summaryrefslogtreecommitdiff
path: root/boost/atomic/detail/ops_gcc_arm.hpp
blob: d2c2f39a2c80616678a670b433855a92de27af09 (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
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
/*
 * 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) 2013 Tim Blechmann
 * Copyright (c) 2014 Andrey Semashev
 */
/*!
 * \file   atomic/detail/ops_gcc_arm.hpp
 *
 * This header contains implementation of the \c operations template.
 */

#ifndef BOOST_ATOMIC_DETAIL_OPS_GCC_ARM_HPP_INCLUDED_
#define BOOST_ATOMIC_DETAIL_OPS_GCC_ARM_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/detail/operations_fwd.hpp>
#include <boost/atomic/detail/ops_extending_cas_based.hpp>
#include <boost/atomic/capabilities.hpp>

#ifdef BOOST_HAS_PRAGMA_ONCE
#pragma once
#endif

namespace boost {
namespace atomics {
namespace detail {

// From the ARM Architecture Reference Manual for architecture v6:
//
// LDREX{<cond>} <Rd>, [<Rn>]
// <Rd> Specifies the destination register for the memory word addressed by <Rd>
// <Rn> Specifies the register containing the address.
//
// STREX{<cond>} <Rd>, <Rm>, [<Rn>]
// <Rd> Specifies the destination register for the returned status value.
//      0  if the operation updates memory
//      1  if the operation fails to update memory
// <Rm> Specifies the register containing the word to be stored to memory.
// <Rn> Specifies the register containing the address.
// Rd must not be the same register as Rm or Rn.
//
// ARM v7 is like ARM v6 plus:
// There are half-word and byte versions of the LDREX and STREX instructions,
// LDREXH, LDREXB, STREXH and STREXB.
// There are also double-word versions, LDREXD and STREXD.
// (Actually it looks like these are available from version 6k onwards.)
// FIXME these are not yet used; should be mostly a matter of copy-and-paste.
// I think you can supply an immediate offset to the address.
//
// A memory barrier is effected using a "co-processor 15" instruction,
// though a separate assembler mnemonic is available for it in v7.
//
// "Thumb 1" is a subset of the ARM instruction set that uses a 16-bit encoding.  It
// doesn't include all instructions and in particular it doesn't include the co-processor
// instruction used for the memory barrier or the load-locked/store-conditional
// instructions.  So, if we're compiling in "Thumb 1" mode, we need to wrap all of our
// asm blocks with code to temporarily change to ARM mode.
//
// You can only change between ARM and Thumb modes when branching using the bx instruction.
// bx takes an address specified in a register.  The least significant bit of the address
// indicates the mode, so 1 is added to indicate that the destination code is Thumb.
// A temporary register is needed for the address and is passed as an argument to these
// macros.  It must be one of the "low" registers accessible to Thumb code, specified
// using the "l" attribute in the asm statement.
//
// Architecture v7 introduces "Thumb 2", which does include (almost?) all of the ARM
// instruction set.  (Actually, there was an extension of v6 called v6T2 which supported
// "Thumb 2" mode, but its architecture manual is no longer available, referring to v7.)
// So in v7 we don't need to change to ARM mode; we can write "universal
// assembler" which will assemble to Thumb 2 or ARM code as appropriate.  The only thing
// we need to do to make this "universal" assembler mode work is to insert "IT" instructions
// to annotate the conditional instructions.  These are ignored in other modes (e.g. v6),
// so they can always be present.

// A note about memory_order_consume. Technically, this architecture allows to avoid
// unnecessary memory barrier after consume load since it supports data dependency ordering.
// However, some compiler optimizations may break a seemingly valid code relying on data
// dependency tracking by injecting bogus branches to aid out of order execution.
// This may happen not only in Boost.Atomic code but also in user's code, which we have no
// control of. See this thread: http://lists.boost.org/Archives/boost/2014/06/213890.php.
// For this reason we promote memory_order_consume to memory_order_acquire.

#if defined(__thumb__) && !defined(__thumb2__)
#define BOOST_ATOMIC_DETAIL_ARM_ASM_START(TMPREG) "adr " #TMPREG ", 8f\n" "bx " #TMPREG "\n" ".arm\n" ".align 4\n" "8:\n"
#define BOOST_ATOMIC_DETAIL_ARM_ASM_END(TMPREG)   "adr " #TMPREG ", 9f + 1\n" "bx " #TMPREG "\n" ".thumb\n" ".align 2\n" "9:\n"
#define BOOST_ATOMIC_DETAIL_ARM_ASM_TMPREG_CONSTRAINT(var) "=&l" (var)
#else
// The tmpreg may be wasted in this case, which is non-optimal.
#define BOOST_ATOMIC_DETAIL_ARM_ASM_START(TMPREG)
#define BOOST_ATOMIC_DETAIL_ARM_ASM_END(TMPREG)
#define BOOST_ATOMIC_DETAIL_ARM_ASM_TMPREG_CONSTRAINT(var) "=&r" (var)
#endif

struct gcc_arm_operations_base
{
    static BOOST_FORCEINLINE void fence_before(memory_order order) BOOST_NOEXCEPT
    {
        if ((order & memory_order_release) != 0)
            hardware_full_fence();
    }

    static BOOST_FORCEINLINE void fence_after(memory_order order) BOOST_NOEXCEPT
    {
        if ((order & (memory_order_consume | memory_order_acquire)) != 0)
            hardware_full_fence();
    }

