/*
 * 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) 2014 Andrey Semashev
 */
/*!
 * \file   atomic/detail/ops_gcc_atomic.hpp
 *
 * This header contains implementation of the \c operations template.
 */

#ifndef BOOST_ATOMIC_DETAIL_OPS_GCC_ATOMIC_HPP_INCLUDED_
#define BOOST_ATOMIC_DETAIL_OPS_GCC_ATOMIC_HPP_INCLUDED_

#include <cstddef>
#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/capabilities.hpp>
#if (defined(__clang__) || (defined(BOOST_GCC) && (BOOST_GCC+0) >= 70000)) && (defined(BOOST_ATOMIC_DETAIL_X86_HAS_CMPXCHG8B) || defined(BOOST_ATOMIC_DETAIL_X86_HAS_CMPXCHG16B))
#include <boost/atomic/detail/ops_gcc_x86_dcas.hpp>
#include <boost/atomic/detail/ops_cas_based.hpp>
#endif

#if __GCC_ATOMIC_LLONG_LOCK_FREE != BOOST_ATOMIC_LLONG_LOCK_FREE || __GCC_ATOMIC_LONG_LOCK_FREE != BOOST_ATOMIC_LONG_LOCK_FREE ||\
    __GCC_ATOMIC_INT_LOCK_FREE != BOOST_ATOMIC_INT_LOCK_FREE || __GCC_ATOMIC_SHORT_LOCK_FREE != BOOST_ATOMIC_SHORT_LOCK_FREE ||\
    __GCC_ATOMIC_CHAR_LOCK_FREE != BOOST_ATOMIC_CHAR_LOCK_FREE || __GCC_ATOMIC_BOOL_LOCK_FREE != BOOST_ATOMIC_BOOL_LOCK_FREE ||\
    __GCC_ATOMIC_WCHAR_T_LOCK_FREE != BOOST_ATOMIC_WCHAR_T_LOCK_FREE
// There are platforms where we need to use larger storage types
#include <boost/atomic/detail/int_sizes.hpp>
#include <boost/atomic/detail/ops_extending_cas_based.hpp>
#endif

#ifdef BOOST_HAS_PRAGMA_ONCE
#pragma once
#endif

#if defined(__INTEL_COMPILER)
// This is used to suppress warning #32013 described below for Intel Compiler.
// In debug builds the compiler does not inline any functions, so basically
// every atomic function call results in this warning. I don't know any other
// way to selectively disable just this one warning.
#pragma system_header
#endif

namespace lslboost {
namespace atomics {
namespace detail {

/*!
 * The function converts \c lslboost::memory_order values to the compiler-specific constants.
 *
 * NOTE: The intention is that the function is optimized away by the compiler, and the
 *       compiler-specific constants are passed to the intrinsics. Unfortunately, constexpr doesn't
 *       work in this case because the standard atomics interface require memory ordering
 *       constants to be passed as function arguments, at which point they stop being constexpr.
 *       However, it is crucial that the compiler sees constants and not runtime values,
 *       because otherwise it just ignores the ordering value and always uses seq_cst.
 *       This is the case with Intel C++ Compiler 14.0.3 (Composer XE 2013 SP1, update 3) and
 *       gcc 4.8.2. Intel Compiler issues a warning in this case:
 *
 *       warning #32013: Invalid memory order specified. Defaulting to seq_cst memory order.
 *
 *       while gcc acts silently.
 *
 *       To mitigate the problem ALL functions, including the atomic<> members must be
 *       declared with BOOST_FORCEINLINE. In this case the compilers are able to see that
 *       all functions are called with constant orderings and call intrinstcts properly.
 *
 *       Unfortunately, this still doesn't work in debug mode as the compiler doesn't
 *       propagate constants even when functions are marked with BOOST_FORCEINLINE. In this case
 *       all atomic operaions will be executed with seq_cst semantics.
 */
BOOST_FORCEINLINE BOOST_CONSTEXPR int convert_memory_order_to_gcc(memory_order order) BOOST_NOEXCEPT
{
    return (order == memory_order_relaxed ? __ATOMIC_RELAXED : (order == memory_order_consume ? __ATOMIC_CONSUME :
        (order == memory_order_acquire ? __ATOMIC_ACQUIRE : (order == memory_order_release ? __ATOMIC_RELEASE :
        (order == memory_order_acq_rel ? __ATOMIC_ACQ_REL : __ATOMIC_SEQ_CST)))));
}

