Re: Floating point arithmetic - C / C++ / MFC Discussion Boards

Re: Floating point arithmetic

CPallini28-Jan-08 23:36

CPallini

28-Jan-08 23:36

OfekSH wrote:
it can't be (i think) a direct precision issue

I think the opposite.

OfekSH wrote:
all arguments and intermediate values are *exactly* representable in single precision.

for instance, when executing s*1023 + s sum, you cannot line up the addition arguments without precision loss.

Smile | :)

If the Lord God Almighty had consulted me before embarking upon the Creation, I would have recommended something simpler.
-- Alfonso the Wise, 13th Century King of Castile.

[my articles]

Re: Floating point arithmetic

Maxwell Chen28-Jan-08 23:39

Maxwell Chen

28-Jan-08 23:39

I replaced all occurrences of float with double, and removed all the tail f, then the results match.

Maxwell Chen

Re: Floating point arithmetic

CPallini28-Jan-08 23:51

CPallini

28-Jan-08 23:51

Maxwell Chen wrote:
I replaced all occurrences of float with double, and removed all the tail f, then the results match.

Of course, because you're cheating!
Big Grin | :-D

double numbers have a 52 bit mantissa (float have a 23 bit one) and, for instance, OP's s*1023 and s numbers can be comfortably lined up for addition without precision loss.
Smile | :)

If the Lord God Almighty had consulted me before embarking upon the Creation, I would have recommended something simpler.
-- Alfonso the Wise, 13th Century King of Castile.

[my articles]

Re: Floating point arithmetic

User 345557728-Jan-08 23:53

User 3455577

28-Jan-08 23:53

Maxwell - thanks for your reply!

what do you mean by 'results match'? match mine, or match each other?
i already tried going double, and still got different results:

double d = 1.0 + pow(2.0,-8.0) + pow(2.0,-18.0);

double dRes1 = 1024 * d;

double dRes2 = 0.0;
for(int i=0; i<1024; ++i)
    dRes2 +=d;

on my machine gives:
dRes1= 1028.0039062500000
dRes2= 1028.0002441406250

which didn't even budge the difference down.
do you get different results? on what machine/compiler?

modified 23-Jan-21 21:03pm.

Re: Floating point arithmetic

Maxwell Chen28-Jan-08 23:57

Maxwell Chen

28-Jan-08 23:57

My result:

dRes1 = 1028.003906250000000
dRes2 = 1028.003906250000000

Visual C++ 2005,
Intel Centrino Core 2 Duo, T2400

Maxwell Chen

Re: Floating point arithmetic

User 345557728-Jan-08 23:57

User 3455577

28-Jan-08 23:57

Pallini - thanks, you're right of course, but i suspect there's more to it. going double didn't make any difference (see my post to Maxwell below).

modified 23-Jan-21 21:03pm.

Re: Floating point arithmetic

CPallini29-Jan-08 0:13

CPallini

29-Jan-08 0:13

I tried and got, like Maxwell Chen, the same result (1028.0039062500000) for both dRes1 and dRes2.
I'm using VC++ 2005 on Windows XP.

If the Lord God Almighty had consulted me before embarking upon the Creation, I would have recommended something simpler.
-- Alfonso the Wise, 13th Century King of Castile.

[my articles]

Re: Floating point arithmetic

User 345557729-Jan-08 1:46

User 3455577

29-Jan-08 1:46

Sorry for the delay - just returned from lunch. Upon further investigation, the difference is between debug and release builds. Debug build compiles to fpu code:

	//double dRes2 = 0.0;
00C1A579  fldz             
00C1A57B  fstp        qword ptr [ebp-48h] 
	//for(int i=0; i<1024; ++i)
00C1A57E  mov         dword ptr [i],0 
00C1A585  jmp         CurrentFunction+0C0h (0C1A590h) 
00C1A587  mov         eax,dword ptr [i] 
00C1A58A  add         eax,1 
00C1A58D  mov         dword ptr [i],eax 
00C1A590  cmp         dword ptr [i],400h 
00C1A597  jge         CurrentFunction+0D4h (0C1A5A4h) 
	//{
	//	dRes2 +=d;
00C1A599  fld         qword ptr [ebp-48h] 
00C1A59C  fadd        qword ptr [ebp-28h] 
00C1A59F  fstp        qword ptr [ebp-48h] 
	//}
00C1A5A2  jmp         CurrentFunction+0B7h (0C1A587h)

which gives the wrong result, while release builds SSE code (along with loop unrolling):

