1/*
2 * Copyright 2009-2015 Samy Al Bahra.
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 *    notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 *    notice, this list of conditions and the following disclaimer in the
12 *    documentation and/or other materials provided with the distribution.
13 *
14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17 * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24 * SUCH DAMAGE.
25 */
26
27#ifndef CK_PR_X86_64_H
28#define CK_PR_X86_64_H
29
30#ifndef CK_PR_H
31#error Do not include this file directly, use ck_pr.h
32#endif
33
34#include <ck_cc.h>
35#include <ck_md.h>
36#include <ck_stdint.h>
37
38/*
39 * The following represent supported atomic operations.
40 * These operations may be emulated.
41 */
42#include "ck_f_pr.h"
43
44/*
45 * Support for TSX extensions.
46 */
47#ifdef CK_MD_RTM_ENABLE
48#include "ck_pr_rtm.h"
49#endif
50
51/* Minimum requirements for the CK_PR interface are met. */
52#define CK_F_PR
53
54#ifdef CK_MD_UMP
55#define CK_PR_LOCK_PREFIX
56#else
57#define CK_PR_LOCK_PREFIX "lock "
58#endif
59
60/*
61 * Prevent speculative execution in busy-wait loops (P4 <=) or "predefined
62 * delay".
63 */
64CK_CC_INLINE static void
65ck_pr_stall(void)
66{
67	__asm__ __volatile__("pause" ::: "memory");
68	return;
69}
70
71#define CK_PR_FENCE(T, I)				\
72	CK_CC_INLINE static void			\
73	ck_pr_fence_strict_##T(void)			\
74	{						\
75		__asm__ __volatile__(I ::: "memory");	\
76	}
77
78/* Atomic operations are always serializing. */
79CK_PR_FENCE(atomic, "")
80CK_PR_FENCE(atomic_store, "")
81CK_PR_FENCE(atomic_load, "")
82CK_PR_FENCE(store_atomic, "")
83CK_PR_FENCE(load_atomic, "")
84
85/* Traditional fence interface. */
86CK_PR_FENCE(load, "lfence")
87CK_PR_FENCE(load_store, "mfence")
88CK_PR_FENCE(store, "sfence")
89CK_PR_FENCE(store_load, "mfence")
90CK_PR_FENCE(memory, "mfence")
91
92/* Below are stdatomic-style fences. */
93
94/*
95 * Provides load-store and store-store ordering. However, Intel specifies that
96 * the WC memory model is relaxed. It is likely an sfence *is* sufficient (in
97 * particular, stores are not re-ordered with respect to prior loads and it is
98 * really just the stores that are subject to re-ordering). However, we take
99 * the conservative route as the manuals are too ambiguous for my taste.
100 */
101CK_PR_FENCE(release, "mfence")
102
103/*
104 * Provides load-load and load-store ordering. The lfence instruction ensures
105 * all prior load operations are complete before any subsequent instructions
106 * actually begin execution. However, the manual also ends up going to describe
107 * WC memory as a relaxed model.
108 */
109CK_PR_FENCE(acquire, "mfence")
110
111CK_PR_FENCE(acqrel, "mfence")
112CK_PR_FENCE(lock, "mfence")
113CK_PR_FENCE(unlock, "mfence")
114
115#undef CK_PR_FENCE
116
117/*
118 * Read for ownership. Older compilers will generate the 32-bit
119 * 3DNow! variant which is binary compatible with x86-64 variant
120 * of prefetchw.
121 */
122#ifndef CK_F_PR_RFO
123#define CK_F_PR_RFO
124CK_CC_INLINE static void
125ck_pr_rfo(const void *m)
126{
127
128	__asm__ __volatile__("prefetchw (%0)"
129	    :
130	    : "r" (m)
131	    : "memory");
132
133	return;
134}
135#endif /* CK_F_PR_RFO */
136
137/*
138 * Atomic fetch-and-store operations.
