atomic.h

Go to the documentation of this file.

/*
    This file is part of Mitsuba, a physically based rendering system.

    Copyright (c) 2007-2014 by Wenzel Jakob and others.

    Mitsuba is free software; you can redistribute it and/or modify
    it under the terms of the GNU General Public License Version 3
    as published by the Free Software Foundation.

    Mitsuba is distributed in the hope that it will be useful,
    but WITHOUT ANY WARRANTY; without even the implied warranty of
    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
    GNU General Public License for more details.

    You should have received a copy of the GNU General Public License
    along with this program. If not, see <http://www.gnu.org/licenses/>.
*/

#pragma once
#if !defined(__MITSUBA_CORE_ATOMIC_H_)
#define __MITSUBA_CORE_ATOMIC_H_

#include <mitsuba/mitsuba.h>

#if defined(__OSX__)
#include <libkern/OSAtomic.h>
#elif defined(_MSC_VER)
#include <intrin.h>
#endif

MTS_NAMESPACE_BEGIN

/**
 * The following implementations are based on PBRT
 *
 * \addtogroup libcore
 */

/*! @{ */

/**
 * \brief Atomically attempt to exchange a pointer with another value
 *
 * \param v Pointer to the pointer in question
 * \param oldValue Last known value of the destination \a v
 * \param newValue Replacement value for the destination \a v
 * \tparam T Base type of the pointer
 * \return \c true if \c *v was equal to \c oldValue and the exchange
 *         was successful.
 */
template <typename T> inline bool atomicCompareAndExchangePtr(T **v, T *newValue, T *oldValue) {
#if defined(_MSC_VER)
  #if defined(WIN64)
    return _InterlockedCompareExchangePointer(
                reinterpret_cast<void * volatile *>(v), newValue, oldValue) == oldValue;
  #else
    return _InterlockedCompareExchange(
                reinterpret_cast<long volatile *>(v),
                reinterpret_cast<long>(newValue), reinterpret_cast<long>(oldValue)) == reinterpret_cast<long>(oldValue);
  #endif
#else
  #if !defined(__clang__) && !defined(__INTEL_COMPILER)
        return __sync_bool_compare_and_swap(v, oldValue, newValue);
  #else
  /* Use the following workaround for clang and icl (Linux/Mac OS) */
  #if __SIZEOF_POINTER__ == 8 || defined(__LP64__)
        return __sync_bool_compare_and_swap(
                reinterpret_cast<long long volatile *>(v), reinterpret_cast<long long>(oldValue),
                reinterpret_cast<long long>(newValue));
  #else
        return __sync_bool_compare_and_swap(
                reinterpret_cast<long volatile *>(v), reinterpret_cast<long>(oldValue),
                reinterpret_cast<long>(newValue));
  #endif
  #endif
#endif
}

/**
 * \brief Atomically attempt to exchange a 32-bit integer with another value
 *
 * \param v Pointer to the memory region in question
 * \param oldValue Last known value of the destination \a v
 * \param newValue Replacement value for the destination \a v
 * \return \c true if \c *v was equal to \c oldValue and the exchange
 *         was successful.
 */

inline bool atomicCompareAndExchange(volatile int32_t *v, int32_t newValue, int32_t oldValue) {
#if defined(_MSC_VER)
    return _InterlockedCompareExchange(
                reinterpret_cast<volatile long *>(v), newValue, oldValue) == oldValue;
#else
        return __sync_bool_compare_and_swap(v, oldValue, newValue);
#endif
}

/**
 * \brief Atomically attempt to exchange a 64-bit integer with another value
 *
 * \param v Pointer to the memory region in question
 * \param oldValue Last known value of the destination \a v
 * \param newValue Replacement value for the destination \a v
 * \return \c true if \c *v was equal to \c oldValue and the exchange
 *         was successful.
 */

inline bool atomicCompareAndExchange(volatile int64_t *v, int64_t newValue, int64_t oldValue) {
#if defined(_MSC_VER)
    return _InterlockedCompareExchange64(
                reinterpret_cast<volatile __int64 *>(v), newValue, oldValue) == oldValue;
#else
        return __sync_bool_compare_and_swap(v, oldValue, newValue);
#endif
}