template< std::size_t Size, bool Signed >
struct gcc_atomic_operations
{
    typedef typename make_storage_type< Size >::type storage_type;
    typedef typename make_storage_type< Size >::aligned aligned_storage_type;

    static BOOST_CONSTEXPR_OR_CONST std::size_t storage_size = Size;
    static BOOST_CONSTEXPR_OR_CONST bool is_signed = Signed;
    static BOOST_CONSTEXPR_OR_CONST bool full_cas_based = false;

    // Note: In the current implementation, gcc_atomic_operations are used only when the particularly sized __atomic
    // intrinsics are always lock-free (i.e. the corresponding LOCK_FREE macro is 2). Therefore it is safe to
    // always set is_always_lock_free to true here.
    static BOOST_CONSTEXPR_OR_CONST bool is_always_lock_free = true;

    static BOOST_FORCEINLINE void store(storage_type volatile& storage, storage_type v, memory_order order) BOOST_NOEXCEPT
    {
        __atomic_store_n(&storage, v, atomics::detail::convert_memory_order_to_gcc(order));
    }

    static BOOST_FORCEINLINE storage_type load(storage_type const volatile& storage, memory_order order) BOOST_NOEXCEPT
    {
        return __atomic_load_n(&storage, atomics::detail::convert_memory_order_to_gcc(order));
    }

    static BOOST_FORCEINLINE storage_type fetch_add(storage_type volatile& storage, storage_type v, memory_order order) BOOST_NOEXCEPT
    {
        return __atomic_fetch_add(&storage, v, atomics::detail::convert_memory_order_to_gcc(order));
    }

    static BOOST_FORCEINLINE storage_type fetch_sub(storage_type volatile& storage, storage_type v, memory_order order) BOOST_NOEXCEPT
    {
        return __atomic_fetch_sub(&storage, v, atomics::detail::convert_memory_order_to_gcc(order));
    }

    static BOOST_FORCEINLINE storage_type exchange(storage_type volatile& storage, storage_type v, memory_order order) BOOST_NOEXCEPT
    {
        return __atomic_exchange_n(&storage, v, atomics::detail::convert_memory_order_to_gcc(order));
    }

    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
    {
        return __atomic_compare_exchange_n
        (
            &storage, &expected, desired, false,
            atomics::detail::convert_memory_order_to_gcc(success_order),
            atomics::detail::convert_memory_order_to_gcc(failure_order)
        );
    }

    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 __atomic_compare_exchange_n
        (
            &storage, &expected, desired, true,
            atomics::detail::convert_memory_order_to_gcc(success_order),
            atomics::detail::convert_memory_order_to_gcc(failure_order)
        );
    }

    static BOOST_FORCEINLINE storage_type fetch_and(storage_type volatile& storage, storage_type v, memory_order order) BOOST_NOEXCEPT
    {
        return __atomic_fetch_and(&storage, v, atomics::detail::convert_memory_order_to_gcc(order));
    }

    static BOOST_FORCEINLINE storage_type fetch_or(storage_type volatile& storage, storage_type v, memory_order order) BOOST_NOEXCEPT
    {
        return __atomic_fetch_or(&storage, v, atomics::detail::convert_memory_order_to_gcc(order));
    }

    static BOOST_FORCEINLINE storage_type fetch_xor(storage_type volatile& storage, storage_type v, memory_order order) BOOST_NOEXCEPT
    {
        return __atomic_fetch_xor(&storage, v, atomics::detail::convert_memory_order_to_gcc(order));
    }