	//double dRes2 = 0.0;
004CAB82  xorps       xmm1,xmm1 
004CAB85  mov         eax,80h 
004CAB8A  lea         ebx,[ebx] 
	//for(int i=0; i<1024; ++i)
004CAB90  sub         eax,1 
	//{
	//	dRes2 +=d;
004CAB93  addsd       xmm1,xmm0 
004CAB97  addsd       xmm1,xmm0 
004CAB9B  addsd       xmm1,xmm0 
004CAB9F  addsd       xmm1,xmm0 
004CABA3  addsd       xmm1,xmm0 
004CABA7  addsd       xmm1,xmm0 
004CABAB  addsd       xmm1,xmm0 
004CABAF  addsd       xmm1,xmm0 
004CABB3  jne         CurrentFunction+80h (4CAB90h) 
004CABB5  movsd       mmword ptr [esp+38h],xmm1 
	//}

which gives an accurate result.

setting the FPU control word to double or extended double precision solved it - i guess either your FPU control word was already set right, or you tested it in release builds.

thanks for everyone's time!

modified 23-Jan-21 21:03pm.

Re: Floating point arithmetic

Maxwell Chen29-Jan-08 4:33

Maxwell Chen

29-Jan-08 4:33

I tested in debug build.

Maxwell Chen

The villain is found!

User 34555776-Feb-08 1:06

User 3455577

6-Feb-08 1:06

The project created earlier a Direct3D device. Apparently [^] this changes by default the FPU precision to single.
Haven't decided yet whether to change creation params to use D3DCREATE_FPU_PRESERVE, or to surround the code where i (think i) need extra precision with manual precision set/restore. Either way - problem found.

Thanks everyone!

modified 23-Jan-21 21:03pm.

Timer Intervals

Derick Magagula28-Jan-08 22:03

Derick Magagula

28-Jan-08 22:03

I want to set a timer to execute in every two minutes in visual C++. Please help

Re: Timer Intervals

Cedric Moonen28-Jan-08 22:09

Cedric Moonen

28-Jan-08 22:09

If you are using MFC, you can use CWnd::SetTimer[^]

BTW, don't forget your friend google...

Cédric Moonen
Software developer

Charting control [v1.2]

Re: Timer Intervals [modified]

Maxwell Chen28-Jan-08 22:09

Maxwell Chen

28-Jan-08 22:09

::SetTimer(1, 120000, MyTest);
// ...
void CALLBACK MyTest(HWND hW, UINT msg, UINT ev, DWORD time) {
  // ... 
}

Maxwell Chen

modified on Tuesday, January 29, 2008 10:35:27 AM

Re: Timer Intervals

David Crow29-Jan-08 3:07

David Crow

29-Jan-08 3:07

Maxwell Chen wrote:
::SetTimer(1, 120, MyTest);

Perhaps you meant to use 120000 instead. Confused | :confused:

"Normal is getting dressed in clothes that you buy for work and driving through traffic in a car that you are still paying for, in order to get to the job you need to pay for the clothes and the car and the house you leave vacant all day so you can afford to live in it." - Ellen Goodman

"To have a respect for ourselves guides our morals; to have deference for others governs our manners." - Laurence Sterne

Re: Timer Intervals

Maxwell Chen29-Jan-08 4:35

Maxwell Chen

29-Jan-08 4:35

DavidCrow wrote:
Perhaps you meant to use 120000 instead.

Yes.

Maxwell Chen

Re: Timer Intervals

CPallini28-Jan-08 22:54

CPallini

28-Jan-08 22:54

Have a look at SetTimer function documentation [^].
Smile | :)

If the Lord God Almighty had consulted me before embarking upon the Creation, I would have recommended something simpler.
-- Alfonso the Wise, 13th Century King of Castile.