/**
 * \brief Atomically add \a delta to the floating point destination \a dst
 *
 * \return The final value written to \a dst
 */
inline float atomicAdd(volatile float *dst, float delta) {
        /* Atomic FP addition from PBRT */
        union bits { float f; int32_t i; };
        bits oldVal, newVal;
        do {
        // On IA32/x64, adding a PAUSE instruction in compare/exchange loops
        // is recommended to improve performance.  (And it does!)
#if (defined(__i386__) || defined(__amd64__))
        __asm__ __volatile__ ("pause\n");
#endif
                oldVal.f = *dst;
                newVal.f = oldVal.f + delta;
        } while (!atomicCompareAndExchange((volatile int32_t *) dst, newVal.i, oldVal.i));
        return newVal.f;
}

/**
 * \brief Atomically add \a delta to the floating point destination \a dst
 *
 * \return The final value written to \a dst
 */
inline double atomicAdd(volatile double *dst, double delta) {
        /* Atomic FP addition from PBRT */
        union bits { double f; int64_t i; };
        bits oldVal, newVal;
        do {
        // On IA64/x64, adding a PAUSE instruction in compare/exchange loops
        // is recommended to improve performance.  (And it does!)
#if (defined(__i386__) || defined(__amd64__))
        __asm__ __volatile__ ("pause\n");
#endif
                oldVal.f = *dst;
                newVal.f = oldVal.f + delta;
        } while (!atomicCompareAndExchange((volatile int64_t *) dst, newVal.i, oldVal.i));
        return newVal.f;
}

/**
 * \brief Atomically add \a delta to the 32-bit integer destination \a dst
 *
 * \return The final value written to \a dst
 */

inline int32_t atomicAdd(volatile int32_t *dst, int32_t delta) {
#if defined(_MSC_VER)
        return _InterlockedExchangeAdd(reinterpret_cast<volatile long *>(dst), delta) + delta;
#else
        return __sync_add_and_fetch(dst, delta);
#endif
}

/**
 * \brief Atomically add \a delta to the 64-bit integer destination \a dst
 *
 * \return The final value written to \a dst
 */

inline int64_t atomicAdd(volatile int64_t *dst, int64_t delta) {
#if defined(_MSC_VER)
  #if defined(_WIN64)
        return _InterlockedExchangeAdd64(reinterpret_cast<volatile __int64 *>(dst), delta) + delta;
  #else
        SLog(EError, "atomicAdd() cannot handle 64-bit integers on WIN32");
        return 0;
  #endif
#else
        return __sync_add_and_fetch(dst, delta);
#endif
}

/**
 * \brief Atomically set \c dst to the maximum of itself and \c value
 * \return The maximum value now stored in \c dst
 */

inline int64_t atomicMaximum(volatile int64_t *dst, int64_t value) {
        int64_t current;
        do {
                current = *dst;
                if (value <= current)
                        return current;
                #if (defined(__i386__) || defined(__amd64__))
                        __asm__ __volatile__ ("pause\n");
                #endif
        } while (!atomicCompareAndExchange(dst, value, current));

        return value;
}

/*
 * \brief Atomically set \c dst to the maximum of itself and \c value
 * \return The maximum value now stored in \c dst
 */

inline int32_t atomicMaximum(volatile int32_t *dst, int32_t value) {
        int32_t current;
        do {
                current = *dst;
                if (value <= current)
                        return current;
                #if (defined(__i386__) || defined(__amd32__))
                        __asm__ __volatile__ ("pause\n");
                #endif
        } while (!atomicCompareAndExchange(dst, value, current));

        return value;
}


/*! }@ */

MTS_NAMESPACE_END

#endif /* __MITSUBA_CORE_ATOMIC_H_ */