    static BOOST_FORCEINLINE void fence_after_store(memory_order order) BOOST_NOEXCEPT
    {
        if (order == memory_order_seq_cst)
            hardware_full_fence();
    }

    static BOOST_FORCEINLINE void hardware_full_fence() BOOST_NOEXCEPT
    {
#if defined(BOOST_ATOMIC_DETAIL_ARM_HAS_DMB)
        // Older binutils (supposedly, older than 2.21.1) didn't support symbolic or numeric arguments of the "dmb" instruction such as "ish" or "#11".
        // As a workaround we have to inject encoded bytes of the instruction. There are two encodings for the instruction: ARM and Thumb. See ARM Architecture Reference Manual, A8.8.43.
        // Since we cannot detect binutils version at compile time, we'll have to always use this hack.
        __asm__ __volatile__
        (
#if defined(__thumb2__)
            ".short 0xF3BF, 0x8F5B\n" // dmb ish
#else
            ".word 0xF57FF05B\n" // dmb ish
#endif
            :
            :
            : "memory"
        );
#else
        int tmp;
        __asm__ __volatile__
        (
            BOOST_ATOMIC_DETAIL_ARM_ASM_START(%0)
            "mcr\tp15, 0, r0, c7, c10, 5\n"
            BOOST_ATOMIC_DETAIL_ARM_ASM_END(%0)
            : "=&l" (tmp)
            :
            : "memory"
        );
#endif
    }
};


template< bool Signed >
struct operations< 4u, Signed > :
    public gcc_arm_operations_base
{
    typedef typename make_storage_type< 4u, Signed >::type storage_type;
    typedef typename make_storage_type< 4u, Signed >::aligned aligned_storage_type;

    static BOOST_FORCEINLINE void store(storage_type volatile& storage, storage_type v, memory_order order) BOOST_NOEXCEPT
    {
        fence_before(order);
        storage = v;
        fence_after_store(order);
    }

    static BOOST_FORCEINLINE storage_type load(storage_type const volatile& storage, memory_order order) BOOST_NOEXCEPT
    {
        storage_type v = storage;
        fence_after(order);
        return v;
    }

    static BOOST_FORCEINLINE storage_type exchange(storage_type volatile& storage, storage_type v, memory_order order) BOOST_NOEXCEPT
    {
        storage_type original;
        fence_before(order);
        uint32_t tmp;
        __asm__ __volatile__
        (
            BOOST_ATOMIC_DETAIL_ARM_ASM_START(%[tmp])
            "1:\n"
            "ldrex %[original], %[storage]\n"          // load the original value
            "strex %[tmp], %[value], %[storage]\n"     // store the replacement, tmp = store failed
            "teq   %[tmp], #0\n"                       // check if store succeeded
            "bne   1b\n"
            BOOST_ATOMIC_DETAIL_ARM_ASM_END(%[tmp])
            : [tmp] "=&l" (tmp), [original] "=&r" (original), [storage] "+Q" (storage)
            : [value] "r" (v)
            : BOOST_ATOMIC_DETAIL_ASM_CLOBBER_CC
        );
        fence_after(order);
        return original;
    }

    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
    {
        fence_before(success_order);
        uint32_t success;
        uint32_t tmp;
        storage_type original;
        __asm__ __volatile__
        (
            BOOST_ATOMIC_DETAIL_ARM_ASM_START(%[tmp])
            "mov     %[success], #0\n"                      // success = 0
            "ldrex   %[original], %[storage]\n"             // original = *(&storage)
            "cmp     %[original], %[expected]\n"            // flags = original==expected
            "itt     eq\n"                                  // [hint that the following 2 instructions are conditional on flags.equal]
            "strexeq %[success], %[desired], %[storage]\n"  // if (flags.equal) *(&storage) = desired, success = store failed
            "eoreq   %[success], %[success], #1\n"          // if (flags.equal) success ^= 1 (i.e. make it 1 if store succeeded)
            BOOST_ATOMIC_DETAIL_ARM_ASM_END(%[tmp])
            : [original] "=&r" (original),  // %0
              [success] "=&r" (success),    // %1
              [tmp] "=&l" (tmp),            // %2
              [storage] "+Q" (storage)      // %3
            : [expected] "r" (expected),    // %4
              [desired] "r" (desired)       // %5
            : BOOST_ATOMIC_DETAIL_ASM_CLOBBER_CC
        );
        if (success)
            fence_after(success_order);
        else
            fence_after(failure_order);
        expected = original;
        return !!success;
    }

    static BOOST_FORCEINLINE bool compare_exchange_strong(
        storage_type volatile& storage, storage_type& expected, storage_type desired, memory_order success_order, memory_order failure_order) BOOST_NOEXCEPT
    {
        fence_before(success_order);
        uint32_t success;
        uint32_t tmp;
        storage_type original;
        __asm__ __volatile__
        (
            BOOST_ATOMIC_DETAIL_ARM_ASM_START(%[tmp])
            "mov     %[success], #0\n"                      // success = 0
            "1:\n"
            "ldrex   %[original], %[storage]\n"             // original = *(&storage)
            "cmp     %[original], %[expected]\n"            // flags = original==expected
            "bne     2f\n"                                  // if (!flags.equal) goto end
            "strex   %[success], %[desired], %[storage]\n"  // *(&storage) = desired, success = store failed
            "eors    %[success], %[success], #1\n"          // success ^= 1 (i.e. make it 1 if store succeeded); flags.equal = success == 0
            "beq     1b\n"                                  // if (flags.equal) goto retry
            "2:\n"
            BOOST_ATOMIC_DETAIL_ARM_ASM_END(%[tmp])
            : [original] "=&r" (original),  // %0
              [success] "=&r" (success),    // %1
              [tmp] "=&l" (tmp),            // %2
              [storage] "+Q" (storage)      // %3
            : [expected] "r" (expected),    // %4
              [desired] "r" (desired)       // %5
            : BOOST_ATOMIC_DETAIL_ASM_CLOBBER_CC
        );
        if (success)
            fence_after(success_order);
        else
            fence_after(failure_order);
        expected = original;
        return !!success;
    }

