1, c2);
526   r[5] = c3;
527   c3 = 0;
528   sqr_add_c(a, 3, c1, c2, c3);
529   r[6] = c1;
530   r[7] = c2;
531 }
532 
533 #undef mul_add
534 #undef mul
535 #undef sqr
536 #undef mul_add_c
537 #undef sqr_add_c
538 #undef mul_add_c2
539 #undef sqr_add_c2
540 
541 #endif  // !NO_ASM && X86_64 && (__GNUC__ || __clang__)
542