/work/x265/source/common/threading.h

Source
/*****************************************************************************
 * Copyright (C) 2013-2020 MulticoreWare, Inc
 *
 * Authors: Steve Borho <steve@borho.org>
 *          Min Chen <chenm003@163.com>
            liwei <liwei@multicorewareinc.com>
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation; either version 2 of the License, or
 * (at your option) any later version.
 *
 * This program 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, write to the Free Software
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02111, USA.
 *
 * This program is also available under a commercial proprietary license.
 * For more information, contact us at license @ x265.com
 *****************************************************************************/

#ifndef X265_THREADING_H
#define X265_THREADING_H

#include "common.h"
#include "x265.h"

#ifdef _WIN32
#include <windows.h>
#include "winxp.h"  // XP workarounds for CONDITION_VARIABLE and ATOMIC_OR
#else
#include <pthread.h>
#include <semaphore.h>
#include <errno.h>
#include <fcntl.h>
#endif

#if MACOS
#include <sys/param.h>
#include <sys/sysctl.h>
#endif

#if NO_ATOMICS

#include <sys/time.h>
#include <unistd.h>

namespace X265_NS {
// x265 private namespace
int no_atomic_or(int* ptr, int mask);
int no_atomic_and(int* ptr, int mask);
int no_atomic_inc(int* ptr);
int no_atomic_dec(int* ptr);
int no_atomic_add(int* ptr, int val);
int64_t no_atomic_add64(int64_t* ptr, int64_t val);
}

#define BSR(id, x)            (id) = ((unsigned long)__builtin_clz(x) ^ 31)
#define BSF(id, x)            (id) = ((unsigned long)__builtin_ctz(x))
#define BSR64(id, x)          (id) = ((unsigned long)__builtin_clzll(x) ^ 63)
#define BSF64(id, x)          (id) = ((unsigned long)__builtin_ctzll(x))
#define ATOMIC_OR(ptr, mask)  no_atomic_or((int*)ptr, mask)
#define ATOMIC_AND(ptr, mask) no_atomic_and((int*)ptr, mask)
#define ATOMIC_INC(ptr)       no_atomic_inc((int*)ptr)
#define ATOMIC_DEC(ptr)       no_atomic_dec((int*)ptr)
#define ATOMIC_ADD(ptr, val)  (sizeof(*(ptr)) == 8 ? \
                               no_atomic_add64((int64_t*)ptr, (int64_t)(val)) : \
                               no_atomic_add((int*)ptr, (int)(val)))
#define GIVE_UP_TIME()        usleep(0)

#elif __GNUC__               /* GCCs builtin atomics */

#include <sys/time.h>
#include <unistd.h>

#define BSR(id, x)            (id) = ((unsigned long)__builtin_clz(x) ^ 31)
#define BSF(id, x)            (id) = ((unsigned long)__builtin_ctz(x))
#define BSR64(id, x)          (id) = ((unsigned long)__builtin_clzll(x) ^ 63)
#define BSF64(id, x)          (id) = ((unsigned long)__builtin_ctzll(x))
#define ATOMIC_OR(ptr, mask)  __sync_fetch_and_or(ptr, mask)
#define ATOMIC_AND(ptr, mask) __sync_fetch_and_and(ptr, mask)
#define ATOMIC_INC(ptr)       __sync_add_and_fetch((volatile int32_t*)ptr, 1)
#define ATOMIC_DEC(ptr)       __sync_add_and_fetch((volatile int32_t*)ptr, -1)
#define ATOMIC_ADD(ptr, val)  __sync_fetch_and_add((volatile __typeof__(*(ptr))*)ptr, (__typeof__(*(ptr) + 0))(val))
#define GIVE_UP_TIME()        usleep(0)

#elif defined(_MSC_VER)       /* Windows atomic intrinsics */

#include <intrin.h>

#define BSR(id, x)            _BitScanReverse(&id, x)
#define BSF(id, x)            _BitScanForward(&id, x)
#define BSR64(id, x)          _BitScanReverse64(&id, x)
#define BSF64(id, x)          _BitScanForward64(&id, x)
#define ATOMIC_INC(ptr)       InterlockedIncrement((volatile LONG*)ptr)
#define ATOMIC_DEC(ptr)       InterlockedDecrement((volatile LONG*)ptr)
#define ATOMIC_ADD(ptr, val)  (sizeof(*(ptr)) == 8 ? \
                               InterlockedExchangeAdd64((volatile LONGLONG*)ptr, (LONGLONG)(val)) : \
                               InterlockedExchangeAdd((volatile LONG*)ptr, (LONG)(val)))
#define ATOMIC_OR(ptr, mask)  _InterlockedOr((volatile LONG*)ptr, (LONG)mask)
#define ATOMIC_AND(ptr, mask) _InterlockedAnd((volatile LONG*)ptr, (LONG)mask)
#define GIVE_UP_TIME()        Sleep(0)

#endif // ifdef __GNUC__

namespace X265_NS {
// x265 private namespace

#ifdef _WIN32

typedef HANDLE ThreadHandle;

class Lock
{
public:

    Lock()
    {
        InitializeCriticalSection(&this->handle);
    }

    ~Lock()
    {
        DeleteCriticalSection(&this->handle);
    }

    void acquire()
    {
        EnterCriticalSection(&this->handle);
    }

    void release()
    {
        LeaveCriticalSection(&this->handle);
    }

protected:

    CRITICAL_SECTION handle;
};

class Event
{
public:

    Event()
    {
        this->handle = CreateEvent(NULL, FALSE, FALSE, NULL);
    }

    ~Event()
    {
        CloseHandle(this->handle);
    }

    void wait()
    {
        WaitForSingleObject(this->handle, INFINITE);
    }

    bool timedWait(uint32_t milliseconds)
    {
        /* returns true if the wait timed out */
        return WaitForSingleObject(this->handle, milliseconds) == WAIT_TIMEOUT;
    }

    void trigger()
    {
        SetEvent(this->handle);
    }

protected:

    HANDLE handle;
};