139 */
140#define CK_PR_FAS(S, M, T, C, I)				\
141	CK_CC_INLINE static T					\
142	ck_pr_fas_##S(M *target, T v)				\
143	{							\
144		__asm__ __volatile__(I " %0, %1"		\
145					: "+m" (*(C *)target),	\
146					  "+q" (v)		\
147					:			\
148					: "memory");		\
149		return v;					\
150	}
151
152CK_PR_FAS(ptr, void, void *, char, "xchgq")
153
154#define CK_PR_FAS_S(S, T, I) CK_PR_FAS(S, T, T, T, I)
155
156#ifndef CK_PR_DISABLE_DOUBLE
157CK_PR_FAS_S(double, double, "xchgq")
158#endif
159CK_PR_FAS_S(char, char, "xchgb")
160CK_PR_FAS_S(uint, unsigned int, "xchgl")
161CK_PR_FAS_S(int, int, "xchgl")
162CK_PR_FAS_S(64, uint64_t, "xchgq")
163CK_PR_FAS_S(32, uint32_t, "xchgl")
164CK_PR_FAS_S(16, uint16_t, "xchgw")
165CK_PR_FAS_S(8,  uint8_t,  "xchgb")
166
167#undef CK_PR_FAS_S
168#undef CK_PR_FAS
169
170/*
171 * Atomic load-from-memory operations.
172 */
173#define CK_PR_LOAD(S, M, T, C, I)				\
174	CK_CC_INLINE static T					\
175	ck_pr_md_load_##S(const M *target)			\
176	{							\
177		T r;						\
178		__asm__ __volatile__(I " %1, %0"		\
179		    : "=q" (r)					\
180		    : "m"  (*(const C *)target)			\
181		    : "memory");				\
182		return (r);					\
183	}
184
185CK_PR_LOAD(ptr, void, void *, char, "movq")
186
187#define CK_PR_LOAD_S(S, T, I) CK_PR_LOAD(S, T, T, T, I)
188
189CK_PR_LOAD_S(char, char, "movb")
190CK_PR_LOAD_S(uint, unsigned int, "movl")
191CK_PR_LOAD_S(int, int, "movl")
192#ifndef CK_PR_DISABLE_DOUBLE
193CK_PR_LOAD_S(double, double, "movq")
194#endif
195CK_PR_LOAD_S(64, uint64_t, "movq")
196CK_PR_LOAD_S(32, uint32_t, "movl")
197CK_PR_LOAD_S(16, uint16_t, "movw")
198CK_PR_LOAD_S(8,  uint8_t,  "movb")
199
200#undef CK_PR_LOAD_S
201#undef CK_PR_LOAD
202
203CK_CC_INLINE static void
204ck_pr_load_64_2(const uint64_t target[2], uint64_t v[2])
205{
206	__asm__ __volatile__("movq %%rdx, %%rcx;"
207			     "movq %%rax, %%rbx;"
208			     CK_PR_LOCK_PREFIX "cmpxchg16b %2;"
209				: "=a" (v[0]),
210				  "=d" (v[1])
211				: "m" (*(const uint64_t *)target)
212				: "rbx", "rcx", "memory", "cc");
213	return;
214}
215
216CK_CC_INLINE static void
217ck_pr_load_ptr_2(const void *t, void *v)
218{
219	ck_pr_load_64_2(CK_CPP_CAST(const uint64_t *, t),
220			CK_CPP_CAST(uint64_t *, v));
221	return;
222}
223
224#define CK_PR_LOAD_2(S, W, T)							\
225	CK_CC_INLINE static void						\
226	ck_pr_md_load_##S##_##W(const T t[2], T v[2])				\
227	{									\
228		ck_pr_load_64_2((const uint64_t *)(const void *)t,		\
229				(uint64_t *)(void *)v);				\
230		return;								\
231	}
232
233CK_PR_LOAD_2(char, 16, char)
234CK_PR_LOAD_2(int, 4, int)
235CK_PR_LOAD_2(uint, 4, unsigned int)
236CK_PR_LOAD_2(32, 4, uint32_t)
237CK_PR_LOAD_2(16, 8, uint16_t)
238CK_PR_LOAD_2(8, 16, uint8_t)
239
240#undef CK_PR_LOAD_2
241
242/*
243 * Atomic store-to-memory operations.
