summaryrefslogtreecommitdiff
path: root/boost/atomic/detail/ops_gcc_x86_dcas.hpp
blob: a6109f926c1da5d96c9f86fc52bcf158385d31d3 (plain)
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
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
/*
 * Distributed under the Boost Software License, Version 1.0.
 * (See accompanying file LICENSE_1_0.txt or copy at
 * http://www.boost.org/LICENSE_1_0.txt)
 *
 * Copyright (c) 2009 Helge Bahmann
 * Copyright (c) 2012 Tim Blechmann
 * Copyright (c) 2014 Andrey Semashev
 */
/*!
 * \file   atomic/detail/ops_gcc_x86_dcas.hpp
 *
 * This header contains implementation of the double-width CAS primitive for x86.
 */

#ifndef BOOST_ATOMIC_DETAIL_OPS_GCC_X86_DCAS_HPP_INCLUDED_
#define BOOST_ATOMIC_DETAIL_OPS_GCC_X86_DCAS_HPP_INCLUDED_

#include <boost/cstdint.hpp>
#include <boost/memory_order.hpp>
#include <boost/atomic/detail/config.hpp>
#include <boost/atomic/detail/storage_type.hpp>
#include <boost/atomic/capabilities.hpp>

#ifdef BOOST_HAS_PRAGMA_ONCE
#pragma once
#endif

namespace boost {
namespace atomics {
namespace detail {

#if defined(BOOST_ATOMIC_DETAIL_X86_HAS_CMPXCHG8B)

template< bool Signed >
struct gcc_dcas_x86
{
    typedef typename make_storage_type< 8u, Signed >::type storage_type;

    static BOOST_FORCEINLINE void store(storage_type volatile& storage, storage_type v, memory_order) BOOST_NOEXCEPT
    {
        if ((((uint32_t)&storage) & 0x00000007) == 0)
        {
#if defined(__SSE2__)
            __asm__ __volatile__
            (
#if defined(__AVX__)
                "vmovq %1, %%xmm4\n\t"
                "vmovq %%xmm4, %0\n\t"
#else
                "movq %1, %%xmm4\n\t"
                "movq %%xmm4, %0\n\t"
#endif
                : "=m" (storage)
                : "m" (v)
                : "memory", "xmm4"
            );
#else
            __asm__ __volatile__
            (
                "fildll %1\n\t"
                "fistpll %0\n\t"
                : "=m" (storage)
                : "m" (v)
                : "memory"
            );
#endif
        }
        else
        {
#if defined(__PIC__)
            uint32_t scratch;
            __asm__ __volatile__
            (
                "movl %%ebx, %[scratch]\n\t"
                "movl %[value_lo], %%ebx\n\t"
                "movl 0(%[dest]), %%eax\n\t"
                "movl 4(%[dest]), %%edx\n\t"
                ".align 16\n\t"
                "1: lock; cmpxchg8b 0(%[dest])\n\t"
                "jne 1b\n\t"
                "movl %[scratch], %%ebx"
#if !defined(BOOST_ATOMIC_DETAIL_NO_ASM_CONSTRAINT_ALTERNATIVES)
                : [scratch] "=m,m" (scratch)
                : [value_lo] "a,a" ((uint32_t)v), "c,c" ((uint32_t)(v >> 32)), [dest] "D,S" (&storage)
#else
                : [scratch] "=m" (scratch)
                : [value_lo] "a" ((uint32_t)v), "c" ((uint32_t)(v >> 32)), [dest] "D" (&storage)
#endif
                : BOOST_ATOMIC_DETAIL_ASM_CLOBBER_CC_COMMA "edx", "memory"
            );
#else
            __asm__ __volatile__
            (
                "movl 0(%[dest]), %%eax\n\t"
                "movl 4(%[dest]), %%edx\n\t"
                ".align 16\n\t"
                "1: lock; cmpxchg8b 0(%[dest])\n\t"
                "jne 1b\n\t"
                :
#if !defined(BOOST_ATOMIC_DETAIL_NO_ASM_CONSTRAINT_ALTERNATIVES)
                : [value_lo] "b,b" ((uint32_t)v), "c,c" ((uint32_t)(v >> 32)), [dest] "D,S" (&storage)
#else
                : [value_lo] "b" ((uint32_t)v), "c" ((uint32_t)(v >> 32)), [dest] "D" (&storage)
#endif
                : BOOST_ATOMIC_DETAIL_ASM_CLOBBER_CC_COMMA "eax", "edx", "memory"
            );
#endif
        }
    }