/* This class is intended for use in signaling state changes safely between CPU
 * cores. One thread should be a writer and multiple threads may be readers. The
 * mutex's main purpose is to serve as a memory fence to ensure writes made by
 * the writer thread are visible prior to readers seeing the m_val change. Its
 * secondary purpose is for use with the condition variable for blocking waits */
class ThreadSafeInteger
{
public:

    ThreadSafeInteger()
    {
        m_val = 0;
        InitializeCriticalSection(&m_cs);
        InitializeConditionVariable(&m_cv);
    }

    ~ThreadSafeInteger()
    {
        DeleteCriticalSection(&m_cs);
        XP_CONDITION_VAR_FREE(&m_cv);
    }

    int waitForChange(int prev)
    {
        EnterCriticalSection(&m_cs);
        if (m_val == prev)
            SleepConditionVariableCS(&m_cv, &m_cs, INFINITE);
        LeaveCriticalSection(&m_cs);
        return m_val;
    }

    int get()
    {
        EnterCriticalSection(&m_cs);
        int ret = m_val;
        LeaveCriticalSection(&m_cs);
        return ret;
    }

    int getIncr(int n = 1)
    {
        EnterCriticalSection(&m_cs);
        int ret = m_val;
        m_val += n;
        LeaveCriticalSection(&m_cs);
        return ret;
    }

    void set(int newval)
    {
        EnterCriticalSection(&m_cs);
        m_val = newval;
        WakeAllConditionVariable(&m_cv);
        LeaveCriticalSection(&m_cs);
    }

    void poke(void)
    {
        /* awaken all waiting threads, but make no change */
        EnterCriticalSection(&m_cs);
        WakeAllConditionVariable(&m_cv);
        LeaveCriticalSection(&m_cs);
    }

    void incr()
    {
        EnterCriticalSection(&m_cs);
        m_val++;
        WakeAllConditionVariable(&m_cv);
        LeaveCriticalSection(&m_cs);
    }

    void decr()
    {
        EnterCriticalSection(&m_cs);
        m_val--;
        WakeAllConditionVariable(&m_cv);
        LeaveCriticalSection(&m_cs);
    }

protected:

    CRITICAL_SECTION   m_cs;
    CONDITION_VARIABLE m_cv;
    int                m_val;
};

class NamedSemaphore
{
public:
    NamedSemaphore() : m_sem(NULL)
    {
    }

    ~NamedSemaphore()
    {
    }

    bool create(const char* name, const int initcnt, const int maxcnt)
    {
        if(!m_sem)
        {
            m_sem = CreateSemaphoreA(NULL, initcnt, maxcnt, name);
        }
        return m_sem != NULL;
    }

    bool give(const int32_t cnt)
    {
        return ReleaseSemaphore(m_sem, (LONG)cnt, NULL) != FALSE;
    }

    bool take(const uint32_t time_out = INFINITE)
    {
        int32_t rt = WaitForSingleObject(m_sem, time_out);
        return rt != WAIT_TIMEOUT && rt != WAIT_FAILED;
    }

    void release()
    {
        CloseHandle(m_sem);
        m_sem = NULL;
    }

private:
    HANDLE m_sem;
};

#else /* POSIX / pthreads */

typedef pthread_t ThreadHandle;

class Lock
{
public:

    Lock()
    {
        pthread_mutex_init(&this->handle, NULL);
    }

    ~Lock()
    {
        pthread_mutex_destroy(&this->handle);
    }

    void acquire()
    {
        pthread_mutex_lock(&this->handle);
    }

    void release()
    {
        pthread_mutex_unlock(&this->handle);
    }

protected:

    pthread_mutex_t handle;
};

class Event
{
public:

    Event()
    {
        m_counter = 0;
        if (pthread_mutex_init(&m_mutex, NULL) ||
            pthread_cond_init(&m_cond, NULL))
        {
            x265_log(NULL, X265_LOG_ERROR, "fatal: unable to initialize conditional variable\n");
        }
    }

    ~Event()
    {
        pthread_cond_destroy(&m_cond);
        pthread_mutex_destroy(&m_mutex);
    }

    void wait()
    {
        pthread_mutex_lock(&m_mutex);

        /* blocking wait on conditional variable, mutex is atomically released
         * while blocked. When condition is signaled, mutex is re-acquired */
        while (!m_counter)
            pthread_cond_wait(&m_cond, &m_mutex);

        m_counter--;
        pthread_mutex_unlock(&m_mutex);
    }

    bool timedWait(uint32_t waitms)
    {
        bool bTimedOut = false;

        pthread_mutex_lock(&m_mutex);
        if (!m_counter)
        {
            struct timeval tv;
            struct timespec ts;
            gettimeofday(&tv, NULL);
            /* convert current time from (sec, usec) to (sec, nsec) */
            ts.tv_sec = tv.tv_sec;
            ts.tv_nsec = tv.tv_usec * 1000;

            ts.tv_nsec += 1000 * 1000 * (waitms % 1000);    /* add ms to tv_nsec */
            ts.tv_sec += ts.tv_nsec / (1000 * 1000 * 1000); /* overflow tv_nsec */
            ts.tv_nsec %= (1000 * 1000 * 1000);             /* clamp tv_nsec */
            ts.tv_sec += waitms / 1000;                     /* add seconds */

            /* blocking wait on conditional variable, mutex is atomically released
             * while blocked. When condition is signaled, mutex is re-acquired.
             * ts is absolute time to stop waiting */
            bTimedOut = pthread_cond_timedwait(&m_cond, &m_mutex, &ts) == ETIMEDOUT;
        }
        if (m_counter > 0)
        {
            m_counter--;
            bTimedOut = false;
        }
        pthread_mutex_unlock(&m_mutex);
        return bTimedOut;
    }

    void trigger()
    {
        pthread_mutex_lock(&m_mutex);
        if (m_counter < UINT_MAX)
            m_counter++;
        /* Signal a single blocking thread */
        pthread_cond_signal(&m_cond);
        pthread_mutex_unlock(&m_mutex);
    }

protected:

    pthread_mutex_t m_mutex;
    pthread_cond_t  m_cond;
    uint32_t        m_counter;
};