244 */
245#define CK_PR_STORE_IMM(S, M, T, C, I, K)				\
246	CK_CC_INLINE static void					\
247	ck_pr_md_store_##S(M *target, T v)				\
248	{								\
249		__asm__ __volatile__(I " %1, %0"			\
250					: "=m" (*(C *)target)		\
251					: K "q" (v)			\
252					: "memory");			\
253		return;							\
254	}
255
256#define CK_PR_STORE(S, M, T, C, I)				\
257	CK_CC_INLINE static void				\
258	ck_pr_md_store_##S(M *target, T v)			\
259	{							\
260		__asm__ __volatile__(I " %1, %0"		\
261					: "=m" (*(C *)target)	\
262					: "q" (v)		\
263					: "memory");		\
264		return;						\
265	}
266
267CK_PR_STORE_IMM(ptr, void, const void *, char, "movq", CK_CC_IMM_U32)
268#ifndef CK_PR_DISABLE_DOUBLE
269CK_PR_STORE(double, double, double, double, "movq")
270#endif
271
272#define CK_PR_STORE_S(S, T, I, K) CK_PR_STORE_IMM(S, T, T, T, I, K)
273
274CK_PR_STORE_S(char, char, "movb", CK_CC_IMM_S32)
275CK_PR_STORE_S(int, int, "movl", CK_CC_IMM_S32)
276CK_PR_STORE_S(uint, unsigned int, "movl", CK_CC_IMM_U32)
277CK_PR_STORE_S(64, uint64_t, "movq", CK_CC_IMM_U32)
278CK_PR_STORE_S(32, uint32_t, "movl", CK_CC_IMM_U32)
279CK_PR_STORE_S(16, uint16_t, "movw", CK_CC_IMM_U32)
280CK_PR_STORE_S(8,  uint8_t, "movb", CK_CC_IMM_U32)
281
282#undef CK_PR_STORE_S
283#undef CK_PR_STORE_IMM
284#undef CK_PR_STORE
285
286/*
287 * Atomic fetch-and-add operations.
288 */
289#define CK_PR_FAA(S, M, T, C, I)					\
290	CK_CC_INLINE static T						\
291	ck_pr_faa_##S(M *target, T d)					\
292	{								\
293		__asm__ __volatile__(CK_PR_LOCK_PREFIX I " %1, %0"	\
294					: "+m" (*(C *)target),		\
295					  "+q" (d)			\
296					:				\
297					: "memory", "cc");		\
298		return (d);						\
299	}
300
301CK_PR_FAA(ptr, void, uintptr_t, char, "xaddq")
302
303#define CK_PR_FAA_S(S, T, I) CK_PR_FAA(S, T, T, T, I)
304
305CK_PR_FAA_S(char, char, "xaddb")
306CK_PR_FAA_S(uint, unsigned int, "xaddl")
307CK_PR_FAA_S(int, int, "xaddl")
308CK_PR_FAA_S(64, uint64_t, "xaddq")
309CK_PR_FAA_S(32, uint32_t, "xaddl")
310CK_PR_FAA_S(16, uint16_t, "xaddw")
311CK_PR_FAA_S(8,  uint8_t,  "xaddb")
312
313#undef CK_PR_FAA_S
314#undef CK_PR_FAA
315
316/*
317 * Atomic store-only unary operations.