    static BOOST_FORCEINLINE storage_type fetch_add(storage_type volatile& storage, storage_type v, memory_order order) BOOST_NOEXCEPT
    {
        fence_before(order);
        uint32_t tmp;
        storage_type original, result;
        __asm__ __volatile__
        (
            BOOST_ATOMIC_DETAIL_ARM_ASM_START(%[tmp])
            "1:\n"
            "ldrex   %[original], %[storage]\n"           // original = *(&storage)
            "add     %[result], %[original], %[value]\n"  // result = original + value
            "strex   %[tmp], %[result], %[storage]\n"     // *(&storage) = result, tmp = store failed
            "teq     %[tmp], #0\n"                        // flags = tmp==0
            "bne     1b\n"                                // if (!flags.equal) goto retry
            BOOST_ATOMIC_DETAIL_ARM_ASM_END(%[tmp])
            : [original] "=&r" (original),  // %0
              [result] "=&r" (result),      // %1
              [tmp] "=&l" (tmp),            // %2
              [storage] "+Q" (storage)      // %3
            : [value] "r" (v)               // %4
            : BOOST_ATOMIC_DETAIL_ASM_CLOBBER_CC
        );
        fence_after(order);
        return original;
    }

    static BOOST_FORCEINLINE storage_type fetch_sub(storage_type volatile& storage, storage_type v, memory_order order) BOOST_NOEXCEPT
    {
        fence_before(order);
        uint32_t tmp;
        storage_type original, result;
        __asm__ __volatile__
        (
            BOOST_ATOMIC_DETAIL_ARM_ASM_START(%[tmp])
            "1:\n"
            "ldrex   %[original], %[storage]\n"           // original = *(&storage)
            "sub     %[result], %[original], %[value]\n"  // result = original - value
            "strex   %[tmp], %[result], %[storage]\n"     // *(&storage) = result, tmp = store failed
            "teq     %[tmp], #0\n"                        // flags = tmp==0
            "bne     1b\n"                                // if (!flags.equal) goto retry
            BOOST_ATOMIC_DETAIL_ARM_ASM_END(%[tmp])
            : [original] "=&r" (original),  // %0
              [result] "=&r" (result),      // %1
              [tmp] "=&l" (tmp),            // %2
              [storage] "+Q" (storage)      // %3
            : [value] "r" (v)               // %4
            : BOOST_ATOMIC_DETAIL_ASM_CLOBBER_CC
        );
        fence_after(order);
        return original;
    }

    static BOOST_FORCEINLINE storage_type fetch_and(storage_type volatile& storage, storage_type v, memory_order order) BOOST_NOEXCEPT
    {
        fence_before(order);
        uint32_t tmp;
        storage_type original, result;
        __asm__ __volatile__
        (
            BOOST_ATOMIC_DETAIL_ARM_ASM_START(%[tmp])
            "1:\n"
            "ldrex   %[original], %[storage]\n"           // original = *(&storage)
            "and     %[result], %[original], %[value]\n"  // result = original & value
            "strex   %[tmp], %[result], %[storage]\n"     // *(&storage) = result, tmp = store failed
            "teq     %[tmp], #0\n"                        // flags = tmp==0
            "bne     1b\n"                                // if (!flags.equal) goto retry
            BOOST_ATOMIC_DETAIL_ARM_ASM_END(%[tmp])
            : [original] "=&r" (original),  // %0
              [result] "=&r" (result),      // %1
              [tmp] "=&l" (tmp),            // %2
              [storage] "+Q" (storage)      // %3
            : [value] "r" (v)               // %4
            : BOOST_ATOMIC_DETAIL_ASM_CLOBBER_CC
        );
        fence_after(order);
        return original;
    }

    static BOOST_FORCEINLINE storage_type fetch_or(storage_type volatile& storage, storage_type v, memory_order order) BOOST_NOEXCEPT
    {
        fence_before(order);
        uint32_t tmp;
        storage_type original, result;
        __asm__ __volatile__
        (
            BOOST_ATOMIC_DETAIL_ARM_ASM_START(%[tmp])
            "1:\n"
            "ldrex   %[original], %[storage]\n"           // original = *(&storage)
            "orr     %[result], %[original], %[value]\n"  // result = original | value
            "strex   %[tmp], %[result], %[storage]\n"     // *(&storage) = result, tmp = store failed
            "teq     %[tmp], #0\n"                        // flags = tmp==0
            "bne     1b\n"                                // if (!flags.equal) goto retry
            BOOST_ATOMIC_DETAIL_ARM_ASM_END(%[tmp])
            : [original] "=&r" (original),  // %0
              [result] "=&r" (result),      // %1
              [tmp] "=&l" (tmp),            // %2
              [storage] "+Q" (storage)      // %3
            : [value] "r" (v)               // %4
            : BOOST_ATOMIC_DETAIL_ASM_CLOBBER_CC
        );
        fence_after(order);
        return original;
    }