/* This class is intended for use in signaling state changes safely between CPU
 * cores. One thread should be a writer and multiple threads may be readers. The
 * mutex's main purpose is to serve as a memory fence to ensure writes made by
 * the writer thread are visible prior to readers seeing the m_val change. Its
 * secondary purpose is for use with the condition variable for blocking waits */
class ThreadSafeInteger
{
public:

    ThreadSafeInteger()
    {
        m_val = 0;
        if (pthread_mutex_init(&m_mutex, NULL) ||
            pthread_cond_init(&m_cond, NULL))
        {
            x265_log(NULL, X265_LOG_ERROR, "fatal: unable to initialize conditional variable\n");
        }
    }

    ~ThreadSafeInteger()
    {
        pthread_cond_destroy(&m_cond);
        pthread_mutex_destroy(&m_mutex);
    }

    int waitForChange(int prev)
    {
        pthread_mutex_lock(&m_mutex);
        if (m_val == prev)
            pthread_cond_wait(&m_cond, &m_mutex);
        pthread_mutex_unlock(&m_mutex);
        return m_val;
    }

    int get()
    {
        pthread_mutex_lock(&m_mutex);
        int ret = m_val;
        pthread_mutex_unlock(&m_mutex);
        return ret;
    }

    int getIncr(int n = 1)
    {
        pthread_mutex_lock(&m_mutex);
        int ret = m_val;
        m_val += n;
        pthread_mutex_unlock(&m_mutex);
        return ret;
    }

    void set(int newval)
    {
        pthread_mutex_lock(&m_mutex);
        m_val = newval;
        pthread_cond_broadcast(&m_cond);
        pthread_mutex_unlock(&m_mutex);
    }

    void poke(void)
    {
        /* awaken all waiting threads, but make no change */
        pthread_mutex_lock(&m_mutex);
        pthread_cond_broadcast(&m_cond);
        pthread_mutex_unlock(&m_mutex);
    }

    void incr()
    {
        pthread_mutex_lock(&m_mutex);
        m_val++;
        pthread_cond_broadcast(&m_cond);
        pthread_mutex_unlock(&m_mutex);
    }

    void decr()
    {
        pthread_mutex_lock(&m_mutex);
        m_val--;
        pthread_cond_broadcast(&m_cond);
        pthread_mutex_unlock(&m_mutex);
    }

protected:

    pthread_mutex_t m_mutex;
    pthread_cond_t  m_cond;
    int             m_val;
};

#define TIMEOUT_INFINITE 0xFFFFFFFF

class NamedSemaphore
{
public:
    NamedSemaphore() 
        : m_sem(NULL)
#ifndef __APPLE__
        , m_name(NULL)
#endif //__APPLE__
    {
    }

    ~NamedSemaphore()
    {
    }

    bool create(const char* name, const int initcnt, const int maxcnt)
    {
        bool ret = false;

        if (initcnt >= maxcnt)
        {
            return false;
        }

#ifdef __APPLE__
        do
        {
            int32_t pshared = name != NULL ? PTHREAD_PROCESS_SHARED : PTHREAD_PROCESS_PRIVATE;

            m_sem = (mac_sem_t *)malloc(sizeof(mac_sem_t));
            if (!m_sem)
            {
                break;
            }

            if (pthread_mutexattr_init(&m_sem->mutexAttr))
            {
                break;
            }

            if (pthread_mutexattr_setpshared(&m_sem->mutexAttr, pshared))
            {
                break;
            }

            if (pthread_condattr_init(&m_sem->condAttr))
            {
                break;
            }

            if (pthread_condattr_setpshared(&m_sem->condAttr, pshared))
            {
                break;
            }

            if (pthread_mutex_init(&m_sem->mutex, &m_sem->mutexAttr))
            {
                break;
            }

            if (pthread_cond_init(&m_sem->cond, &m_sem->condAttr))
            {
                break;
            }

            m_sem->curCnt = initcnt;
            m_sem->maxCnt = maxcnt;

            ret = true;
        } while (0);
        
        if (!ret)
        {
            release();
        }

#else  //__APPLE__
        m_sem = sem_open(name, O_CREAT | O_EXCL, 0666, initcnt);
        if (m_sem != SEM_FAILED) 
        {
            m_name = strdup(name);
            ret = true;
        }
        else 
        {
            if (EEXIST == errno) 
            {
                m_sem = sem_open(name, 0);
                if (m_sem != SEM_FAILED) 
                {
                    m_name = strdup(name);
                    ret = true;
                }
            }
        }
#endif //__APPLE__

        return ret;
    }

    bool give(const int32_t cnt)
    {
        if (!m_sem)
        {
            return false;
        }

#ifdef __APPLE__
        if (pthread_mutex_lock(&m_sem->mutex))
        {
            return false;
        }

        int oldCnt = m_sem->curCnt;
        m_sem->curCnt += cnt;
        if (m_sem->curCnt > m_sem->maxCnt)
        {
            m_sem->curCnt = m_sem->maxCnt;
        }

        bool ret = true;
        if (!oldCnt)
        {
            ret = 0 == pthread_cond_broadcast(&m_sem->cond);
        }

        if (pthread_mutex_unlock(&m_sem->mutex))
        {
            return false;
        }

        return ret;
#else //__APPLE__
        int ret = 0;
        int32_t curCnt = cnt;
        while (curCnt-- && !ret) {
            ret = sem_post(m_sem);
        }

        return 0 == ret;
#endif //_APPLE__
    }

    bool take(const uint32_t time_out = TIMEOUT_INFINITE)
    {
        if (!m_sem)
        {
            return false;
        }