318 */
319#define CK_PR_UNARY(K, S, T, C, I)				\
320	CK_PR_UNARY_R(K, S, T, C, I)				\
321	CK_PR_UNARY_V(K, S, T, C, I)
322
323#define CK_PR_UNARY_R(K, S, T, C, I)				\
324	CK_CC_INLINE static void				\
325	ck_pr_##K##_##S(T *target)				\
326	{							\
327		__asm__ __volatile__(CK_PR_LOCK_PREFIX I " %0"	\
328					: "+m" (*(C *)target)	\
329					:			\
330					: "memory", "cc");	\
331		return;						\
332	}
333
334#define CK_PR_UNARY_V(K, S, T, C, I)					\
335	CK_CC_INLINE static bool					\
336	ck_pr_##K##_##S##_is_zero(T *target)				\
337	{								\
338		bool ret;						\
339		__asm__ __volatile__(CK_PR_LOCK_PREFIX I " %0; setz %1"	\
340					: "+m" (*(C *)target),		\
341					  "=rm" (ret)			\
342					:				\
343					: "memory", "cc");		\
344		return ret;						\
345	}
346
347#define CK_PR_UNARY_S(K, S, T, I) CK_PR_UNARY(K, S, T, T, I)
348
349#define CK_PR_GENERATE(K)				\
350	CK_PR_UNARY(K, ptr, void, char, #K "q") 	\
351	CK_PR_UNARY_S(K, char, char, #K "b")		\
352	CK_PR_UNARY_S(K, int, int, #K "l")		\
353	CK_PR_UNARY_S(K, uint, unsigned int, #K "l")	\
354	CK_PR_UNARY_S(K, 64, uint64_t, #K "q")		\
355	CK_PR_UNARY_S(K, 32, uint32_t, #K "l")		\
356	CK_PR_UNARY_S(K, 16, uint16_t, #K "w")		\
357	CK_PR_UNARY_S(K, 8, uint8_t, #K "b")
358
359CK_PR_GENERATE(inc)
360CK_PR_GENERATE(dec)
361CK_PR_GENERATE(neg)
362
363/* not does not affect condition flags. */
364#undef CK_PR_UNARY_V
365#define CK_PR_UNARY_V(a, b, c, d, e)
366CK_PR_GENERATE(not)
367
368#undef CK_PR_GENERATE
369#undef CK_PR_UNARY_S
370#undef CK_PR_UNARY_V
371#undef CK_PR_UNARY_R
372#undef CK_PR_UNARY
373
374/*
375 * Atomic store-only binary operations.
376 */
377#define CK_PR_BINARY(K, S, M, T, C, I, O)				\
378	CK_CC_INLINE static void					\
379	ck_pr_##K##_##S(M *target, T d)					\
380	{								\
381		__asm__ __volatile__(CK_PR_LOCK_PREFIX I " %1, %0"	\
382					: "+m" (*(C *)target)		\
383					: O "q" (d)			\
384					: "memory", "cc");		\
385		return;							\
386	}
387
388#define CK_PR_BINARY_S(K, S, T, I, O) CK_PR_BINARY(K, S, T, T, T, I, O)
389
390#define CK_PR_GENERATE(K)							\
391	CK_PR_BINARY(K, ptr, void, uintptr_t, char, #K "q", CK_CC_IMM_U32)	\
392	CK_PR_BINARY_S(K, char, char, #K "b", CK_CC_IMM_S32)			\
393	CK_PR_BINARY_S(K, int, int, #K "l", CK_CC_IMM_S32)			\
394	CK_PR_BINARY_S(K, uint, unsigned int, #K "l", CK_CC_IMM_U32)		\
395	CK_PR_BINARY_S(K, 64, uint64_t, #K "q", CK_CC_IMM_U32)			\
396	CK_PR_BINARY_S(K, 32, uint32_t, #K "l", CK_CC_IMM_U32)			\
397	CK_PR_BINARY_S(K, 16, uint16_t, #K "w", CK_CC_IMM_U32)			\
398	CK_PR_BINARY_S(K, 8, uint8_t, #K "b", CK_CC_IMM_U32)
399
400CK_PR_GENERATE(add)
401CK_PR_GENERATE(sub)
402CK_PR_GENERATE(and)
403CK_PR_GENERATE(or)
404CK_PR_GENERATE(xor)
405
406#undef CK_PR_GENERATE
407#undef CK_PR_BINARY_S
408#undef CK_PR_BINARY
409
410/*
411 * Atomic compare and swap.
