Delphi的三种延时及其精度分析

Yourtion 创作于:2010-09-19     全文约 5303 字, 预计阅读时间为 16 分钟

因为PingSwitch项目要不间断Ping交换机所以Ping的速度和评论不能太大,所以要进行延时操作,一般用Sleep,但是好奇就找了一下资料:

在Delphi中,通常可以用以下三种方法来实现程序的延时,即TTtimer控件,Sleep函数,GetTickCount函数。但是其精度是各不相同的。 一、三种方法的简单介绍

1)TTtimer控件

TTtimer控件的实质是调用Windows API定时函数SetTimerKillTimer来实现的,并简化了对WM_TIMER 消息的处理过程。通过设置OnTimer事件和Interval属性,我们可以很方便的产生一些简单的定时事件。

2)Sleep函数

其中Sleep函数用来使程序的执行延时给定的时间值。Sleep的调用形式为Sleep(milliseconds),暂停当前的进程milliseconds毫秒。Sleep的实现方法其实也是调用Windows API的Sleep函数。例如:

sleep(1000); //延迟1000毫秒

Sleep会引起程序停滞,如果你延迟的时间较长的话,你的程序将不能够响应延时期间的发生的其他消息,所以程序看起来好像暂时死机。

3)GetTickCount函数

在主程序中延时,为了达到延时和响应消息这两个目的,GetTickCount()构成的循环就是一种广为流传的方法。例如:

procedure Delay(MSecs: Longint);
//延时函数,MSecs单位为毫秒(千分之1秒)
var
FirstTickCount, Now: Longint;
begin
	FirstTickCount := GetTickCount();
	repeat
	Application.ProcessMessages;
	Now := GetTickCount();
	until (Now - FirstTickCount >= MSecs) or (Now < FirstTickCount);
end;

二、高精度的微妙级性能计数器(high-resolution performance counter)介绍

为了比较以上方法的精度,首先需要找到一个参考的定时器。在这里,我提供了两个参考的定时器。一是用单片机每隔1.024ms产生一个实时中断RTI,作为计数器;二是选用了一个高精度的微妙级性能计数器

(参见:http://msdn.microsoft.com/msdnmag/issues/04/03/HighResolutionTimer/default.aspx ,或者 http://community.csdn.net/Expert/FAQ/FAQ_Index.asp?id=200249 )

1)计数器的Delphi源代码

{
	A high-precision counter/timer. Retrieves time differences
	downto microsec.
	Quick Reference:
	THPCounter inherits from TComponent.

	Key-Methods:
	Start: Starts the counter. Place this call just before the
	code you want to measure.

	Read: Reads the counter as a string. Place this call just
	after the code you want to measure.

	ReadInt: Reads the counter as an Int64. Place this call just
	after the code you want to measure.
}
unit HPCounter;

interface

uses
SysUtils, WinTypes, WinProcs, Messages, Classes, Graphics, Controls,
Forms, Dialogs, StdCtrls, ExtCtrls;

type
TInt64 = TLargeInteger;
THPCounter = class(TComponent)
private
Frequency: TLargeInteger;
lpPerformanceCount1: TLargeInteger;
lpPerformanceCount2: TLargeInteger;
fAbout: string;
procedure SetAbout(Value: string);
{ Private declarations }
public
constructor Create(AOwner: TComponent); override;
destructor Destroy; override;
procedure Start;
function Read: string;
function ReadInt: TLargeInteger;
{ Private declarations }
published
property About: string read fAbout write SetAbout;
{ Published declarations }
end;

procedure Register;

implementation

procedure Register;
begin
	RegisterComponents('MAs Prod.', [THPCounter]);
end;

constructor THPCounter.Create(AOwner: TComponent);
begin
	inherited Create(AOwner);
	fAbout:= 'Version 1.1, 2000® Mats Asplund, EMail: masprod@telia.com, Site: http://go.to/masdp';
end;

destructor THPCounter.Destroy;
begin
	inherited Destroy;
end;

function THPCounter.Read: string;
begin
	QueryPerformanceCounter(TInt64((@lpPerformanceCount2)^));
	QueryPerformanceFrequency(TInt64((@Frequency)^));
	Result:=IntToStr(Round(1000000 * (lpPerformanceCount2 -
	lpPerformanceCount1) / Frequency));
end;