#ifdef __APPLE__

        if (pthread_mutex_lock(&m_sem->mutex))
        {
            return false;
        }

        bool ret = true;
        if (TIMEOUT_INFINITE == time_out) 
        {
            if (!m_sem->curCnt)
            {
                if (pthread_cond_wait(&m_sem->cond, &m_sem->mutex))
                {
                    ret = false;
                } 
            }

            if (m_sem->curCnt && ret)
            {
                m_sem->curCnt--;
            }
        }
        else
        {
            if (0 == time_out)
            {
                if (m_sem->curCnt)
                {
                    m_sem->curCnt--;
                }
                else
                {
                    ret = false;
                }
            }
            else
            {
                if (!m_sem->curCnt)
                {
                    struct timespec ts;
                    ts.tv_sec = time_out / 1000L;
                    ts.tv_nsec = (time_out * 1000000L) - ts.tv_sec * 1000 * 1000 * 1000;

                    if (pthread_cond_timedwait(&m_sem->cond, &m_sem->mutex, &ts))
                    {
                        ret = false;
                    }
                }

                if (m_sem->curCnt && ret)
                {
                    m_sem->curCnt--;
                }
            }
        }

        if (pthread_mutex_unlock(&m_sem->mutex))
        {
            return false;
        }

        return ret;
#else //__APPLE__
        if (TIMEOUT_INFINITE == time_out) 
        {
            return 0 == sem_wait(m_sem);
        }
        else 
        {
            if (0 == time_out)
            {
                return 0 == sem_trywait(m_sem);
            }
            else
            {
                struct timespec ts;
                ts.tv_sec = time_out / 1000L;
                ts.tv_nsec = (time_out * 1000000L) - ts.tv_sec * 1000 * 1000 * 1000;
                return 0 == sem_timedwait(m_sem, &ts);
            }
        }
#endif //_APPLE__
    }

    void release()
    {
        if (m_sem)
        {
#ifdef __APPLE__
            pthread_condattr_destroy(&m_sem->condAttr);
            pthread_mutexattr_destroy(&m_sem->mutexAttr);
            pthread_mutex_destroy(&m_sem->mutex);
            pthread_cond_destroy(&m_sem->cond);
            free(m_sem);
            m_sem = NULL;
#else //__APPLE__
            sem_close(m_sem);
            sem_unlink(m_name);
            m_sem = NULL;
            free(m_name);
            m_name = NULL;
#endif //__APPLE__
        }
    }

private:
#ifdef __APPLE__
    typedef struct
    {
        pthread_mutex_t     mutex;
        pthread_cond_t      cond;
        pthread_mutexattr_t mutexAttr;
        pthread_condattr_t  condAttr;
        uint32_t            curCnt;
        uint32_t            maxCnt;
    }mac_sem_t;
    mac_sem_t *m_sem;
#else // __APPLE__
    sem_t *m_sem;
    char  *m_name;
#endif // __APPLE_
};

#endif // ifdef _WIN32

class ScopedLock
{
public:

    ScopedLock(Lock &instance) : inst(instance)
    {
        this->inst.acquire();
    }

    ~ScopedLock()
    {
        this->inst.release();
    }

protected:

    // do not allow assignments
    ScopedLock &operator =(const ScopedLock &);

    Lock &inst;
};

// Utility class which adds elapsed time of the scope of the object into the
// accumulator provided to the constructor
struct ScopedElapsedTime
{
    ScopedElapsedTime(int64_t& accum) : accumlatedTime(accum) { startTime = x265_mdate(); }

    ~ScopedElapsedTime() { accumlatedTime += x265_mdate() - startTime; }

protected:

    int64_t  startTime;
    int64_t& accumlatedTime;

    // do not allow assignments
    ScopedElapsedTime &operator =(const ScopedElapsedTime &);
};

//< Simplistic portable thread class.  Shutdown signalling left to derived class
class Thread
{
private:

    ThreadHandle thread;

public:

    Thread();

    virtual ~Thread();

    //< Derived class must implement ThreadMain.
    virtual void threadMain() = 0;

    //< Returns true if thread was successfully created
    bool start();