412 */
413#define CK_PR_CAS(S, M, T, C, I)						\
414	CK_CC_INLINE static bool						\
415	ck_pr_cas_##S(M *target, T compare, T set)				\
416	{									\
417		bool z;								\
418		__asm__ __volatile__(CK_PR_LOCK_PREFIX I " %2, %0; setz %1"	\
419					: "+m"  (*(C *)target),			\
420					  "=a"  (z)				\
421					: "q"   (set),				\
422					  "a"   (compare)			\
423					: "memory", "cc");			\
424		return z;							\
425	}
426
427CK_PR_CAS(ptr, void, void *, char, "cmpxchgq")
428
429#define CK_PR_CAS_S(S, T, I) CK_PR_CAS(S, T, T, T, I)
430
431CK_PR_CAS_S(char, char, "cmpxchgb")
432CK_PR_CAS_S(int, int, "cmpxchgl")
433CK_PR_CAS_S(uint, unsigned int, "cmpxchgl")
434#ifndef CK_PR_DISABLE_DOUBLE
435CK_PR_CAS_S(double, double, "cmpxchgq")
436#endif
437CK_PR_CAS_S(64, uint64_t, "cmpxchgq")
438CK_PR_CAS_S(32, uint32_t, "cmpxchgl")
439CK_PR_CAS_S(16, uint16_t, "cmpxchgw")
440CK_PR_CAS_S(8,  uint8_t,  "cmpxchgb")
441
442#undef CK_PR_CAS_S
443#undef CK_PR_CAS
444
445/*
446 * Compare and swap, set *v to old value of target.
447 */
448#define CK_PR_CAS_O(S, M, T, C, I, R)						\
449	CK_CC_INLINE static bool						\
450	ck_pr_cas_##S##_value(M *target, T compare, T set, M *v)		\
451	{									\
452		bool z;								\
453		__asm__ __volatile__(CK_PR_LOCK_PREFIX "cmpxchg" I " %3, %0;"	\
454				     "mov %% " R ", %2;"			\
455				     "setz %1;"					\
456					: "+m"  (*(C *)target),			\
457					  "=a"  (z),				\
458					  "=m"  (*(C *)v)			\
459					: "q"   (set),				\
460					  "a"   (compare)			\
461					: "memory", "cc");			\
462		return z;							\
463	}
464
465CK_PR_CAS_O(ptr, void, void *, char, "q", "rax")
466
467#define CK_PR_CAS_O_S(S, T, I, R)	\
468	CK_PR_CAS_O(S, T, T, T, I, R)
469
470CK_PR_CAS_O_S(char, char, "b", "al")
471CK_PR_CAS_O_S(int, int, "l", "eax")
472CK_PR_CAS_O_S(uint, unsigned int, "l", "eax")
473#ifndef CK_PR_DISABLE_DOUBLE
474CK_PR_CAS_O_S(double, double, "q", "rax")
475#endif
476CK_PR_CAS_O_S(64, uint64_t, "q", "rax")
477CK_PR_CAS_O_S(32, uint32_t, "l", "eax")
478CK_PR_CAS_O_S(16, uint16_t, "w", "ax")
479CK_PR_CAS_O_S(8,  uint8_t,  "b", "al")
480
481#undef CK_PR_CAS_O_S
482#undef CK_PR_CAS_O
483
484/*
485 * Contrary to C-interface, alignment requirements are that of uint64_t[2].