    static BOOST_FORCEINLINE storage_type load(storage_type const volatile& storage, memory_order) BOOST_NOEXCEPT
    {
        storage_type value;

        if ((((uint32_t)&storage) & 0x00000007) == 0)
        {
#if defined(__SSE2__)
            __asm__ __volatile__
            (
#if defined(__AVX__)
                "vmovq %1, %%xmm4\n\t"
                "vmovq %%xmm4, %0\n\t"
#else
                "movq %1, %%xmm4\n\t"
                "movq %%xmm4, %0\n\t"
#endif
                : "=m" (value)
                : "m" (storage)
                : "memory", "xmm4"
            );
#else
            __asm__ __volatile__
            (
                "fildll %1\n\t"
                "fistpll %0\n\t"
                : "=m" (value)
                : "m" (storage)
                : "memory"
            );
#endif
        }
        else
        {
#if defined(__clang__)
            // Clang cannot allocate eax:edx register pairs but it has sync intrinsics
            value = __sync_val_compare_and_swap(&storage, (storage_type)0, (storage_type)0);
#else
            // We don't care for comparison result here; the previous value will be stored into value anyway.
            // Also we don't care for ebx and ecx values, they just have to be equal to eax and edx before cmpxchg8b.
            __asm__ __volatile__
            (
                "movl %%ebx, %%eax\n\t"
                "movl %%ecx, %%edx\n\t"
                "lock; cmpxchg8b %[storage]"
                : "=&A" (value)
                : [storage] "m" (storage)
                : BOOST_ATOMIC_DETAIL_ASM_CLOBBER_CC_COMMA "memory"
            );
#endif
        }

        return value;
    }

    static BOOST_FORCEINLINE bool compare_exchange_strong(
        storage_type volatile& storage, storage_type& expected, storage_type desired, memory_order, memory_order) BOOST_NOEXCEPT
    {
#if defined(__clang__)
        // Clang cannot allocate eax:edx register pairs but it has sync intrinsics
        storage_type old_expected = expected;
        expected = __sync_val_compare_and_swap(&storage, old_expected, desired);
        return expected == old_expected;
#elif defined(__PIC__)
        // Make sure ebx is saved and restored properly in case
        // of position independent code. To make this work
        // setup register constraints such that ebx can not be
        // used by accident e.g. as base address for the variable
        // to be modified. Accessing "scratch" should always be okay,
        // as it can only be placed on the stack (and therefore
        // accessed through ebp or esp only).
        //
        // In theory, could push/pop ebx onto/off the stack, but movs
        // to a prepared stack slot turn out to be faster.

        uint32_t scratch;
        bool success;
        __asm__ __volatile__
        (
            "movl %%ebx, %[scratch]\n\t"
            "movl %[desired_lo], %%ebx\n\t"
            "lock; cmpxchg8b %[dest]\n\t"
            "movl %[scratch], %%ebx\n\t"
            "sete %[success]"
#if !defined(BOOST_ATOMIC_DETAIL_NO_ASM_CONSTRAINT_ALTERNATIVES)
            : "+A,A,A,A,A,A" (expected), [dest] "+m,m,m,m,m,m" (storage), [scratch] "=m,m,m,m,m,m" (scratch), [success] "=q,m,q,m,q,m" (success)
            : [desired_lo] "S,S,D,D,m,m" ((uint32_t)desired), "c,c,c,c,c,c" ((uint32_t)(desired >> 32))
#else
            : "+A" (expected), [dest] "+m" (storage), [scratch] "=m" (scratch), [success] "=q" (success)
            : [desired_lo] "S" ((uint32_t)desired), "c" ((uint32_t)(desired >> 32))
#endif
            : BOOST_ATOMIC_DETAIL_ASM_CLOBBER_CC_COMMA "memory"
        );
        return success;
#else
        bool success;
        __asm__ __volatile__
        (
            "lock; cmpxchg8b %[dest]\n\t"
            "sete %[success]"
#if !defined(BOOST_ATOMIC_DETAIL_NO_ASM_CONSTRAINT_ALTERNATIVES)
            : "+A,A" (expected), [dest] "+m,m" (storage), [success] "=q,m" (success)
            : "b,b" ((uint32_t)desired), "c,c" ((uint32_t)(desired >> 32))
#else
            : "+A" (expected), [dest] "+m" (storage), [success] "=q" (success)
            : "b" ((uint32_t)desired), "c" ((uint32_t)(desired >> 32))
#endif
            : BOOST_ATOMIC_DETAIL_ASM_CLOBBER_CC_COMMA "memory"
        );
        return success;
#endif
    }

    static BOOST_FORCEINLINE bool compare_exchange_weak(
        storage_type volatile& storage, storage_type& expected, storage_type desired, memory_order success_order, memory_order failure_order) BOOST_NOEXCEPT
    {
        return compare_exchange_strong(storage, expected, desired, success_order, failure_order);
    }

    static BOOST_FORCEINLINE bool is_lock_free(storage_type const volatile&) BOOST_NOEXCEPT
    {
        return true;
    }
};