    void stop();
};
} // end namespace X265_NS

#endif // ifndef X265_THREADING_H

Coverage Report

Created: 2026-03-08 06:41

Line	Count	Source
1		/*****************************************************************************
2		* Copyright (C) 2013-2020 MulticoreWare, Inc
3		*
4		* Authors: Steve Borho <steve@borho.org>
5		* Min Chen <chenm003@163.com>
6		liwei <liwei@multicorewareinc.com>
7		*
8		* This program is free software; you can redistribute it and/or modify
9		* it under the terms of the GNU General Public License as published by
10		* the Free Software Foundation; either version 2 of the License, or
11		* (at your option) any later version.
12		*
13		* This program is distributed in the hope that it will be useful,
14		* but WITHOUT ANY WARRANTY; without even the implied warranty of
15		* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16		* GNU General Public License for more details.
17		*
18		* You should have received a copy of the GNU General Public License
19		* along with this program; if not, write to the Free Software
20		* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA.
21		*
22		* This program is also available under a commercial proprietary license.
23		* For more information, contact us at license @ x265.com
24		*****************************************************************************/
25
26		#ifndef X265_THREADING_H
27		#define X265_THREADING_H
28
29		#include "common.h"
30		#include "x265.h"
31
32		#ifdef _WIN32
33		#include <windows.h>
34		#include "winxp.h" // XP workarounds for CONDITION_VARIABLE and ATOMIC_OR
35		#else
36		#include <pthread.h>
37		#include <semaphore.h>
38		#include <errno.h>
39		#include <fcntl.h>
40		#endif
41
42		#if MACOS
43		#include <sys/param.h>
44		#include <sys/sysctl.h>
45		#endif
46
47		#if NO_ATOMICS
48
49		#include <sys/time.h>
50		#include <unistd.h>
51
52		namespace X265_NS {
53		// x265 private namespace
54		int no_atomic_or(int* ptr, int mask);
55		int no_atomic_and(int* ptr, int mask);
56		int no_atomic_inc(int* ptr);
57		int no_atomic_dec(int* ptr);
58		int no_atomic_add(int* ptr, int val);
59		int64_t no_atomic_add64(int64_t* ptr, int64_t val);
60		}
61
62		#define BSR(id, x) (id) = ((unsigned long)__builtin_clz(x) ^ 31)
63		#define BSF(id, x) (id) = ((unsigned long)__builtin_ctz(x))
64		#define BSR64(id, x) (id) = ((unsigned long)__builtin_clzll(x) ^ 63)
65		#define BSF64(id, x) (id) = ((unsigned long)__builtin_ctzll(x))
66		#define ATOMIC_OR(ptr, mask) no_atomic_or((int*)ptr, mask)
67		#define ATOMIC_AND(ptr, mask) no_atomic_and((int*)ptr, mask)
68		#define ATOMIC_INC(ptr) no_atomic_inc((int*)ptr)
69		#define ATOMIC_DEC(ptr) no_atomic_dec((int*)ptr)
70		#define ATOMIC_ADD(ptr, val) (sizeof(*(ptr)) == 8 ? \
71		no_atomic_add64((int64_t*)ptr, (int64_t)(val)) : \
72		no_atomic_add((int*)ptr, (int)(val)))
73		#define GIVE_UP_TIME() usleep(0)
74
75		#elif __GNUC__ /* GCCs builtin atomics */
76
77		#include <sys/time.h>
78		#include <unistd.h>
79
80	2.28M	#define BSR(id, x) (id) = ((unsigned long)__builtin_clz(x) ^ 31)
81	166k	#define BSF(id, x) (id) = ((unsigned long)__builtin_ctz(x))
82		#define BSR64(id, x) (id) = ((unsigned long)__builtin_clzll(x) ^ 63)
83		#define BSF64(id, x) (id) = ((unsigned long)__builtin_ctzll(x))
84	12.1k	#define ATOMIC_OR(ptr, mask) __sync_fetch_and_or(ptr, mask)
85	6.42k	#define ATOMIC_AND(ptr, mask) __sync_fetch_and_and(ptr, mask)
86	28.0k	#define ATOMIC_INC(ptr) __sync_add_and_fetch((volatile int32_t*)ptr, 1)
87	7.07k	#define ATOMIC_DEC(ptr) __sync_add_and_fetch((volatile int32_t*)ptr, -1)
88	13.7k	#define ATOMIC_ADD(ptr, val) __sync_fetch_and_add((volatile __typeof__((ptr)))ptr, (__typeof__(*(ptr) + 0))(val))
89	0	#define GIVE_UP_TIME() usleep(0)
90
91		#elif defined(_MSC_VER) /* Windows atomic intrinsics */
92
93		#include <intrin.h>
94
95		#define BSR(id, x) _BitScanReverse(&id, x)
96		#define BSF(id, x) _BitScanForward(&id, x)
97		#define BSR64(id, x) _BitScanReverse64(&id, x)
98		#define BSF64(id, x) _BitScanForward64(&id, x)
99		#define ATOMIC_INC(ptr) InterlockedIncrement((volatile LONG*)ptr)
100		#define ATOMIC_DEC(ptr) InterlockedDecrement((volatile LONG*)ptr)
101		#define ATOMIC_ADD(ptr, val) (sizeof(*(ptr)) == 8 ? \
102		InterlockedExchangeAdd64((volatile LONGLONG*)ptr, (LONGLONG)(val)) : \
103		InterlockedExchangeAdd((volatile LONG*)ptr, (LONG)(val)))
104		#define ATOMIC_OR(ptr, mask) _InterlockedOr((volatile LONG*)ptr, (LONG)mask)
105		#define ATOMIC_AND(ptr, mask) _InterlockedAnd((volatile LONG*)ptr, (LONG)mask)
106		#define GIVE_UP_TIME() Sleep(0)
107
108		#endif // ifdef __GNUC__
109
110		namespace X265_NS {
111		// x265 private namespace
112
113		#ifdef _WIN32
114
115		typedef HANDLE ThreadHandle;
116
117		class Lock
118		{
119		public:
120
121		Lock()
122		{
123		InitializeCriticalSection(&this->handle);
124		}
125
126		~Lock()
127		{
128		DeleteCriticalSection(&this->handle);
129		}
130
131		void acquire()
132		{
133		EnterCriticalSection(&this->handle);
134		}
135
136		void release()
137		{
138		LeaveCriticalSection(&this->handle);