486 */
487CK_CC_INLINE static bool
488ck_pr_cas_64_2(uint64_t target[2], uint64_t compare[2], uint64_t set[2])
489{
490	bool z;
491
492	__asm__ __volatile__("movq 0(%4), %%rax;"
493			     "movq 8(%4), %%rdx;"
494			     CK_PR_LOCK_PREFIX "cmpxchg16b %0; setz %1"
495				: "+m" (*target),
496				  "=q" (z)
497				: "b"  (set[0]),
498				  "c"  (set[1]),
499				  "q"  (compare)
500				: "memory", "cc", "%rax", "%rdx");
501	return z;
502}
503
504CK_CC_INLINE static bool
505ck_pr_cas_ptr_2(void *t, void *c, void *s)
506{
507	return ck_pr_cas_64_2(CK_CPP_CAST(uint64_t *, t),
508			      CK_CPP_CAST(uint64_t *, c),
509			      CK_CPP_CAST(uint64_t *, s));
510}
511
512CK_CC_INLINE static bool
513ck_pr_cas_64_2_value(uint64_t target[2],
514		     uint64_t compare[2],
515		     uint64_t set[2],
516		     uint64_t v[2])
517{
518	bool z;
519
520	__asm__ __volatile__(CK_PR_LOCK_PREFIX "cmpxchg16b %0;"
521			     "setz %3"
522				: "+m" (*target),
523				  "=a" (v[0]),
524				  "=d" (v[1]),
525				  "=q" (z)
526				: "a" (compare[0]),
527				  "d" (compare[1]),
528				  "b" (set[0]),
529				  "c" (set[1])
530				: "memory", "cc");
531	return z;
532}
533
534CK_CC_INLINE static bool
535ck_pr_cas_ptr_2_value(void *t, void *c, void *s, void *v)
536{
537	return ck_pr_cas_64_2_value(CK_CPP_CAST(uint64_t *,t),
538				    CK_CPP_CAST(uint64_t *,c),
539				    CK_CPP_CAST(uint64_t *,s),
540				    CK_CPP_CAST(uint64_t *,v));
541}
542
543#define CK_PR_CAS_V(S, W, T)					\
544CK_CC_INLINE static bool					\
545ck_pr_cas_##S##_##W(T t[W], T c[W], T s[W])			\
546{								\
547	return ck_pr_cas_64_2((uint64_t *)(void *)t,		\
548			      (uint64_t *)(void *)c,		\
549			      (uint64_t *)(void *)s);		\
550}								\
551CK_CC_INLINE static bool					\
552ck_pr_cas_##S##_##W##_value(T *t, T c[W], T s[W], T *v)		\
553{								\
554	return ck_pr_cas_64_2_value((uint64_t *)(void *)t,	\
555				    (uint64_t *)(void *)c,	\
556				    (uint64_t *)(void *)s,	\
557				    (uint64_t *)(void *)v);	\
558}
559
560#ifndef CK_PR_DISABLE_DOUBLE
561CK_PR_CAS_V(double, 2, double)
562#endif
563CK_PR_CAS_V(char, 16, char)
564CK_PR_CAS_V(int, 4, int)
565CK_PR_CAS_V(uint, 4, unsigned int)
566CK_PR_CAS_V(32, 4, uint32_t)
567CK_PR_CAS_V(16, 8, uint16_t)
568CK_PR_CAS_V(8, 16, uint8_t)
569
570#undef CK_PR_CAS_V
571
572/*
573 * Atomic bit test operations.
574 */
575#define CK_PR_BT(K, S, T, P, C, I)					\
576	CK_CC_INLINE static bool					\
577	ck_pr_##K##_##S(T *target, unsigned int b)			\
578	{								\
579		bool c;							\
580		__asm__ __volatile__(CK_PR_LOCK_PREFIX I "; setc %1"	\
581					: "+m" (*(C *)target),		\
582					  "=q" (c)			\
583					: "q"  ((P)b)			\
584					: "memory", "cc");		\
585		return c;						\
586	}
587
588#define CK_PR_BT_S(K, S, T, I) CK_PR_BT(K, S, T, T, T, I)
589
590#define CK_PR_GENERATE(K)					\
591	CK_PR_BT(K, ptr, void, uint64_t, char, #K "q %2, %0")	\
592	CK_PR_BT_S(K, uint, unsigned int, #K "l %2, %0")	\
593	CK_PR_BT_S(K, int, int, #K "l %2, %0")			\
594	CK_PR_BT_S(K, 64, uint64_t, #K "q %2, %0")		\
595	CK_PR_BT_S(K, 32, uint32_t, #K "l %2, %0")		\
596	CK_PR_BT_S(K, 16, uint16_t, #K "w %w2, %0")
597
598CK_PR_GENERATE(btc)
599CK_PR_GENERATE(bts)
600CK_PR_GENERATE(btr)
601
602#undef CK_PR_GENERATE
603#undef CK_PR_BT
604
605#endif /* CK_PR_X86_64_H */
606
607