    static BOOST_FORCEINLINE bool test_and_set(storage_type volatile& storage, memory_order order) BOOST_NOEXCEPT
    {
        return __atomic_test_and_set(&storage, atomics::detail::convert_memory_order_to_gcc(order));
    }

    static BOOST_FORCEINLINE void clear(storage_type volatile& storage, memory_order order) BOOST_NOEXCEPT
    {
        __atomic_clear(const_cast< storage_type* >(&storage), atomics::detail::convert_memory_order_to_gcc(order));
    }
};

#if BOOST_ATOMIC_INT128_LOCK_FREE > 0
#if (defined(__clang__) || (defined(BOOST_GCC) && (BOOST_GCC+0) >= 70000)) && defined(BOOST_ATOMIC_DETAIL_X86_HAS_CMPXCHG16B)

// Workaround for clang bug: http://llvm.org/bugs/show_bug.cgi?id=19149
// Clang 3.4 does not implement 128-bit __atomic* intrinsics even though it defines __GCC_HAVE_SYNC_COMPARE_AND_SWAP_16
// A similar problem exists with gcc 7 as well, as it requires to link with libatomic to use 16-byte intrinsics:
// https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80878
template< bool Signed >
struct operations< 16u, Signed > :
    public cas_based_operations< gcc_dcas_x86_64< Signed > >
{
    static BOOST_CONSTEXPR_OR_CONST std::size_t storage_size = 16u;
    static BOOST_CONSTEXPR_OR_CONST bool is_signed = Signed;
};

#else

template< bool Signed >
struct operations< 16u, Signed > :
    public gcc_atomic_operations< 16u, Signed >
{
};

#endif
#endif


#if BOOST_ATOMIC_INT64_LOCK_FREE > 0
#if defined(__clang__) && defined(BOOST_ATOMIC_DETAIL_X86_HAS_CMPXCHG8B)

// Workaround for clang bug http://llvm.org/bugs/show_bug.cgi?id=19355
template< bool Signed >
struct operations< 8u, Signed > :
    public cas_based_operations< gcc_dcas_x86< Signed > >
{
    static BOOST_CONSTEXPR_OR_CONST std::size_t storage_size = 8u;
    static BOOST_CONSTEXPR_OR_CONST bool is_signed = Signed;
};

#elif (BOOST_ATOMIC_DETAIL_SIZEOF_LLONG == 8 && __GCC_ATOMIC_LLONG_LOCK_FREE != BOOST_ATOMIC_LLONG_LOCK_FREE) ||\
    (BOOST_ATOMIC_DETAIL_SIZEOF_LONG == 8 && __GCC_ATOMIC_LONG_LOCK_FREE != BOOST_ATOMIC_LONG_LOCK_FREE) ||\
    (BOOST_ATOMIC_DETAIL_SIZEOF_INT == 8 && __GCC_ATOMIC_INT_LOCK_FREE != BOOST_ATOMIC_INT_LOCK_FREE) ||\
    (BOOST_ATOMIC_DETAIL_SIZEOF_SHORT == 8 && __GCC_ATOMIC_SHORT_LOCK_FREE != BOOST_ATOMIC_SHORT_LOCK_FREE) ||\
    (BOOST_ATOMIC_DETAIL_SIZEOF_WCHAR_T == 8 && __GCC_ATOMIC_WCHAR_T_LOCK_FREE != BOOST_ATOMIC_WCHAR_T_LOCK_FREE)

#define BOOST_ATOMIC_DETAIL_INT64_EXTENDED

template< bool Signed >
struct operations< 8u, Signed > :
    public extending_cas_based_operations< gcc_atomic_operations< 16u, Signed >, 8u, Signed >
{
};

#else

template< bool Signed >
struct operations< 8u, Signed > :
    public gcc_atomic_operations< 8u, Signed >
{
};