    static BOOST_FORCEINLINE storage_type fetch_xor(storage_type volatile& storage, storage_type v, memory_order order) BOOST_NOEXCEPT
    {
        fence_before(order);
        uint32_t tmp;
        storage_type original, result;
        __asm__ __volatile__
        (
            BOOST_ATOMIC_DETAIL_ARM_ASM_START(%[tmp])
            "1:\n"
            "ldrex   %[original], %[storage]\n"           // original = *(&storage)
            "eor     %[result], %[original], %[value]\n"  // result = original ^ value
            "strex   %[tmp], %[result], %[storage]\n"     // *(&storage) = result, tmp = store failed
            "teq     %[tmp], #0\n"                        // flags = tmp==0
            "bne     1b\n"                                // if (!flags.equal) goto retry
            BOOST_ATOMIC_DETAIL_ARM_ASM_END(%[tmp])
            : [original] "=&r" (original),  // %0
              [result] "=&r" (result),      // %1
              [tmp] "=&l" (tmp),            // %2
              [storage] "+Q" (storage)      // %3
            : [value] "r" (v)               // %4
            : BOOST_ATOMIC_DETAIL_ASM_CLOBBER_CC
        );
        fence_after(order);
        return original;
    }

    static BOOST_FORCEINLINE bool test_and_set(storage_type volatile& storage, memory_order order) BOOST_NOEXCEPT
    {
        return !!exchange(storage, (storage_type)1, order);
    }

    static BOOST_FORCEINLINE void clear(storage_type volatile& storage, memory_order order) BOOST_NOEXCEPT
    {
        store(storage, 0, order);
    }

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


template< >
struct operations< 1u, false > :
    public operations< 4u, false >
{
    typedef operations< 4u, false > base_type;
    typedef base_type::storage_type storage_type;

    static BOOST_FORCEINLINE storage_type fetch_add(storage_type volatile& storage, storage_type v, memory_order order) BOOST_NOEXCEPT
    {
        fence_before(order);
        uint32_t tmp;
        storage_type original, result;
        __asm__ __volatile__
        (
            BOOST_ATOMIC_DETAIL_ARM_ASM_START(%[tmp])
            "1:\n"
            "ldrex   %[original], %[storage]\n"           // original = *(&storage)
            "add     %[result], %[original], %[value]\n"  // result = original + value
            "uxtb    %[result], %[result]\n"              // zero extend result from 8 to 32 bits
            "strex   %[tmp], %[result], %[storage]\n"     // *(&storage) = result, tmp = store failed
            "teq     %[tmp], #0\n"                        // flags = tmp==0
            "bne     1b\n"                                // if (!flags.equal) goto retry
            BOOST_ATOMIC_DETAIL_ARM_ASM_END(%[tmp])
            : [original] "=&r" (original),  // %0
              [result] "=&r" (result),      // %1
              [tmp] "=&l" (tmp),            // %2
              [storage] "+Q" (storage)      // %3
            : [value] "r" (v)               // %4
            : BOOST_ATOMIC_DETAIL_ASM_CLOBBER_CC
        );
        fence_after(order);
        return original;
    }

    static BOOST_FORCEINLINE storage_type fetch_sub(storage_type volatile& storage, storage_type v, memory_order order) BOOST_NOEXCEPT
    {
        fence_before(order);
        uint32_t tmp;
        storage_type original, result;
        __asm__ __volatile__
        (
            BOOST_ATOMIC_DETAIL_ARM_ASM_START(%[tmp])
            "1:\n"
            "ldrex   %[original], %[storage]\n"           // original = *(&storage)
            "sub     %[result], %[original], %[value]\n"  // result = original - value
            "uxtb    %[result], %[result]\n"              // zero extend result from 8 to 32 bits
            "strex   %[tmp], %[result], %[storage]\n"     // *(&storage) = result, tmp = store failed
            "teq     %[tmp], #0\n"                        // flags = tmp==0
            "bne     1b\n"                                // if (!flags.equal) goto retry
            BOOST_ATOMIC_DETAIL_ARM_ASM_END(%[tmp])
            : [original] "=&r" (original),  // %0
              [result] "=&r" (result),      // %1
              [tmp] "=&l" (tmp),            // %2
              [storage] "+Q" (storage)      // %3
            : [value] "r" (v)               // %4
            : BOOST_ATOMIC_DETAIL_ASM_CLOBBER_CC
        );
        fence_after(order);
        return original;
    }
};

template< >
struct operations< 1u, true > :
    public operations< 4u, true >
{
    typedef operations< 4u, true > base_type;
    typedef base_type::storage_type storage_type;

    static BOOST_FORCEINLINE storage_type fetch_add(storage_type volatile& storage, storage_type v, memory_order order) BOOST_NOEXCEPT
    {
        fence_before(order);
        uint32_t tmp;
        storage_type original, result;
        __asm__ __volatile__
        (
            BOOST_ATOMIC_DETAIL_ARM_ASM_START(%[tmp])
            "1:\n"
            "ldrex   %[original], %[storage]\n"           // original = *(&storage)
            "add     %[result], %[original], %[value]\n"  // result = original + value
            "sxtb    %[result], %[result]\n"              // sign extend result from 8 to 32 bits
            "strex   %[tmp], %[result], %[storage]\n"     // *(&storage) = result, tmp = store failed
            "teq     %[tmp], #0\n"                        // flags = tmp==0
            "bne     1b\n"                                // if (!flags.equal) goto retry
            BOOST_ATOMIC_DETAIL_ARM_ASM_END(%[tmp])
            : [original] "=&r" (original),  // %0
              [result] "=&r" (result),      // %1
              [tmp] "=&l" (tmp),            // %2
              [storage] "+Q" (storage)      // %3
            : [value] "r" (v)               // %4
            : BOOST_ATOMIC_DETAIL_ASM_CLOBBER_CC
        );
        fence_after(order);
        return original;
    }

