/src/x265/source/common/threading.h

Source (jump to first uncovered line)
/*****************************************************************************
 * 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);
}

#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)  no_atomic_add((int*)ptr, 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 int32_t*)ptr, 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)  InterlockedExchangeAdd((volatile LONG*)ptr, 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: 2025-07-23 08:18

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