C++多執行緒（二）

f66666602

一 簡單實例（來自codeprojct：http://www.codeproject.com/useritems/MultithreadingTutorial.asp）
主執行緒創建2個執行緒t1和t2，創建時2個執行緒就被掛起，後來調用ResumeThread恢復2個執行緒，是其開始執行，調用WaitForSingleObject等待2個執行緒執行完，然後推出主執行緒即結束進程。
/**//*  file Main.cpp
*
*  This program is an adaptation of the code Rex Jaeschke showed in
*  Listing 1 of his Oct 2005 C/C++ User's Journal article entitled
*  "C++/CLI Threading: Part I".  I changed it from C++/CLI (managed)
*  code to standard C++.
*
*  One hassle is the fact that C++ must employ a free (C) function
*  or a static class member function as the thread entry function.
*
*  This program must be compiled with a multi-threaded C run-time
*  (/MT for LIBCMT.LIB in a release build or /MTd for LIBCMTD.LIB
*  in a debug build).
*
*                                      John Kopplin  7/2006
*/


#include <stdio.h>
#include <string>
// for STL string class
#include <windows.h>
// for HANDLE
#include <process.h>
// for _beginthread()

using
namespace std;


class ThreadX
{
private:
  int loopStart;
  int loopEnd;
  int dispFrequency;

public:
  string threadName;

  ThreadX( int startValue、int endValue、int frequency )
  {
    loopStart = startValue;
    loopEnd = endValue;
    dispFrequency = frequency;
  }

  // In C++ you must employ a free (C) function or a static
  // class member function as the thread entry-point-function.
  // Furthermore、_beginthreadex() demands that the thread
  // entry function signature take a single (void*) and returned
  // an unsigned.

static unsigned __stdcall ThreadStaticEntryPoint(void
* pThis)
  {
      ThreadX * pthX = (ThreadX*)pThis;   // the tricky cast
      pthX->ThreadEntryPoint();           // now call the true entry-point-function

      // A thread terminates automatically if it completes execution,
      // or it can terminate itself with a call to _endthread().

      return
1;          // the thread exit code
  }

  void ThreadEntryPoint()
  {
     // This is the desired entry-point-function but to get
     // here we have to use a 2 step procedure involving
     // the ThreadStaticEntryPoint() function.

    for (int i = loopStart; i <= loopEnd; ++i)
    {
      if (i % dispFrequency ==
0)
      {
          printf( "%s: i = %d\n"、threadName.c_str()、i );
      }
    }
    printf( "%s thread terminating\n"、threadName.c_str() );
  }
};


int main()
{
    // All processes get a primary thread automatically. This primary
    // thread can generate additional threads.  In this program the
    // primary thread creates 2 additional threads and all 3 threads
    // then run simultaneously without any synchronization.  No data
    // is shared between the threads.

    // We instantiate an object of the ThreadX class. Next we will
    // create a thread and specify that the thread is to begin executing
    // the function ThreadEntryPoint() on object o1. Once started,
    // this thread will execute until that function terminates or
    // until the overall process terminates.

    ThreadX * o1 =
new ThreadX( 0、1、2000 );

    // When developing a multithreaded WIN32-based application with
    // Visual C++、you need to use the CRT thread functions to create
    // any threads that call CRT functions. Hence to create and terminate
    // threads、use _beginthreadex() and _endthreadex() instead of
    // the Win32 APIs CreateThread() and EndThread().

    // The multithread library LIBCMT.LIB includes the _beginthread()
    // and _endthread() functions. The _beginthread() function performs
    // initialization without which many C run-time functions will fail.
    // You must use _beginthread() instead of CreateThread() in C programs
    // built with LIBCMT.LIB if you intend to call C run-time functions.

    // Unlike the thread handle returned by _beginthread()、the thread handle
    // returned by _beginthreadex() can be used with the synchronization APIs.

    HANDLE   hth1;
    unsigned  uiThread1ID;

    hth1 = (HANDLE)_beginthreadex( NULL,         // security

0,            // stack size
                                   ThreadX::ThreadStaticEntryPoint,
                                   o1,           // arg list
                                   CREATE_SUSPENDED,  // so we can later call ResumeThread()

&uiThread1ID );

    if ( hth1 ==
0 )
        printf("Failed to create thread 1\n");

    DWORD   dwExitCode;

    GetExitCodeThread( hth1、&dwExitCode );  // should be STILL_ACTIVE = 0x00000103 = 259
    printf( "initial thread 1 exit code = %u\n"、dwExitCode );

    // The System::Threading::Thread object in C++/CLI has a "Name" property.
    // To create the equivalent functionality in C++ I added a public data member
    // named threadName.

    o1->threadName =
"t1";

    ThreadX * o2 =
new ThreadX( -1000000、0、2000 );

    HANDLE   hth2;
    unsigned  uiThread2ID;

    hth2 = (HANDLE)_beginthreadex( NULL,         // security

0,            // stack size
                                   ThreadX::ThreadStaticEntryPoint,
                                   o2,           // arg list
                                   CREATE_SUSPENDED,  // so we can later call ResumeThread()

&uiThread2ID );

    if ( hth2 ==
0 )
        printf("Failed to create thread 2\n");

    GetExitCodeThread( hth2、&dwExitCode );  // should be STILL_ACTIVE = 0x00000103 = 259
    printf( "initial thread 2 exit code = %u\n"、dwExitCode );

    o2->threadName =
"t2";

    // If we hadn't specified CREATE_SUSPENDED in the call to _beginthreadex()
    // we wouldn't now need to call ResumeThread().