139		}
140
141		protected:
142
143		CRITICAL_SECTION handle;
144		};
145
146		class Event
147		{
148		public:
149
150		Event()
151		{
152		this->handle = CreateEvent(NULL, FALSE, FALSE, NULL);
153		}
154
155		~Event()
156		{
157		CloseHandle(this->handle);
158		}
159
160		void wait()
161		{
162		WaitForSingleObject(this->handle, INFINITE);
163		}
164
165		bool timedWait(uint32_t milliseconds)
166		{
167		/* returns true if the wait timed out */
168		return WaitForSingleObject(this->handle, milliseconds) == WAIT_TIMEOUT;
169		}
170
171		void trigger()
172		{
173		SetEvent(this->handle);
174		}
175
176		protected:
177
178		HANDLE handle;
179		};
180
181		/* This class is intended for use in signaling state changes safely between CPU
182		* cores. One thread should be a writer and multiple threads may be readers. The
183		* mutex's main purpose is to serve as a memory fence to ensure writes made by
184		* the writer thread are visible prior to readers seeing the m_val change. Its
185		* secondary purpose is for use with the condition variable for blocking waits */
186		class ThreadSafeInteger
187		{
188		public:
189
190		ThreadSafeInteger()
191		{
192		m_val = 0;
193		InitializeCriticalSection(&m_cs);
194		InitializeConditionVariable(&m_cv);
195		}
196
197		~ThreadSafeInteger()
198		{
199		DeleteCriticalSection(&m_cs);
200		XP_CONDITION_VAR_FREE(&m_cv);
201		}
202
203		int waitForChange(int prev)
204		{
205		EnterCriticalSection(&m_cs);
206		if (m_val == prev)
207		SleepConditionVariableCS(&m_cv, &m_cs, INFINITE);
208		LeaveCriticalSection(&m_cs);
209		return m_val;
210		}
211
212		int get()
213		{
214		EnterCriticalSection(&m_cs);
215		int ret = m_val;
216		LeaveCriticalSection(&m_cs);
217		return ret;
218		}
219
220		int getIncr(int n = 1)
221		{
222		EnterCriticalSection(&m_cs);
223		int ret = m_val;
224		m_val += n;
225		LeaveCriticalSection(&m_cs);
226		return ret;
227		}
228
229		void set(int newval)
230		{
231		EnterCriticalSection(&m_cs);
232		m_val = newval;
233		WakeAllConditionVariable(&m_cv);
234		LeaveCriticalSection(&m_cs);
235		}
236
237		void poke(void)
238		{
239		/* awaken all waiting threads, but make no change */
240		EnterCriticalSection(&m_cs);
241		WakeAllConditionVariable(&m_cv);
242		LeaveCriticalSection(&m_cs);
243		}
244
245		void incr()
246		{
247		EnterCriticalSection(&m_cs);
248		m_val++;
249		WakeAllConditionVariable(&m_cv);
250		LeaveCriticalSection(&m_cs);
251		}
252
253		void decr()
254		{
255		EnterCriticalSection(&m_cs);
256		m_val--;
257		WakeAllConditionVariable(&m_cv);
258		LeaveCriticalSection(&m_cs);
259		}
260
261		protected:
262
263		CRITICAL_SECTION m_cs;
264		CONDITION_VARIABLE m_cv;
265		int m_val;
266		};
267
268		class NamedSemaphore
269		{
270		public:
271		NamedSemaphore() : m_sem(NULL)
272		{
273		}
274
275		~NamedSemaphore()
276		{
277		}
278
279		bool create(const char* name, const int initcnt, const int maxcnt)
280		{
281		if(!m_sem)
282		{
283		m_sem = CreateSemaphoreA(NULL, initcnt, maxcnt, name);
284		}
285		return m_sem != NULL;
286		}
287
288		bool give(const int32_t cnt)
289		{
290		return ReleaseSemaphore(m_sem, (LONG)cnt, NULL) != FALSE;
291		}
292
293		bool take(const uint32_t time_out = INFINITE)
294		{
295		int32_t rt = WaitForSingleObject(m_sem, time_out);
296		return rt != WAIT_TIMEOUT && rt != WAIT_FAILED;
297		}
298
299		void release()
300		{
301		CloseHandle(m_sem);
302		m_sem = NULL;
303		}
304
305		private:
306		HANDLE m_sem;
307		};
308
309		#else /* POSIX / pthreads */
310
311		typedef pthread_t ThreadHandle;
312
313		class Lock
314		{
315		public:
316
317		Lock()
318	90.3k	{
319	90.3k	pthread_mutex_init(&this->handle, NULL);
320	90.3k	}
321
322		~Lock()
323	90.3k	{
324	90.3k	pthread_mutex_destroy(&this->handle);
325	90.3k	}
326
327		void acquire()
328	200k	{
329	200k	pthread_mutex_lock(&this->handle);
330	200k	}
331
332		void release()
333	200k	{
334	200k	pthread_mutex_unlock(&this->handle);
335	200k	}
336
337		protected:
338
339		pthread_mutex_t handle;
340		};
341
342		class Event
343		{
344		public:
345
346		Event()
347	31.5k	{
348	31.5k	m_counter = 0;
349	31.5k	if (pthread_mutex_init(&m_mutex, NULL) \|\|
350	31.5k	pthread_cond_init(&m_cond, NULL))
351	0	{
352	0	x265_log(NULL, X265_LOG_ERROR, "fatal: unable to initialize conditional variable\n");
353	0	}
354	31.5k	}
355
356		~Event()
357	31.5k	{
358	31.5k	pthread_cond_destroy(&m_cond);
359	31.5k	pthread_mutex_destroy(&m_mutex);
360	31.5k	}
361
362		void wait()
363	37.9k	{
364	37.9k	pthread_mutex_lock(&m_mutex);
365
366		/* blocking wait on conditional variable, mutex is atomically released
367		* while blocked. When condition is signaled, mutex is re-acquired */
368	75.8k	while (!m_counter)
369	37.9k	pthread_cond_wait(&m_cond, &m_mutex);
370
371	37.9k	m_counter--;
372	37.9k	pthread_mutex_unlock(&m_mutex);
373	37.9k	}
374
375		bool timedWait(uint32_t waitms)
376	531	{
377	531	bool bTimedOut = false;
378
379	531	pthread_mutex_lock(&m_mutex);
380	531	if (!m_counter)
381	531	{
382	531	struct timeval tv;
383	531	struct timespec ts;
