/usr/include/atomic_ops/sysdeps/sunc/x86_64.h is in libatomic-ops-dev 7.2~alpha5+cvs20101124-1ubuntu1.
This file is owned by root:root, with mode 0o644.
The actual contents of the file can be viewed below.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 | /*
* Copyright (c) 1991-1994 by Xerox Corporation. All rights reserved.
* Copyright (c) 1996-1999 by Silicon Graphics. All rights reserved.
* Copyright (c) 1999-2003 by Hewlett-Packard Company. All rights reserved.
*
*
* THIS MATERIAL IS PROVIDED AS IS, WITH ABSOLUTELY NO WARRANTY EXPRESSED
* OR IMPLIED. ANY USE IS AT YOUR OWN RISK.
*
* Permission is hereby granted to use or copy this program
* for any purpose, provided the above notices are retained on all copies.
* Permission to modify the code and to distribute modified code is granted,
* provided the above notices are retained, and a notice that the code was
* modified is included with the above copyright notice.
*
* Some of the machine specific code was borrowed from our GC distribution.
*/
#include "../all_aligned_atomic_load_store.h"
/* Real X86 implementations, appear */
/* to enforce ordering between memory operations, EXCEPT that a later */
/* read can pass earlier writes, presumably due to the visible */
/* presence of store buffers. */
/* We ignore the fact that the official specs */
/* seem to be much weaker (and arguably too weak to be usable). */
#include "../ordered_except_wr.h"
#include "../test_and_set_t_is_char.h"
#include "../standard_ao_double_t.h"
AO_INLINE void
AO_nop_full(void)
{
/* Note: "mfence" (SSE2) is supported on all x86_64/amd64 chips. */
__asm__ __volatile__("mfence" : : : "memory");
}
#define AO_HAVE_nop_full
/* As far as we can tell, the lfence and sfence instructions are not */
/* currently needed or useful for cached memory accesses. */
AO_INLINE AO_t
AO_fetch_and_add_full (volatile AO_t *p, AO_t incr)
{
AO_t result;
__asm__ __volatile__ ("lock; xaddq %0, %1" :
"=r" (result), "=m" (*p) : "0" (incr) /* , "m" (*p) */
: "memory");
return result;
}
#define AO_HAVE_fetch_and_add_full
AO_INLINE unsigned char
AO_char_fetch_and_add_full (volatile unsigned char *p, unsigned char incr)
{
unsigned char result;
__asm__ __volatile__ ("lock; xaddb %0, %1" :
"=q" (result), "=m" (*p) : "0" (incr) /* , "m" (*p) */
: "memory");
return result;
}
#define AO_HAVE_char_fetch_and_add_full
AO_INLINE unsigned short
AO_short_fetch_and_add_full (volatile unsigned short *p, unsigned short incr)
{
unsigned short result;
__asm__ __volatile__ ("lock; xaddw %0, %1" :
"=r" (result), "=m" (*p) : "0" (incr) /* , "m" (*p) */
: "memory");
return result;
}
#define AO_HAVE_short_fetch_and_add_full
AO_INLINE unsigned int
AO_int_fetch_and_add_full (volatile unsigned int *p, unsigned int incr)
{
unsigned int result;
__asm__ __volatile__ ("lock; xaddl %0, %1" :
"=r" (result), "=m" (*p) : "0" (incr) /* , "m" (*p) */
: "memory");
return result;
}
#define AO_HAVE_int_fetch_and_add_full
AO_INLINE void
AO_or_full (volatile AO_t *p, AO_t incr)
{
__asm__ __volatile__ ("lock; orq %1, %0" :
"=m" (*p) : "r" (incr) /* , "m" (*p) */
: "memory");
}
#define AO_HAVE_or_full
AO_INLINE AO_TS_VAL_t
AO_test_and_set_full(volatile AO_TS_t *addr)
{
AO_TS_t oldval;
/* Note: the "xchg" instruction does not need a "lock" prefix */
__asm__ __volatile__("xchg %0, %1"
: "=q"(oldval), "=m"(*addr)
: "0"(0xff) /* , "m"(*addr) */
: "memory");
return (AO_TS_VAL_t)oldval;
}
#define AO_HAVE_test_and_set_full
/* Returns nonzero if the comparison succeeded. */
AO_INLINE int
AO_compare_and_swap_full(volatile AO_t *addr,
AO_t old, AO_t new_val)
{
char result;
__asm__ __volatile__("lock; cmpxchgq %2, %0; setz %1"
: "=m"(*addr), "=a"(result)
: "r" (new_val), "a"(old) : "memory");
return (int) result;
}
#define AO_HAVE_compare_and_swap_full
#ifdef AO_CMPXCHG16B_AVAILABLE
/* NEC LE-IT: older AMD Opterons are missing this instruction.
* On these machines SIGILL will be thrown.
* Define AO_WEAK_DOUBLE_CAS_EMULATION to have an emulated
* (lock based) version available */
/* HB: Changed this to not define either by default. There are
* enough machines and tool chains around on which cmpxchg16b
* doesn't work. And the emulation is unsafe by our usual rules.
* Hoewever both are clearly useful in certain cases.
*/
AO_INLINE int
AO_compare_double_and_swap_double_full(volatile AO_double_t *addr,
AO_t old_val1, AO_t old_val2,
AO_t new_val1, AO_t new_val2)
{
char result;
__asm__ __volatile__("lock; cmpxchg16b %0; setz %1"
: "=m"(*addr), "=a"(result)
: "m"(*addr), "d" (old_val2), "a" (old_val1),
"c" (new_val2), "b" (new_val1) : "memory");
return (int) result;
}
#define AO_HAVE_compare_double_and_swap_double_full
#else
/* this one provides spinlock based emulation of CAS implemented in */
/* atomic_ops.c. We probably do not want to do this here, since it is */
/* not atomic with respect to other kinds of updates of *addr. On the */
/* other hand, this may be a useful facility on occasion. */
#ifdef AO_WEAK_DOUBLE_CAS_EMULATION
int AO_compare_double_and_swap_double_emulation(volatile AO_double_t *addr,
AO_t old_val1, AO_t old_val2,
AO_t new_val1, AO_t new_val2);
AO_INLINE int
AO_compare_double_and_swap_double_full(volatile AO_double_t *addr,
AO_t old_val1, AO_t old_val2,
AO_t new_val1, AO_t new_val2)
{
return AO_compare_double_and_swap_double_emulation(addr,
old_val1, old_val2,
new_val1, new_val2);
}
#define AO_HAVE_compare_double_and_swap_double_full
#endif /* AO_WEAK_DOUBLE_CAS_EMULATION */
#endif /* AO_CMPXCHG16B_AVAILABLE */
|