v/thirdparty/libatomic_ops/atomic_ops/sysdeps/gcc/aarch64.h

/*
 * 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.
 * Copyright (c) 2013-2017 Ivan Maidanski
 *
 * 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.
 *
 */

/* As of clang-5.0 (and gcc-5.4), __atomic_thread_fence is always       */
/* translated to DMB (which is inefficient for AO_nop_write).           */
/* TODO: Update it for newer Clang and GCC releases. */
#if !defined(AO_PREFER_BUILTIN_ATOMICS) && !defined(AO_THREAD_SANITIZER) \
    && !defined(AO_UNIPROCESSOR)
  AO_INLINE void
  AO_nop_write(void)
  {
    __asm__ __volatile__("dmb ishst" : : : "memory");
  }
# define AO_HAVE_nop_write
#endif

/* There were some bugs in the older clang releases (related to         */
/* optimization of functions dealing with __int128 values, supposedly), */
/* so even asm-based implementation did not work correctly.             */
#if !defined(__clang__) || AO_CLANG_PREREQ(3, 9)

# include "../standard_ao_double_t.h"

/* As of gcc-5.4, all built-in load/store and CAS atomics for double    */
/* word require -latomic, are not lock-free and cause test_stack        */
/* failure, so the asm-based implementation is used for now.            */
/* TODO: Update it for newer GCC releases. */
#if (!defined(__ILP32__) && !defined(__clang__)) \
    || defined(AO_AARCH64_ASM_LOAD_STORE_CAS)

# ifndef AO_PREFER_GENERALIZED
    AO_INLINE AO_double_t
    AO_double_load(const volatile AO_double_t *addr)
    {
      AO_double_t result;
      int status;

      /* Note that STXP cannot be discarded because LD[A]XP is not      */
      /* single-copy atomic (unlike LDREXD for 32-bit ARM).             */
      do {
        __asm__ __volatile__("//AO_double_load\n"
#       ifdef __ILP32__
          "       ldxp  %w0, %w1, %3\n"
          "       stxp %w2, %w0, %w1, %3"
#       else
          "       ldxp  %0, %1, %3\n"
          "       stxp %w2, %0, %1, %3"
#       endif
        : "=&r" (result.AO_val1), "=&r" (result.AO_val2), "=&r" (status)
        : "Q" (*addr));
      } while (AO_EXPECT_FALSE(status));
      return result;
    }
#   define AO_HAVE_double_load

    AO_INLINE AO_double_t
    AO_double_load_acquire(const volatile AO_double_t *addr)
    {
      AO_double_t result;
      int status;

      do {
        __asm__ __volatile__("//AO_double_load_acquire\n"
#       ifdef __ILP32__
          "       ldaxp  %w0, %w1, %3\n"
          "       stxp %w2, %w0, %w1, %3"
#       else
          "       ldaxp  %0, %1, %3\n"
          "       stxp %w2, %0, %1, %3"
#       endif
        : "=&r" (result.AO_val1), "=&r" (result.AO_val2), "=&r" (status)
        : "Q" (*addr));
      } while (AO_EXPECT_FALSE(status));
      return result;
    }
#   define AO_HAVE_double_load_acquire

    AO_INLINE void
    AO_double_store(volatile AO_double_t *addr, AO_double_t value)
    {
      AO_double_t old_val;
      int status;

      do {
        __asm__ __volatile__("//AO_double_store\n"
#       ifdef __ILP32__
          "       ldxp  %w0, %w1, %3\n"
          "       stxp %w2, %w4, %w5, %3"
#       else
          "       ldxp  %0, %1, %3\n"
          "       stxp %w2, %4, %5, %3"
#       endif
        : "=&r" (old_val.AO_val1), "=&r" (old_val.AO_val2), "=&r" (status),
          "=Q" (*addr)
        : "r" (value.AO_val1), "r" (value.AO_val2));
        /* Compared to the arm.h implementation, the 'cc' (flags) are   */
        /* not clobbered because A64 has no concept of conditional      */
        /* execution.                                                   */
      } while (AO_EXPECT_FALSE(status));
    }
#   define AO_HAVE_double_store

    AO_INLINE void
    AO_double_store_release(volatile AO_double_t *addr, AO_double_t value)
    {
      AO_double_t old_val;
      int status;

      do {
        __asm__ __volatile__("//AO_double_store_release\n"
#       ifdef __ILP32__
          "       ldxp  %w0, %w1, %3\n"
          "       stlxp %w2, %w4, %w5, %3"
#       else
          "       ldxp  %0, %1, %3\n"
          "       stlxp %w2, %4, %5, %3"
#       endif
        : "=&r" (old_val.AO_val1), "=&r" (old_val.AO_val2), "=&r" (status),
          "=Q" (*addr)
        : "r" (value.AO_val1), "r" (value.AO_val2));
      } while (AO_EXPECT_FALSE(status));
    }
#   define AO_HAVE_double_store_release
# endif /* !AO_PREFER_GENERALIZED */