#endif // defined(BOOST_ATOMIC_DETAIL_X86_HAS_CMPXCHG8B)

#if defined(BOOST_ATOMIC_DETAIL_X86_HAS_CMPXCHG16B)

template< bool Signed >
struct gcc_dcas_x86_64
{
    typedef typename make_storage_type< 16u, Signed >::type storage_type;

    static BOOST_FORCEINLINE void store(storage_type volatile& storage, storage_type v, memory_order) BOOST_NOEXCEPT
    {
        uint64_t const* p_value = (uint64_t const*)&v;
        __asm__ __volatile__
        (
            "movq 0(%[dest]), %%rax\n\t"
            "movq 8(%[dest]), %%rdx\n\t"
            ".align 16\n\t"
            "1: lock; cmpxchg16b 0(%[dest])\n\t"
            "jne 1b"
            :
            : "b" (p_value[0]), "c" (p_value[1]), [dest] "r" (&storage)
            : BOOST_ATOMIC_DETAIL_ASM_CLOBBER_CC_COMMA "rax", "rdx", "memory"
        );
    }

    static BOOST_FORCEINLINE storage_type load(storage_type const volatile& storage, memory_order) BOOST_NOEXCEPT
    {
#if defined(__clang__)
        // Clang cannot allocate rax:rdx register pairs but it has sync intrinsics
        storage_type value = storage_type();
        return __sync_val_compare_and_swap(&storage, value, value);
#else
        storage_type value;

        // We don't care for comparison result here; the previous value will be stored into value anyway.
        // Also we don't care for rbx and rcx values, they just have to be equal to rax and rdx before cmpxchg16b.
        __asm__ __volatile__
        (
            "movq %%rbx, %%rax\n\t"
            "movq %%rcx, %%rdx\n\t"
            "lock; cmpxchg16b %[storage]"
            : "=&A" (value)
            : [storage] "m" (storage)
            : BOOST_ATOMIC_DETAIL_ASM_CLOBBER_CC_COMMA "memory"
        );

        return value;
#endif
    }

    static BOOST_FORCEINLINE bool compare_exchange_strong(
        storage_type volatile& storage, storage_type& expected, storage_type desired, memory_order, memory_order) BOOST_NOEXCEPT
    {
#if defined(__clang__)
        // Clang cannot allocate rax:rdx register pairs but it has sync intrinsics
        storage_type old_expected = expected;
        expected = __sync_val_compare_and_swap(&storage, old_expected, desired);
        return expected == old_expected;
#else
        uint64_t const* p_desired = (uint64_t const*)&desired;
        bool success;
        __asm__ __volatile__
        (
            "lock; cmpxchg16b %[dest]\n\t"
            "sete %[success]"
#if !defined(BOOST_ATOMIC_DETAIL_NO_ASM_CONSTRAINT_ALTERNATIVES)
            : "+A,A" (expected), [dest] "+m,m" (storage), [success] "=q,m" (success)
            : "b,b" (p_desired[0]), "c,c" (p_desired[1])
#else
            : "+A" (expected), [dest] "+m" (storage), [success] "=q" (success)
            : "b" (p_desired[0]), "c" (p_desired[1])
#endif
            : BOOST_ATOMIC_DETAIL_ASM_CLOBBER_CC_COMMA "memory"
        );
        return success;
#endif
    }

    static BOOST_FORCEINLINE bool compare_exchange_weak(
        storage_type volatile& storage, storage_type& expected, storage_type desired, memory_order success_order, memory_order failure_order) BOOST_NOEXCEPT
    {
        return compare_exchange_strong(storage, expected, desired, success_order, failure_order);
    }

    static BOOST_FORCEINLINE bool is_lock_free(storage_type const volatile&) BOOST_NOEXCEPT
    {
        return true;
    }
};

#endif // defined(BOOST_ATOMIC_DETAIL_X86_HAS_CMPXCHG16B)

} // namespace detail
} // namespace atomics
} // namespace boost

#endif // BOOST_ATOMIC_DETAIL_OPS_GCC_X86_DCAS_HPP_INCLUDED_