    static BOOST_FORCEINLINE storage_type fetch_sub(storage_type volatile& storage, storage_type v, memory_order order) BOOST_NOEXCEPT
    {
        fence_before(order);
        uint32_t tmp;
        storage_type original, result;
        __asm__ __volatile__
        (
            BOOST_ATOMIC_DETAIL_ARM_ASM_START(%[tmp])
            "1:\n"
            "ldrex   %[original], %[storage]\n"           // original = *(&storage)
            "sub     %[result], %[original], %[value]\n"  // result = original - value
            "sxtb    %[result], %[result]\n"              // sign extend result from 8 to 32 bits
            "strex   %[tmp], %[result], %[storage]\n"     // *(&storage) = result, tmp = store failed
            "teq     %[tmp], #0\n"                        // flags = tmp==0
            "bne     1b\n"                                // if (!flags.equal) goto retry
            BOOST_ATOMIC_DETAIL_ARM_ASM_END(%[tmp])
            : [original] "=&r" (original),  // %0
              [result] "=&r" (result),      // %1
              [tmp] "=&l" (tmp),            // %2
              [storage] "+Q" (storage)      // %3
            : [value] "r" (v)               // %4
            : BOOST_ATOMIC_DETAIL_ASM_CLOBBER_CC
        );
        fence_after(order);
        return original;
    }
};


template< >
struct operations< 2u, false > :
    public operations< 4u, false >
{
    typedef operations< 4u, false > base_type;
    typedef base_type::storage_type storage_type;

    static BOOST_FORCEINLINE storage_type fetch_add(storage_type volatile& storage, storage_type v, memory_order order) BOOST_NOEXCEPT
    {
        fence_before(order);
        uint32_t tmp;
        storage_type original, result;
        __asm__ __volatile__
        (
            BOOST_ATOMIC_DETAIL_ARM_ASM_START(%[tmp])
            "1:\n"
            "ldrex   %[original], %[storage]\n"           // original = *(&storage)
            "add     %[result], %[original], %[value]\n"  // result = original + value
            "uxth    %[result], %[result]\n"              // zero extend result from 16 to 32 bits
            "strex   %[tmp], %[result], %[storage]\n"     // *(&storage) = result, tmp = store failed
            "teq     %[tmp], #0\n"                        // flags = tmp==0
            "bne     1b\n"                                // if (!flags.equal) goto retry
            BOOST_ATOMIC_DETAIL_ARM_ASM_END(%[tmp])
            : [original] "=&r" (original),  // %0
              [result] "=&r" (result),      // %1
              [tmp] "=&l" (tmp),            // %2
              [storage] "+Q" (storage)      // %3
            : [value] "r" (v)               // %4
            : BOOST_ATOMIC_DETAIL_ASM_CLOBBER_CC
        );
        fence_after(order);
        return original;
    }

    static BOOST_FORCEINLINE storage_type fetch_sub(storage_type volatile& storage, storage_type v, memory_order order) BOOST_NOEXCEPT
    {
        fence_before(order);
        uint32_t tmp;
        storage_type original, result;
        __asm__ __volatile__
        (
            BOOST_ATOMIC_DETAIL_ARM_ASM_START(%[tmp])
            "1:\n"
            "ldrex   %[original], %[storage]\n"           // original = *(&storage)
            "sub     %[result], %[original], %[value]\n"  // result = original - value
            "uxth    %[result], %[result]\n"              // zero extend result from 16 to 32 bits
            "strex   %[tmp], %[result], %[storage]\n"     // *(&storage) = result, tmp = store failed
            "teq     %[tmp], #0\n"                        // flags = tmp==0
            "bne     1b\n"                                // if (!flags.equal) goto retry
            BOOST_ATOMIC_DETAIL_ARM_ASM_END(%[tmp])
            : [original] "=&r" (original),  // %0
              [result] "=&r" (result),      // %1
              [tmp] "=&l" (tmp),            // %2
              [storage] "+Q" (storage)      // %3
            : [value] "r" (v)               // %4
            : BOOST_ATOMIC_DETAIL_ASM_CLOBBER_CC
        );
        fence_after(order);
        return original;
    }
};

template< >
struct operations< 2u, true > :
    public operations< 4u, true >
{
    typedef operations< 4u, true > base_type;
    typedef base_type::storage_type storage_type;

    static BOOST_FORCEINLINE storage_type fetch_add(storage_type volatile& storage, storage_type v, memory_order order) BOOST_NOEXCEPT
    {
        fence_before(order);
        uint32_t tmp;
        storage_type original, result;
        __asm__ __volatile__
        (
            BOOST_ATOMIC_DETAIL_ARM_ASM_START(%[tmp])
            "1:\n"
            "ldrex   %[original], %[storage]\n"           // original = *(&storage)
            "add     %[result], %[original], %[value]\n"  // result = original + value
            "sxth    %[result], %[result]\n"              // sign extend result from 16 to 32 bits
            "strex   %[tmp], %[result], %[storage]\n"     // *(&storage) = result, tmp = store failed
            "teq     %[tmp], #0\n"                        // flags = tmp==0
            "bne     1b\n"                                // if (!flags.equal) goto retry
            BOOST_ATOMIC_DETAIL_ARM_ASM_END(%[tmp])
            : [original] "=&r" (original),  // %0
              [result] "=&r" (result),      // %1
              [tmp] "=&l" (tmp),            // %2
              [storage] "+Q" (storage)      // %3
            : [value] "r" (v)               // %4
            : BOOST_ATOMIC_DETAIL_ASM_CLOBBER_CC
        );
        fence_after(order);
        return original;
    }