  AO_INLINE int
  AO_double_compare_and_swap(volatile AO_double_t *addr,
                             AO_double_t old_val, AO_double_t new_val)
  {
    AO_double_t tmp;
    int result = 1;

    do {
      __asm__ __volatile__("//AO_double_compare_and_swap\n"
#       ifdef __ILP32__
          "       ldxp  %w0, %w1, %2\n"
#       else
          "       ldxp  %0, %1, %2\n"
#       endif
        : "=&r" (tmp.AO_val1), "=&r" (tmp.AO_val2)
        : "Q" (*addr));
      if (tmp.AO_val1 != old_val.AO_val1 || tmp.AO_val2 != old_val.AO_val2)
        break;
      __asm__ __volatile__(
#       ifdef __ILP32__
          "       stxp %w0, %w2, %w3, %1\n"
#       else
          "       stxp %w0, %2, %3, %1\n"
#       endif
        : "=&r" (result), "=Q" (*addr)
        : "r" (new_val.AO_val1), "r" (new_val.AO_val2));
    } while (AO_EXPECT_FALSE(result));
    return !result;
  }
# define AO_HAVE_double_compare_and_swap

  AO_INLINE int
  AO_double_compare_and_swap_acquire(volatile AO_double_t *addr,
                                     AO_double_t old_val, AO_double_t new_val)
  {
    AO_double_t tmp;
    int result = 1;

    do {
      __asm__ __volatile__("//AO_double_compare_and_swap_acquire\n"
#       ifdef __ILP32__
          "       ldaxp  %w0, %w1, %2\n"
#       else
          "       ldaxp  %0, %1, %2\n"
#       endif
        : "=&r" (tmp.AO_val1), "=&r" (tmp.AO_val2)
        : "Q" (*addr));
      if (tmp.AO_val1 != old_val.AO_val1 || tmp.AO_val2 != old_val.AO_val2)
        break;
      __asm__ __volatile__(
#       ifdef __ILP32__
          "       stxp %w0, %w2, %w3, %1\n"
#       else
          "       stxp %w0, %2, %3, %1\n"
#       endif
        : "=&r" (result), "=Q" (*addr)
        : "r" (new_val.AO_val1), "r" (new_val.AO_val2));
    } while (AO_EXPECT_FALSE(result));
    return !result;
  }
# define AO_HAVE_double_compare_and_swap_acquire

  AO_INLINE int
  AO_double_compare_and_swap_release(volatile AO_double_t *addr,
                                     AO_double_t old_val, AO_double_t new_val)
  {
    AO_double_t tmp;
    int result = 1;

    do {
      __asm__ __volatile__("//AO_double_compare_and_swap_release\n"
#       ifdef __ILP32__
          "       ldxp  %w0, %w1, %2\n"
#       else
          "       ldxp  %0, %1, %2\n"
#       endif
        : "=&r" (tmp.AO_val1), "=&r" (tmp.AO_val2)
        : "Q" (*addr));
      if (tmp.AO_val1 != old_val.AO_val1 || tmp.AO_val2 != old_val.AO_val2)
        break;
      __asm__ __volatile__(
#       ifdef __ILP32__
          "       stlxp %w0, %w2, %w3, %1\n"
#       else
          "       stlxp %w0, %2, %3, %1\n"
#       endif
        : "=&r" (result), "=Q" (*addr)
        : "r" (new_val.AO_val1), "r" (new_val.AO_val2));
    } while (AO_EXPECT_FALSE(result));
    return !result;
  }
# define AO_HAVE_double_compare_and_swap_release

  AO_INLINE int
  AO_double_compare_and_swap_full(volatile AO_double_t *addr,
                                  AO_double_t old_val, AO_double_t new_val)
  {
    AO_double_t tmp;
    int result = 1;

    do {
      __asm__ __volatile__("//AO_double_compare_and_swap_full\n"
#       ifdef __ILP32__
          "       ldaxp  %w0, %w1, %2\n"
#       else
          "       ldaxp  %0, %1, %2\n"
#       endif
        : "=&r" (tmp.AO_val1), "=&r" (tmp.AO_val2)
        : "Q" (*addr));
      if (tmp.AO_val1 != old_val.AO_val1 || tmp.AO_val2 != old_val.AO_val2)
        break;
      __asm__ __volatile__(
#       ifdef __ILP32__
          "       stlxp %w0, %w2, %w3, %1\n"
#       else
          "       stlxp %w0, %2, %3, %1\n"
#       endif
        : "=&r" (result), "=Q" (*addr)
        : "r" (new_val.AO_val1), "r" (new_val.AO_val2));
    } while (AO_EXPECT_FALSE(result));
    return !result;
  }
# define AO_HAVE_double_compare_and_swap_full

#endif /* !__ILP32__ && !__clang__ || AO_AARCH64_ASM_LOAD_STORE_CAS */

/* As of clang-5.0 and gcc-8.1, __GCC_HAVE_SYNC_COMPARE_AND_SWAP_16     */
/* macro is still missing (while the double-word CAS is available).     */
# ifndef __ILP32__
#   define AO_GCC_HAVE_double_SYNC_CAS
# endif

#endif /* !__clang__ || AO_CLANG_PREREQ(3, 9) */

#if (defined(__clang__) && !AO_CLANG_PREREQ(3, 8)) || defined(__APPLE_CC__)
  /* __GCC_HAVE_SYNC_COMPARE_AND_SWAP_n macros are missing.     */
# define AO_GCC_FORCE_HAVE_CAS
#endif

#include "generic.h"

#undef AO_GCC_FORCE_HAVE_CAS
#undef AO_GCC_HAVE_double_SYNC_CAS