384	531	gettimeofday(&tv, NULL);
385		/* convert current time from (sec, usec) to (sec, nsec) */
386	531	ts.tv_sec = tv.tv_sec;
387	531	ts.tv_nsec = tv.tv_usec * 1000;
388
389	531	ts.tv_nsec += 1000 * 1000 * (waitms % 1000); /* add ms to tv_nsec */
390	531	ts.tv_sec += ts.tv_nsec / (1000 * 1000 * 1000); /* overflow tv_nsec */
391	531	ts.tv_nsec %= (1000 * 1000 * 1000); /* clamp tv_nsec */
392	531	ts.tv_sec += waitms / 1000; /* add seconds */
393
394		/* blocking wait on conditional variable, mutex is atomically released
395		* while blocked. When condition is signaled, mutex is re-acquired.
396		* ts is absolute time to stop waiting */
397	531	bTimedOut = pthread_cond_timedwait(&m_cond, &m_mutex, &ts) == ETIMEDOUT;
398	531	}
399	531	if (m_counter > 0)
400	531	{
401	531	m_counter--;
402	531	bTimedOut = false;
403	531	}
404	531	pthread_mutex_unlock(&m_mutex);
405	531	return bTimedOut;
406	531	}
407
408		void trigger()
409	38.6k	{
410	38.6k	pthread_mutex_lock(&m_mutex);
411	38.6k	if (m_counter < UINT_MAX)
412	38.6k	m_counter++;
413		/* Signal a single blocking thread */
414	38.6k	pthread_cond_signal(&m_cond);
415	38.6k	pthread_mutex_unlock(&m_mutex);
416	38.6k	}
417
418		protected:
419
420		pthread_mutex_t m_mutex;
421		pthread_cond_t m_cond;
422		uint32_t m_counter;
423		};
424
425		/* This class is intended for use in signaling state changes safely between CPU
426		* cores. One thread should be a writer and multiple threads may be readers. The
427		* mutex's main purpose is to serve as a memory fence to ensure writes made by
428		* the writer thread are visible prior to readers seeing the m_val change. Its
429		* secondary purpose is for use with the condition variable for blocking waits */
430		class ThreadSafeInteger
431		{
432		public:
433
434		ThreadSafeInteger()
435	70.8k	{
436	70.8k	m_val = 0;
437	70.8k	if (pthread_mutex_init(&m_mutex, NULL) \|\|
438	70.8k	pthread_cond_init(&m_cond, NULL))
439	0	{
440	0	x265_log(NULL, X265_LOG_ERROR, "fatal: unable to initialize conditional variable\n");
441	0	}
442	70.8k	}
443
444		~ThreadSafeInteger()
445	70.8k	{
446	70.8k	pthread_cond_destroy(&m_cond);
447	70.8k	pthread_mutex_destroy(&m_mutex);
448	70.8k	}
449
450		int waitForChange(int prev)
451	0	{
452	0	pthread_mutex_lock(&m_mutex);
453	0	if (m_val == prev)
454	0	pthread_cond_wait(&m_cond, &m_mutex);
455	0	pthread_mutex_unlock(&m_mutex);
456	0	return m_val;
457	0	}
458
459		int get()
460	48.8k	{
461	48.8k	pthread_mutex_lock(&m_mutex);
462	48.8k	int ret = m_val;
463	48.8k	pthread_mutex_unlock(&m_mutex);
464	48.8k	return ret;
465	48.8k	}
466
467		int getIncr(int n = 1)
468	0	{
469	0	pthread_mutex_lock(&m_mutex);
470	0	int ret = m_val;
471	0	m_val += n;
472	0	pthread_mutex_unlock(&m_mutex);
473	0	return ret;
474	0	}
475
476		void set(int newval)
477	65.6k	{
478	65.6k	pthread_mutex_lock(&m_mutex);
479	65.6k	m_val = newval;
480	65.6k	pthread_cond_broadcast(&m_cond);
481	65.6k	pthread_mutex_unlock(&m_mutex);
482	65.6k	}
483
484		void poke(void)
485	3.64k	{
486		/* awaken all waiting threads, but make no change */
487	3.64k	pthread_mutex_lock(&m_mutex);
488	3.64k	pthread_cond_broadcast(&m_cond);
489	3.64k	pthread_mutex_unlock(&m_mutex);
490	3.64k	}
491
492		void incr()
493	16.2k	{
494	16.2k	pthread_mutex_lock(&m_mutex);
495	16.2k	m_val++;
496	16.2k	pthread_cond_broadcast(&m_cond);
497	16.2k	pthread_mutex_unlock(&m_mutex);
498	16.2k	}
499
500		void decr()
501	0	{
502	0	pthread_mutex_lock(&m_mutex);
503	0	m_val--;
504	0	pthread_cond_broadcast(&m_cond);
505	0	pthread_mutex_unlock(&m_mutex);
506	0	}
507
508		protected:
509
510		pthread_mutex_t m_mutex;
511		pthread_cond_t m_cond;
512		int m_val;
513		};
514
515	0	#define TIMEOUT_INFINITE 0xFFFFFFFF
516
517		class NamedSemaphore
518		{
519		public:
520		NamedSemaphore()
521	0	: m_sem(NULL)
522	0	#ifndef __APPLE__
523	0	, m_name(NULL)
524	0	#endif //__APPLE__
525	0	{
526	0	}
527
528		~NamedSemaphore()
529	0	{
530	0	}
531
532		bool create(const char* name, const int initcnt, const int maxcnt)
533	0	{
534	0	bool ret = false;
535
536	0	if (initcnt >= maxcnt)
537	0	{
538	0	return false;
539	0	}
540
541		#ifdef __APPLE__
542		do
543		{
544		int32_t pshared = name != NULL ? PTHREAD_PROCESS_SHARED : PTHREAD_PROCESS_PRIVATE;
545
546		m_sem = (mac_sem_t *)malloc(sizeof(mac_sem_t));
547		if (!m_sem)
548		{
549		break;
550		}
551
552		if (pthread_mutexattr_init(&m_sem->mutexAttr))
553		{
554		break;
555		}
556
557		if (pthread_mutexattr_setpshared(&m_sem->mutexAttr, pshared))
558		{
559		break;
560		}
561
562		if (pthread_condattr_init(&m_sem->condAttr))
563		{
564		break;
565		}
566
567		if (pthread_condattr_setpshared(&m_sem->condAttr, pshared))
568		{
569		break;
570		}
571
572		if (pthread_mutex_init(&m_sem->mutex, &m_sem->mutexAttr))
573		{
574		break;
575		}
576
577		if (pthread_cond_init(&m_sem->cond, &m_sem->condAttr))
578		{
579		break;
580		}
581
582		m_sem->curCnt = initcnt;
583		m_sem->maxCnt = maxcnt;
584
585		ret = true;
586		} while (0);
587
588		if (!ret)
589		{
590		release();
591		}
592
593		#else //__APPLE__