[my articles]

Working with CFile!!!!

rowdy_vc++28-Jan-08 21:57

rowdy_vc++

28-Jan-08 21:57

Hi Friends....

i have some doubts regarding with my project ..i am trouble and i dont know how to finish up this one..

Tis is my queries:

1.i have one file that file contains list of data.first i wanna convert

that data Cfile into CStringArray how its possible.i used these function

CFile Lfile;
CStringArray data;
data=Lfile.GetLenght();

is it right conversion to data converting CFile to CStringArray..its shows one error :
error C2679: binary '=' : no operator found which takes a right-hand operand of type 'ULONGLONG' (or there is no acceptable conversion)

how to cure it ? r wat method is suitable for converting CFile to CString Array.?

Thanks

raju.k

Re: Working with CFile!!!!

Maxwell Chen28-Jan-08 22:04

Maxwell Chen

28-Jan-08 22:04

CFile file("data.txt", CFile::modeReadWrite | CFile::modeCreate);
CStringArray ar;
CString s;
s.format("%d", file.GetLength());
ar.Add(s);

Maxwell Chen

Re: Working with CFile!!!!

Cedric Moonen28-Jan-08 22:12

Cedric Moonen

28-Jan-08 22:12

I think the OP wanted to read all the strings from the file and convert them into a CStringArray, not adding the file size to the string array.
Am I wrong ?

Cédric Moonen
Software developer

Charting control [v1.2]

Re: Working with CFile!!!!

Maxwell Chen28-Jan-08 22:16

Maxwell Chen

28-Jan-08 22:16

Maxwell Chen

Re: Working with CFile!!!!

Cedric Moonen28-Jan-08 22:06

Cedric Moonen

28-Jan-08 22:06

You already asked this question yesterday and people told you that it is not the way to go. The GetLenght function only returns the size of the file. How could the compiler convert it to a string array Confused | :confused:

?

No, you really need to read your file yourself and extract each string, create a CString object and push it into the CStringArray.

Cédric Moonen
Software developer

Charting control [v1.2]

Re: Working with CFile!!!!

rowdy_vc++28-Jan-08 22:38

rowdy_vc++

28-Jan-08 22:38

Thanks guys its working ...i am really happy ..

Thank u very much

raju.k

Re: Working with CFile!!!!

Maxwell Chen28-Jan-08 23:05

Maxwell Chen

28-Jan-08 23:05

You really took my solution?! OMG | :OMG:

Maxwell Chen

Re: Working with CFile!!!!

Rajesh R Subramanian28-Jan-08 22:51

Rajesh R Subramanian

28-Jan-08 22:51

May be you need to use CStdioFile instead of CFile ?

CStdioFile csf;
CFileException cfe;
CString str;
CStringArray csa;

csf.Open(_T("E:\\myfile.txt"), CFile::modeRead, &cfe);

while(csf.ReadString(str))//ReadString will read one line at a time.
{
    csa.Add(str);//Add the new line to our CStringArray.
}
AfxMessageBox(csa.GetAt(0));

Nobody can give you wiser advice than yourself. - Cicero
.·´¯`·->Rajesh<-·´¯`·.
Codeproject.com: Visual C++ MVP

Problem in AfxBeginThread

gReaen28-Jan-08 21:35

gReaen

28-Jan-08 21:35

I need to call the function AfxBeginThread in one of my function (say, myFunc). myFunc should receive a function name which should become a thread.

My code is as follows:
void myFunc(void* fthread)
{
HANDLE cmd_thread;
CWinThread* pCmdThread = AfxBeginThread(fthread, NULL);
cmd_thread = pCmdThread->m_hThread;
..........
}

But i get an error that: cannot convert parameter one of AfxBeginThread from void* to unsigned int(_cdecl*)(void *).....
What should the parameter type be?

Use Ctrl+Left/Right to switch messages, Ctrl+Up/Down to switch threads, Ctrl+Shift+Left/Right to switch pages.