    static BOOST_FORCEINLINE storage_type fetch_sub(storage_type volatile& storage, storage_type v, memory_order order) BOOST_NOEXCEPT
    {
        fence_before(order);
        uint32_t tmp;
        storage_type original, result;
        __asm__ __volatile__
        (
            BOOST_ATOMIC_DETAIL_ARM_ASM_START(%[tmp])
            "1:\n"
            "ldrex   %[original], %[storage]\n"           // original = *(&storage)
            "sub     %[result], %[original], %[value]\n"  // result = original - value
            "sxth    %[result], %[result]\n"              // sign extend result from 16 to 32 bits
            "strex   %[tmp], %[result], %[storage]\n"     // *(&storage) = result, tmp = store failed
            "teq     %[tmp], #0\n"                        // flags = tmp==0
            "bne     1b\n"                                // if (!flags.equal) goto retry
            BOOST_ATOMIC_DETAIL_ARM_ASM_END(%[tmp])
            : [original] "=&r" (original),  // %0
              [result] "=&r" (result),      // %1
              [tmp] "=&l" (tmp),            // %2
              [storage] "+Q" (storage)      // %3
            : [value] "r" (v)               // %4
            : BOOST_ATOMIC_DETAIL_ASM_CLOBBER_CC
        );
        fence_after(order);
        return original;
    }
};


#if defined(BOOST_ATOMIC_DETAIL_ARM_HAS_LDREXD_STREXD)

// Unlike 32-bit operations, for 64-bit loads and stores we must use ldrexd/strexd.
// Any other instructions result in a non-atomic sequence of 32-bit accesses.
// See "ARM Architecture Reference Manual ARMv7-A and ARMv7-R edition",
// Section A3.5.3 "Atomicity in the ARM architecture".

// In the asm blocks below we have to use 32-bit register pairs to compose 64-bit values.
// In order to pass the 64-bit operands to/from asm blocks, we use undocumented gcc feature:
// the lower half (Rt) of the operand is accessible normally, via the numbered placeholder (e.g. %0),
// and the upper half (Rt2) - via the same placeholder with an 'H' after the '%' sign (e.g. %H0).
// See: http://hardwarebug.org/2010/07/06/arm-inline-asm-secrets/

template< bool Signed >
struct operations< 8u, Signed > :
    public gcc_arm_operations_base
{
    typedef typename make_storage_type< 8u, Signed >::type storage_type;
    typedef typename make_storage_type< 8u, Signed >::aligned aligned_storage_type;

    static BOOST_FORCEINLINE void store(storage_type volatile& storage, storage_type v, memory_order order) BOOST_NOEXCEPT
    {
        exchange(storage, v, order);
    }

    static BOOST_FORCEINLINE storage_type load(storage_type const volatile& storage, memory_order order) BOOST_NOEXCEPT
    {
        storage_type original;
        uint32_t tmp;
        __asm__ __volatile__
        (
            BOOST_ATOMIC_DETAIL_ARM_ASM_START(%0)
            "ldrexd %1, %H1, [%2]\n"
            BOOST_ATOMIC_DETAIL_ARM_ASM_END(%0)
            : BOOST_ATOMIC_DETAIL_ARM_ASM_TMPREG_CONSTRAINT(tmp), // %0
              "=&r" (original)   // %1
            : "r" (&storage)     // %2
        );
        fence_after(order);
        return original;
    }

    static BOOST_FORCEINLINE storage_type exchange(storage_type volatile& storage, storage_type v, memory_order order) BOOST_NOEXCEPT
    {
        storage_type original;
        fence_before(order);
        uint32_t tmp;
        __asm__ __volatile__
        (
            BOOST_ATOMIC_DETAIL_ARM_ASM_START(%0)
            "1:\n"
            "ldrexd %1, %H1, [%3]\n"        // load the original value
            "strexd %0, %2, %H2, [%3]\n"    // store the replacement, tmp = store failed
            "teq    %0, #0\n"               // check if store succeeded
            "bne    1b\n"
            BOOST_ATOMIC_DETAIL_ARM_ASM_END(%0)
            : BOOST_ATOMIC_DETAIL_ARM_ASM_TMPREG_CONSTRAINT(tmp), // %0
              "=&r" (original)   // %1
            : "r" (v),           // %2
              "r" (&storage)     // %3
            : BOOST_ATOMIC_DETAIL_ASM_CLOBBER_CC_COMMA "memory"
        );
        fence_after(order);
        return original;
    }