594	0	m_sem = sem_open(name, O_CREAT \| O_EXCL, 0666, initcnt);
595	0	if (m_sem != SEM_FAILED)
596	0	{
597	0	m_name = strdup(name);
598	0	ret = true;
599	0	}
600	0	else
601	0	{
602	0	if (EEXIST == errno)
603	0	{
604	0	m_sem = sem_open(name, 0);
605	0	if (m_sem != SEM_FAILED)
606	0	{
607	0	m_name = strdup(name);
608	0	ret = true;
609	0	}
610	0	}
611	0	}
612	0	#endif //__APPLE__
613
614	0	return ret;
615	0	}
616
617		bool give(const int32_t cnt)
618	0	{
619	0	if (!m_sem)
620	0	{
621	0	return false;
622	0	}
623
624		#ifdef __APPLE__
625		if (pthread_mutex_lock(&m_sem->mutex))
626		{
627		return false;
628		}
629
630		int oldCnt = m_sem->curCnt;
631		m_sem->curCnt += cnt;
632		if (m_sem->curCnt > m_sem->maxCnt)
633		{
634		m_sem->curCnt = m_sem->maxCnt;
635		}
636
637		bool ret = true;
638		if (!oldCnt)
639		{
640		ret = 0 == pthread_cond_broadcast(&m_sem->cond);
641		}
642
643		if (pthread_mutex_unlock(&m_sem->mutex))
644		{
645		return false;
646		}
647
648		return ret;
649		#else //__APPLE__
650	0	int ret = 0;
651	0	int32_t curCnt = cnt;
652	0	while (curCnt-- && !ret) {
653	0	ret = sem_post(m_sem);
654	0	}
655
656	0	return 0 == ret;
657	0	#endif //_APPLE__
658	0	}
659
660		bool take(const uint32_t time_out = TIMEOUT_INFINITE)
661	0	{
662	0	if (!m_sem)
663	0	{
664	0	return false;
665	0	}
666
667		#ifdef __APPLE__
668
669		if (pthread_mutex_lock(&m_sem->mutex))
670		{
671		return false;
672		}
673
674		bool ret = true;
675		if (TIMEOUT_INFINITE == time_out)
676		{
677		if (!m_sem->curCnt)
678		{
679		if (pthread_cond_wait(&m_sem->cond, &m_sem->mutex))
680		{
681		ret = false;
682		}
683		}
684
685		if (m_sem->curCnt && ret)
686		{
687		m_sem->curCnt--;
688		}
689		}
690		else
691		{
692		if (0 == time_out)
693		{
694		if (m_sem->curCnt)
695		{
696		m_sem->curCnt--;
697		}
698		else
699		{
700		ret = false;
701		}
702		}
703		else
704		{
705		if (!m_sem->curCnt)
706		{
707		struct timespec ts;
708		ts.tv_sec = time_out / 1000L;
709		ts.tv_nsec = (time_out * 1000000L) - ts.tv_sec * 1000 * 1000 * 1000;
710
711		if (pthread_cond_timedwait(&m_sem->cond, &m_sem->mutex, &ts))
712		{
713		ret = false;
714		}
715		}
716
717		if (m_sem->curCnt && ret)
718		{
719		m_sem->curCnt--;
720		}
721		}
722		}
723
724		if (pthread_mutex_unlock(&m_sem->mutex))
725		{
726		return false;
727		}
728
729		return ret;
730		#else //__APPLE__
731	0	if (TIMEOUT_INFINITE == time_out)
732	0	{
733	0	return 0 == sem_wait(m_sem);
734	0	}
735	0	else
736	0	{
737	0	if (0 == time_out)
738	0	{
739	0	return 0 == sem_trywait(m_sem);
740	0	}
741	0	else
742	0	{
743	0	struct timespec ts;
744	0	ts.tv_sec = time_out / 1000L;
745	0	ts.tv_nsec = (time_out * 1000000L) - ts.tv_sec * 1000 * 1000 * 1000;
746	0	return 0 == sem_timedwait(m_sem, &ts);
747	0	}
748	0	}
749	0	#endif //_APPLE__
750	0	}
751
752		void release()
753	0	{
754	0	if (m_sem)
755	0	{
756		#ifdef __APPLE__
757		pthread_condattr_destroy(&m_sem->condAttr);
758		pthread_mutexattr_destroy(&m_sem->mutexAttr);
759		pthread_mutex_destroy(&m_sem->mutex);
760		pthread_cond_destroy(&m_sem->cond);
761		free(m_sem);
762		m_sem = NULL;
763		#else //__APPLE__
764	0	sem_close(m_sem);
765	0	sem_unlink(m_name);
766	0	m_sem = NULL;
767	0	free(m_name);
768	0	m_name = NULL;
769	0	#endif //__APPLE__
770	0	}
771	0	}
772
773		private:
774		#ifdef __APPLE__
775		typedef struct
776		{
777		pthread_mutex_t mutex;
778		pthread_cond_t cond;
779		pthread_mutexattr_t mutexAttr;
780		pthread_condattr_t condAttr;
781		uint32_t curCnt;
782		uint32_t maxCnt;
783		}mac_sem_t;
784		mac_sem_t *m_sem;
785		#else // __APPLE__
786		sem_t *m_sem;
787		char *m_name;
788		#endif // __APPLE_
789		};
790
791		#endif // ifdef _WIN32
792
793		class ScopedLock
794		{
795		public:
796
797	183k	ScopedLock(Lock &instance) : inst(instance)
798	183k	{
799	183k	this->inst.acquire();
800	183k	}
801
802		~ScopedLock()
803	183k	{
804	183k	this->inst.release();
805	183k	}
806
807		protected:
808
809		// do not allow assignments
810		ScopedLock &operator =(const ScopedLock &);
811
812		Lock &inst;
813		};
814
815		// Utility class which adds elapsed time of the scope of the object into the
816		// accumulator provided to the constructor
817		struct ScopedElapsedTime
818		{
819	0	ScopedElapsedTime(int64_t& accum) : accumlatedTime(accum) { startTime = x265_mdate(); }
820
821	0	~ScopedElapsedTime() { accumlatedTime += x265_mdate() - startTime; }
822
823		protected:
824
825		int64_t startTime;
826		int64_t& accumlatedTime;
827
828		// do not allow assignments
829		ScopedElapsedTime &operator =(const ScopedElapsedTime &);
830		};
831
832		//< Simplistic portable thread class. Shutdown signalling left to derived class
833		class Thread
834		{
835		private:
836
837		ThreadHandle thread;
838
839		public:
840
841		Thread();
842
843		virtual ~Thread();
844
845		//< Derived class must implement ThreadMain.
846		virtual void threadMain() = 0;
847
848		//< Returns true if thread was successfully created
849		bool start();
850
851		void stop();
852		};
853		} // end namespace X265_NS
854
855		#endif // ifndef X265_THREADING_H