    ResumeThread( hth1 );   // serves the purpose of Jaeschke's t1->Start()

    ResumeThread( hth2 );

    // In C++/CLI the process continues until the last thread exits.
    // That is、the thread's have independent lifetimes. Hence
    // Jaeschke's original code was designed to show that the primary
    // thread could exit and not influence the other threads.

    // However in C++ the process terminates when the primary thread exits
    // and when the process terminates all its threads are then terminated.
    // Hence if you comment out the following waits、the non-primary
    // threads will never get a chance to run.

    WaitForSingleObject( hth1、INFINITE );
    WaitForSingleObject( hth2、INFINITE );

    GetExitCodeThread( hth1、&dwExitCode );
    printf( "thread 1 exited with code %u\n"、dwExitCode );

    GetExitCodeThread( hth2、&dwExitCode );
    printf( "thread 2 exited with code %u\n"、dwExitCode );

    // The handle returned by _beginthreadex() has to be closed
    // by the caller of _beginthreadex().

    CloseHandle( hth1 );
    CloseHandle( hth2 );

    delete o1;
    o1 = NULL;

    delete o2;
    o2 = NULL;

    printf("Primary thread terminating.\n");
}

二解釋
1）如果你正在編寫C/C++代碼，決不應該調用CreateThread。相反，應該使用VisualC++運行期庫函數_beginthreadex，推出也應該使用_endthreadex。如果不使用Microsoft的VisualC++編譯器，你的編譯器供應商有它自己的CreateThred替代函數。不管這個替代函數是什麼，你都必須使用。
2）因為_beginthreadex和_endthreadex是CRT執行緒函數，所以必須注意編譯選項runtimelibaray的選擇，使用MT或MTD。
3)_beginthreadex函數的參數列表與CreateThread函數的參數列表是相同的，但是參數名和類型並不完全相同。這是因為Microsoft的C/C++運行期庫的開發小組認為，C/C++運行期函數不應該對Windows資料類型有任何依賴。_beginthreadex函數也像CreateThread那樣，返回新創建的執行緒的句柄。
下面是關於_beginthreadex的一些要點：
．每個執行緒均獲得由C/C++運行期庫的堆棧分配的自己的tiddata記憶體結構。（tiddata結構位於Mtdll.h檔案中的VisualC++來源碼中）。
．傳遞給_beginthreadex的執行緒函數的地址保存在tiddata記憶體塊中。傳遞給該函數的參數也保存在該資料塊中。
．_beginthreadex確實從內部調用CreateThread，因為這是操作系統瞭解如何創建新執行緒的唯一方法。
．當調用CreatetThread時，它被告知通過調用_threadstartex而不是pfnStartAddr來啟動執行新執行緒。還有，傳遞給執行緒函數的參數是tiddata結構而不是pvParam的地址。
．如果一切順利，就會像CreateThread那樣返回執行緒句柄。如果任何操作失敗了，便返回NULL。
4) _endthreadex的一些要點：
．C運行期庫的_getptd函數內部調用操作系統的TlsGetValue函數，該函數負責檢索調用執行緒的tiddata記憶體塊的地址。
．然後該資料塊被釋放，而操作系統的ExitThread函數被調用，以便真正撤消該執行緒。當然，退出代碼要正確地設置和傳遞。
5)雖然也提供了簡化版的的_beginthread和_endthread，但是可控制性太差，所以一般不使用。
6）執行緒handle因為是內核對象，所以需要在最後closehandle。
7）更多的API：HANDLE GetCurrentProcess();HANDLE GetCurrentThread();DWORDGetCurrentProcessId();DWORD GetCurrentThreadId()。DWORDSetThreadIdealProcessor(HANDLE hThread,DWORD dwIdealProcessor);BOOLSetThreadPriority(HANDLE hThread,int nPriority);BOOLSetPriorityClass(GetCurrentProcess(),  IDLE_PRIORITY_CLASS);BOOLGetThreadContext(HANDLE hThread,PCONTEXT pContext);BOOLSwitchToThread();

三注意
1）C++主執行緒的終止，同時也會終止所有主執行緒創建的子執行緒，不管子執行緒有沒有執行完畢。所以上面的代碼中如果不調用WaitForSingleObject，則2個子執行緒t1和t2可能並沒有執行完畢或根本沒有執行。
2）如果某執行緒掛起，然後有調用WaitForSingleObject等待該執行緒，就會導致死鎖。所以上面的代碼如果不調用resumethread，則會死鎖。