    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
    {
        fence_before(success_order);
        uint32_t tmp;
        storage_type original, old_val = expected;
        __asm__ __volatile__
        (
            BOOST_ATOMIC_DETAIL_ARM_ASM_START(%0)
            "ldrexd   %1, %H1, [%3]\n"               // original = *(&storage)
            "cmp      %1, %2\n"                      // flags = original.lo==old_val.lo
            "ittt     eq\n"                          // [hint that the following 3 instructions are conditional on flags.equal]
            "cmpeq    %H1, %H2\n"                    // if (flags.equal) flags = original.hi==old_val.hi
            "strexdeq %0, %4, %H4, [%3]\n"           // if (flags.equal) *(&storage) = desired, tmp = store failed
            "teqeq    %0, #0\n"                      // if (flags.equal) flags = tmp==0
            "ite      eq\n"                          // [hint that the following 2 instructions are conditional on flags.equal]
            "moveq    %2, #1\n"                      // if (flags.equal) old_val.lo = 1
            "movne    %2, #0\n"                      // if (!flags.equal) old_val.lo = 0
            BOOST_ATOMIC_DETAIL_ARM_ASM_END(%0)
            : BOOST_ATOMIC_DETAIL_ARM_ASM_TMPREG_CONSTRAINT(tmp), // %0
              "=&r" (original),  // %1
              "+r" (old_val)     // %2
            : "r" (&storage),    // %3
              "r" (desired)      // %4
            : BOOST_ATOMIC_DETAIL_ASM_CLOBBER_CC_COMMA "memory"
        );
        const uint32_t success = (uint32_t)old_val;
        if (success)
            fence_after(success_order);
        else
            fence_after(failure_order);
        expected = original;
        return !!success;
    }

    static BOOST_FORCEINLINE bool compare_exchange_strong(
        storage_type volatile& storage, storage_type& expected, storage_type desired, memory_order success_order, memory_order failure_order) BOOST_NOEXCEPT
    {
        fence_before(success_order);
        uint32_t tmp;
        storage_type original, old_val = expected;
        __asm__ __volatile__
        (
            BOOST_ATOMIC_DETAIL_ARM_ASM_START(%0)
            "1:\n"
            "ldrexd  %1, %H1, [%3]\n"               // original = *(&storage)
            "cmp     %1, %2\n"                      // flags = original.lo==old_val.lo
            "it      eq\n"                          // [hint that the following instruction is conditional on flags.equal]
            "cmpeq   %H1, %H2\n"                    // if (flags.equal) flags = original.hi==old_val.hi
            "bne     2f\n"                          // if (!flags.equal) goto end
            "strexd  %0, %4, %H4, [%3]\n"           // *(&storage) = desired, tmp = store failed
            "teq     %0, #0\n"                      // flags.equal = tmp == 0
            "bne     1b\n"                          // if (flags.equal) goto retry
            "2:\n"
            "ite      eq\n"                         // [hint that the following 2 instructions are conditional on flags.equal]
            "moveq    %2, #1\n"                     // if (flags.equal) old_val.lo = 1
            "movne    %2, #0\n"                     // if (!flags.equal) old_val.lo = 0
            BOOST_ATOMIC_DETAIL_ARM_ASM_END(%0)
            : BOOST_ATOMIC_DETAIL_ARM_ASM_TMPREG_CONSTRAINT(tmp), // %0
              "=&r" (original),  // %1
              "+r" (old_val)     // %2
            : "r" (&storage),    // %3
              "r" (desired)      // %4
            : BOOST_ATOMIC_DETAIL_ASM_CLOBBER_CC_COMMA "memory"
        );
        const uint32_t success = (uint32_t)old_val;
        if (success)
            fence_after(success_order);
        else
            fence_after(failure_order);
        expected = original;
        return !!success;
    }

    static BOOST_FORCEINLINE storage_type fetch_add(storage_type volatile& storage, storage_type v, memory_order order) BOOST_NOEXCEPT
    {
        fence_before(order);
        storage_type original, result;
        uint32_t tmp;
        __asm__ __volatile__
        (
            BOOST_ATOMIC_DETAIL_ARM_ASM_START(%0)
            "1:\n"
            "ldrexd  %1, %H1, [%3]\n"               // original = *(&storage)
            "adds    %2, %1, %4\n"                  // result = original + value
            "adc     %H2, %H1, %H4\n"
            "strexd  %0, %2, %H2, [%3]\n"           // *(&storage) = result, tmp = store failed
            "teq     %0, #0\n"                      // flags = tmp==0
            "bne     1b\n"                          // if (!flags.equal) goto retry
            BOOST_ATOMIC_DETAIL_ARM_ASM_END(%0)
            : BOOST_ATOMIC_DETAIL_ARM_ASM_TMPREG_CONSTRAINT(tmp), // %0
              "=&r" (original),  // %1
              "=&r" (result)     // %2
            : "r" (&storage),    // %3
              "r" (v)            // %4
            : BOOST_ATOMIC_DETAIL_ASM_CLOBBER_CC_COMMA "memory"
        );
        fence_after(order);
        return original;
    }

    static BOOST_FORCEINLINE storage_type fetch_sub(storage_type volatile& storage, storage_type v, memory_order order) BOOST_NOEXCEPT
    {
        fence_before(order);
        storage_type original, result;
        uint32_t tmp;
        __asm__ __volatile__
        (
            BOOST_ATOMIC_DETAIL_ARM_ASM_START(%0)
            "1:\n"
            "ldrexd  %1, %H1, [%3]\n"               // original = *(&storage)
            "subs    %2, %1, %4\n"                  // result = original - value
            "sbc     %H2, %H1, %H4\n"
            "strexd  %0, %2, %H2, [%3]\n"           // *(&storage) = result, tmp = store failed
            "teq     %0, #0\n"                      // flags = tmp==0
            "bne     1b\n"                          // if (!flags.equal) goto retry
            BOOST_ATOMIC_DETAIL_ARM_ASM_END(%0)
            : BOOST_ATOMIC_DETAIL_ARM_ASM_TMPREG_CONSTRAINT(tmp), // %0
              "=&r" (original),  // %1
              "=&r" (result)     // %2
            : "r" (&storage),    // %3
              "r" (v)            // %4
            : BOOST_ATOMIC_DETAIL_ASM_CLOBBER_CC_COMMA "memory"
        );
        fence_after(order);
        return original;
    }