function THPCounter.ReadInt: TLargeInteger;
begin
	QueryPerformanceCounter(TInt64((@lpPerformanceCount2)^));
	QueryPerformanceFrequency(TInt64((@Frequency)^));
	Result:=Round(1000000 * (lpPerformanceCount2 -
	lpPerformanceCount1) / Frequency);
end;

procedure THPCounter.SetAbout(Value: string);
begin
	Exit;
end;

procedure THPCounter.Start;
begin
	QueryPerformanceCounter(TInt64((@lpPerformanceCount1)^));
end;

end.

2)使用方法:

unit Unit1;

interface

uses
Windows, Messages, SysUtils, Classes, Graphics, Controls, Forms, Dialogs,
HPCounter, StdCtrls;

type
	TForm1 = class(TForm)
	Button1: TButton;
	Edit1: TEdit;
	Label1: TLabel;
	Label2: TLabel;
	procedure Button1Click(Sender: TObject);
	private
	{ Private declarations }
	public
	{ Public declarations }
end;

var
Form1: TForm1;

implementation

{$R *.DFM}

procedure TForm1.Button1Click(Sender: TObject);
begin
	Edit1.Text:= '';
	Application.ProcessMessages;
	with THPCounter.Create(Self) do
	begin
		Start;
		// Place code to measure here
		Sleep(1000);
		// Place code to measure here
		Edit1.Text:=Read;
		Free;
	end;
end;

end.

二、三种方法的精度比较

为了比较,采用以上3种方法,分别设置延时时间为1ms、2ms、5ms、10ms、20ms、50ms、100ms、200ms、500ms、1000ms,循环次数为5次,得到实际的延时时间。

1)TTtimer控件

实际延时时间(ms)
1ms: 8.012 21.551 6.875 21.647 9.809
2ms: 9.957 20.675 14.671 11.903 20.551
5ms: 9.952 20.605 9.924 20.705 12.682
10ms:14.852 9.96 21.547 9.82 20.634
20ms:27.512 34.291 26.427 31.244 30.398
50ms:61.196 61.307 64.027 62.048 63.059
100ms:102.495 108.408 112.318 110.322 102.531
200ms:193.955 202.135 207.016 205.082 202.194
500ms:496.659 500.534 503.398 495.551 500.394
1000ms:999.699 1003.576 993.698 1004.443 995.625

2)Sleep函数

1ms: 1.895 1.895 1.896 1.897 1.898
2ms: 2.868 2.874 2.852 2.872 2.869
5ms: 5.8 5.797 5.79 5.79 5.791
10ms:10.675 10.683 10.611 10.669 10.67
20ms:20.404 20.434 20.447 20.477 20.368
50ms:50.67 50.691 50.69 50.682 50.671
100ms:100.515 100.469 100.484 100.481 100.484
200ms:200.101 200.126 199.892 200.066 200.108
500ms:499.961 499.961 499.958 499.961 499.96
1000ms:1000.034 1000.04 1000.03 1000.018 1000.029

3)GetTickCount函数

1ms: 15.54 15.596 15.527 15.566 15.838
2ms: 15.561 15.563 15.603 15.477 15.571
5ms: 15.519 15.549 15.569 15.666 15.394
10ms:15.558 15.561 15.522 15.568 15.518
20ms:31.186 31.137 31.17 31.17 31.19
50ms:62.445 62.4 63.893 60.88 62.404
100ms:109.276 109.298 109.273 109.28 109.28
200ms:203.027 203.084 203.021 203.027 203.046
500ms:499.959 499.961 499.963 499.967 499.965
1000ms:1000.023 1000.022 1000.026 1000.029 1000.021

可见,相对而言,Sleep的精度最高,尤其是在10ms以内的延时,只有sleep函数才能够做到。TTimer控件的定时精度最差,而且稳定性不好,波动很大。GetTickCount函数所能实现的最短延时为15ms左右,稳定性相对TTimer要好一些。

原文链接:https://blog.yourtion.com/delphi-accuracy-delay-analysis.html