#endif
#endif

#if BOOST_ATOMIC_INT32_LOCK_FREE > 0
#if (BOOST_ATOMIC_DETAIL_SIZEOF_LLONG == 4 && __GCC_ATOMIC_LLONG_LOCK_FREE != BOOST_ATOMIC_LLONG_LOCK_FREE) ||\
    (BOOST_ATOMIC_DETAIL_SIZEOF_LONG == 4 && __GCC_ATOMIC_LONG_LOCK_FREE != BOOST_ATOMIC_LONG_LOCK_FREE) ||\
    (BOOST_ATOMIC_DETAIL_SIZEOF_INT == 4 && __GCC_ATOMIC_INT_LOCK_FREE != BOOST_ATOMIC_INT_LOCK_FREE) ||\
    (BOOST_ATOMIC_DETAIL_SIZEOF_SHORT == 4 && __GCC_ATOMIC_SHORT_LOCK_FREE != BOOST_ATOMIC_SHORT_LOCK_FREE) ||\
    (BOOST_ATOMIC_DETAIL_SIZEOF_WCHAR_T == 4 && __GCC_ATOMIC_WCHAR_T_LOCK_FREE != BOOST_ATOMIC_WCHAR_T_LOCK_FREE)

#define BOOST_ATOMIC_DETAIL_INT32_EXTENDED

#if !defined(BOOST_ATOMIC_DETAIL_INT64_EXTENDED)

template< bool Signed >
struct operations< 4u, Signed > :
    public extending_cas_based_operations< gcc_atomic_operations< 8u, Signed >, 4u, Signed >
{
};

#else // !defined(BOOST_ATOMIC_DETAIL_INT64_EXTENDED)

template< bool Signed >
struct operations< 4u, Signed > :
    public extending_cas_based_operations< gcc_atomic_operations< 16u, Signed >, 4u, Signed >
{
};

#endif // !defined(BOOST_ATOMIC_DETAIL_INT64_EXTENDED)

#else

template< bool Signed >
struct operations< 4u, Signed > :
    public gcc_atomic_operations< 4u, Signed >
{
};

#endif
#endif

#if BOOST_ATOMIC_INT16_LOCK_FREE > 0
#if (BOOST_ATOMIC_DETAIL_SIZEOF_LLONG == 2 && __GCC_ATOMIC_LLONG_LOCK_FREE != BOOST_ATOMIC_LLONG_LOCK_FREE) ||\
    (BOOST_ATOMIC_DETAIL_SIZEOF_LONG == 2 && __GCC_ATOMIC_LONG_LOCK_FREE != BOOST_ATOMIC_LONG_LOCK_FREE) ||\
    (BOOST_ATOMIC_DETAIL_SIZEOF_INT == 2 && __GCC_ATOMIC_INT_LOCK_FREE != BOOST_ATOMIC_INT_LOCK_FREE) ||\
    (BOOST_ATOMIC_DETAIL_SIZEOF_SHORT == 2 && __GCC_ATOMIC_SHORT_LOCK_FREE != BOOST_ATOMIC_SHORT_LOCK_FREE) ||\
    (BOOST_ATOMIC_DETAIL_SIZEOF_WCHAR_T == 2 && __GCC_ATOMIC_WCHAR_T_LOCK_FREE != BOOST_ATOMIC_WCHAR_T_LOCK_FREE)

#define BOOST_ATOMIC_DETAIL_INT16_EXTENDED

#if !defined(BOOST_ATOMIC_DETAIL_INT32_EXTENDED)

template< bool Signed >
struct operations< 2u, Signed > :
    public extending_cas_based_operations< gcc_atomic_operations< 4u, Signed >, 2u, Signed >
{
};

#elif !defined(BOOST_ATOMIC_DETAIL_INT64_EXTENDED)

template< bool Signed >
struct operations< 2u, Signed > :
    public extending_cas_based_operations< gcc_atomic_operations< 8u, Signed >, 2u, Signed >
{
};