    static BOOST_FORCEINLINE storage_type fetch_and(storage_type volatile& storage, storage_type v, memory_order order) BOOST_NOEXCEPT
    {
        fence_before(order);
        storage_type original, result;
        uint32_t tmp;
        __asm__ __volatile__
        (
            BOOST_ATOMIC_DETAIL_ARM_ASM_START(%0)
            "1:\n"
            "ldrexd  %1, %H1, [%3]\n"               // original = *(&storage)
            "and     %2, %1, %4\n"                  // result = original & value
            "and     %H2, %H1, %H4\n"
            "strexd  %0, %2, %H2, [%3]\n"           // *(&storage) = result, tmp = store failed
            "teq     %0, #0\n"                      // flags = tmp==0
            "bne     1b\n"                          // if (!flags.equal) goto retry
            BOOST_ATOMIC_DETAIL_ARM_ASM_END(%0)
            : BOOST_ATOMIC_DETAIL_ARM_ASM_TMPREG_CONSTRAINT(tmp), // %0
              "=&r" (original),  // %1
              "=&r" (result)     // %2
            : "r" (&storage),    // %3
              "r" (v)            // %4
            : BOOST_ATOMIC_DETAIL_ASM_CLOBBER_CC_COMMA "memory"
        );
        fence_after(order);
        return original;
    }

    static BOOST_FORCEINLINE storage_type fetch_or(storage_type volatile& storage, storage_type v, memory_order order) BOOST_NOEXCEPT
    {
        fence_before(order);
        storage_type original, result;
        uint32_t tmp;
        __asm__ __volatile__
        (
            BOOST_ATOMIC_DETAIL_ARM_ASM_START(%0)
            "1:\n"
            "ldrexd  %1, %H1, [%3]\n"               // original = *(&storage)
            "orr     %2, %1, %4\n"                  // result = original | value
            "orr     %H2, %H1, %H4\n"
            "strexd  %0, %2, %H2, [%3]\n"           // *(&storage) = result, tmp = store failed
            "teq     %0, #0\n"                      // flags = tmp==0
            "bne     1b\n"                          // if (!flags.equal) goto retry
            BOOST_ATOMIC_DETAIL_ARM_ASM_END(%0)
            : BOOST_ATOMIC_DETAIL_ARM_ASM_TMPREG_CONSTRAINT(tmp), // %0
              "=&r" (original),  // %1
              "=&r" (result)     // %2
            : "r" (&storage),    // %3
              "r" (v)            // %4
            : BOOST_ATOMIC_DETAIL_ASM_CLOBBER_CC_COMMA "memory"
        );
        fence_after(order);
        return original;
    }

    static BOOST_FORCEINLINE storage_type fetch_xor(storage_type volatile& storage, storage_type v, memory_order order) BOOST_NOEXCEPT
    {
        fence_before(order);
        storage_type original, result;
        uint32_t tmp;
        __asm__ __volatile__
        (
            BOOST_ATOMIC_DETAIL_ARM_ASM_START(%0)
            "1:\n"
            "ldrexd  %1, %H1, [%3]\n"               // original = *(&storage)
            "eor     %2, %1, %4\n"                  // result = original ^ value
            "eor     %H2, %H1, %H4\n"
            "strexd  %0, %2, %H2, [%3]\n"           // *(&storage) = result, tmp = store failed
            "teq     %0, #0\n"                      // flags = tmp==0
            "bne     1b\n"                          // if (!flags.equal) goto retry
            BOOST_ATOMIC_DETAIL_ARM_ASM_END(%0)
            : BOOST_ATOMIC_DETAIL_ARM_ASM_TMPREG_CONSTRAINT(tmp), // %0
              "=&r" (original),  // %1
              "=&r" (result)     // %2
            : "r" (&storage),    // %3
              "r" (v)            // %4
            : BOOST_ATOMIC_DETAIL_ASM_CLOBBER_CC_COMMA "memory"
        );
        fence_after(order);
        return original;
    }

    static BOOST_FORCEINLINE bool test_and_set(storage_type volatile& storage, memory_order order) BOOST_NOEXCEPT
    {
        return !!exchange(storage, (storage_type)1, order);
    }

    static BOOST_FORCEINLINE void clear(storage_type volatile& storage, memory_order order) BOOST_NOEXCEPT
    {
        store(storage, 0, order);
    }

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

#endif // defined(BOOST_ATOMIC_DETAIL_ARM_HAS_LDREXD_STREXD)


BOOST_FORCEINLINE void thread_fence(memory_order order) BOOST_NOEXCEPT
{
    if (order != memory_order_relaxed)
        gcc_arm_operations_base::hardware_full_fence();
}

BOOST_FORCEINLINE void signal_fence(memory_order order) BOOST_NOEXCEPT
{
    if (order != memory_order_relaxed)
        __asm__ __volatile__ ("" ::: "memory");
}

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

#endif // BOOST_ATOMIC_DETAIL_OPS_GCC_ARM_HPP_INCLUDED_