Windows下用MFC实现微秒级延时

   一帮情况可以使用Sleep()函数来实现延时，但Windows不是实时的，是操作系统来分配处理器给多个线程的，而不会线程一直拥有处理器的使用权。比如延时50ms，不论采用什么方式来延时，50ms以后，操作系统未必就正好把处理器分配给这个线程使用。
   使用一个Sleep(50)，这下这个线程就暂停了，50ms以内操作系统就不会把处理器给这个线程用，50ms以后这个线程要求处理器，操作系统才会可能给他分配，但是仅仅是可能，也有可能这时候有一个优先级更高的线程也要用处理器，操作系统就不一定给他。可能等到这个更高优先级的线程用完了，这时候已经过了55ms了，就不是精确的了。

   在Windows平台下，常用的计时器还有另外两种，一种是timeGetTime多媒体计时器，它可以提供毫秒级的计时。但这个精度对很多应用场合而言还是太粗糙了。另一种是QueryPerformanceCount计数器，随系统的不同可以提供微秒级的计数。对于实时图形处理、多媒体数据流处理、或者实时系统构造的程序员，善用QueryPerformanceCount/QueryPerformanceFrequency是一项基本功。

  可以使用 QueryPerformanceCount/QueryPerformanceFrequency实现微秒级的延时，例子如下：

void DelayUs(double uDelay)
{
	LARGE_INTEGER litmp;
    LONGLONG QPart1,QPart2;

      double dfMinus,dfFreq,dfTim; //定义变量

      /*
         Pointer to a variable that the function sets, in counts per second, to the current performance-counter frequency.
         If the installed hardware does not support a high-resolution performance counter,
         the value passed back through this pointer can be zero.

     */
     QueryPerformanceFrequency(&litmp); //获取当前电脑的时钟频率

     dfFreq = (double)litmp.QuadPart;

     /*
         Pointer to a variable that the function sets, in counts, to the current performance-counter value.
     */
     QueryPerformanceCounter(&litmp); //获取延时前的CPU执行次数

     QPart1 = litmp.QuadPart;
     do
     {
           QueryPerformanceCounter(&litmp);//获取延时后的CPU执行次数
         QPart2 = litmp.QuadPart;
         dfMinus = (double)(QPart2-QPart1);//延时后的CPU执行次数 减去  延时前的CPU执行次数
		 if (dfMinus < 0)
			 break;
         dfTim = dfMinus/dfFreq * 1000000;
		 //dfTim = dfMinus/dfFreq;
      }while(dfTim< uDelay); //延时时间小于预定值，继续进行延时

}

  对延时函数进行测试：

#include <stdio.h>
#include <Windows.h>

void DelayUs(double uDelay)
{
	LARGE_INTEGER litmp;
    LONGLONG QPart1,QPart2;

      double dfMinus,dfFreq,dfTim;

      /*
         Pointer to a variable that the function sets, in counts per second, to the current performance-counter frequency.
         If the installed hardware does not support a high-resolution performance counter,
         the value passed back through this pointer can be zero.

     */
     QueryPerformanceFrequency(&litmp);

     dfFreq = (double)litmp.QuadPart;

     /*
         Pointer to a variable that the function sets, in counts, to the current performance-counter value.
     */
     QueryPerformanceCounter(&litmp);

     QPart1 = litmp.QuadPart;
     do
     {
           QueryPerformanceCounter(&litmp);
         QPart2 = litmp.QuadPart;
         dfMinus = (double)(QPart2-QPart1);
		 if (dfMinus < 0)
			 break;
         dfTim = dfMinus/dfFreq * 1000000;
		 //dfTim = dfMinus/dfFreq;
      }while(dfTim< uDelay);

}

void main()
{
LARGE_INTEGER t1, t2, tc;
QueryPerformanceFrequency(&tc);
printf("Frequency: %u \n", tc.QuadPart);
QueryPerformanceCounter(&t1);
//Sleep(1000);
DelayUs(20000);
QueryPerformanceCounter(&t2);
printf("Begin Time: %u \n", t1.QuadPart);
printf("End Time: %u \n", t2.QuadPart);
printf("Lasting Time: %u \n",( t2.QuadPart- t1.QuadPart));
}

程序执行结果：

 

机器的CPU频率为3117382Hz

延时前CPU执行次数为：985262162

延时后CPU执行次数为：985324514

延时函数总共耗费CPU执行次数为：62352

 实际延时时间为：62352/3117382 ≈0.02000139860947423190356523518773

延时误差在0.2微秒以内，基本能达到微秒级的延时要求。