#else

template< bool Signed >
struct operations< 2u, Signed > :
    public extending_cas_based_operations< gcc_atomic_operations< 16u, Signed >, 2u, Signed >
{
};

#endif

#else

template< bool Signed >
struct operations< 2u, Signed > :
    public gcc_atomic_operations< 2u, Signed >
{
};

#endif
#endif

#if BOOST_ATOMIC_INT8_LOCK_FREE > 0
#if (BOOST_ATOMIC_DETAIL_SIZEOF_LLONG == 1 && __GCC_ATOMIC_LLONG_LOCK_FREE != BOOST_ATOMIC_LLONG_LOCK_FREE) ||\
    (BOOST_ATOMIC_DETAIL_SIZEOF_LONG == 1 && __GCC_ATOMIC_LONG_LOCK_FREE != BOOST_ATOMIC_LONG_LOCK_FREE) ||\
    (BOOST_ATOMIC_DETAIL_SIZEOF_INT == 1 && __GCC_ATOMIC_INT_LOCK_FREE != BOOST_ATOMIC_INT_LOCK_FREE) ||\
    (BOOST_ATOMIC_DETAIL_SIZEOF_SHORT == 1 && __GCC_ATOMIC_SHORT_LOCK_FREE != BOOST_ATOMIC_SHORT_LOCK_FREE) ||\
    (BOOST_ATOMIC_DETAIL_SIZEOF_WCHAR_T == 1 && __GCC_ATOMIC_WCHAR_T_LOCK_FREE != BOOST_ATOMIC_WCHAR_T_LOCK_FREE) ||\
    (__GCC_ATOMIC_CHAR_LOCK_FREE != BOOST_ATOMIC_CHAR_LOCK_FREE) ||\
    (__GCC_ATOMIC_BOOL_LOCK_FREE != BOOST_ATOMIC_BOOL_LOCK_FREE)

#if !defined(BOOST_ATOMIC_DETAIL_INT16_EXTENDED)

template< bool Signed >
struct operations< 1u, Signed > :
    public extending_cas_based_operations< gcc_atomic_operations< 2u, Signed >, 1u, Signed >
{
};

#elif !defined(BOOST_ATOMIC_DETAIL_INT32_EXTENDED)

template< bool Signed >
struct operations< 1u, Signed > :
    public extending_cas_based_operations< gcc_atomic_operations< 4u, Signed >, 1u, Signed >
{
};

#elif !defined(BOOST_ATOMIC_DETAIL_INT64_EXTENDED)

template< bool Signed >
struct operations< 1u, Signed > :
    public extending_cas_based_operations< gcc_atomic_operations< 8u, Signed >, 1u, Signed >
{
};

#else

template< bool Signed >
struct operations< 1u, Signed > :
    public extending_cas_based_operations< gcc_atomic_operations< 16u, Signed >, 1u, Signed >
{
};

#endif

#else

template< bool Signed >
struct operations< 1u, Signed > :
    public gcc_atomic_operations< 1u, Signed >
{
};

#endif
#endif

#undef BOOST_ATOMIC_DETAIL_INT16_EXTENDED
#undef BOOST_ATOMIC_DETAIL_INT32_EXTENDED
#undef BOOST_ATOMIC_DETAIL_INT64_EXTENDED

BOOST_FORCEINLINE void thread_fence(memory_order order) BOOST_NOEXCEPT
{
    __atomic_thread_fence(atomics::detail::convert_memory_order_to_gcc(order));
}

BOOST_FORCEINLINE void signal_fence(memory_order order) BOOST_NOEXCEPT
{
    __atomic_signal_fence(atomics::detail::convert_memory_order_to_gcc(order));
}

} // namespace detail
} // namespace atomics
} // namespace lslboost

#endif // BOOST_ATOMIC_DETAIL_OPS_GCC_ATOMIC_HPP_INCLUDED_