<Solved> Strange result when multiplying two double numbers

[Doc_QuicknDirty]

Hello,

Can anyone please explain me the logic behind this in Orge 1.7:

Code: Select all

double tmp = 47.6 * 6378137.0; // tmp  = 303599321.19999999

O.k. that's what I expected, but:

Code: Select all

double con = 6378137.0;
double tmp = 47.6 * con; // tmp = 303599328.00000000

The multiplication doesn't work no matter if OGRE_DOUBLE_PRECISION is set to 1 or 0.

I would be grateful for any hint.

Best wishes

Ralf

[Fish]

What does this have to do with Ogre? Is this not a computer science issue: an issue of double/float precision?

[madmarx]

I could answer you with the following lines :
"
Why does this result be the same even when you don't use OGRE ?
Why did you not discriminate between a static optimisation and a code that is evaluated at run time ?
Why didn't you read the generated assembly to check that?
And most of all, why did you assume that doubles have more than 8 valid digits on your platform ?
"

So please change the offensive title .

Anyway, hope this helps.

[haffax]

What Every Computer Scientist Should Know About Floating-Point Arithmetic

[Kojack]

Are you using the direct3d renderer? If so, what option do you have set for the floating point mode in the config dialog? (This is unrelated to the OGRE_DOUBLE_PRECISION setting)
The default behaviour of Direct3D is to turn the FPU to single precision mode, even for doubles. Ogre can prevent that, but only if you put the mode to "consistent" instead of "fastest".

[Doc_QuicknDirty]

Hello,

I forgot: platform Windows 7 with VS2005

@madmax
Title changed.

The result is NOT the same when I don't use Ogre.

Code: Select all

INT WINAPI WinMain( HINSTANCE hInst, HINSTANCE, LPSTR strCmdLine, INT )
{

   double tmp1 = 47.6 * 6378137.0; // tmp1  = 303599321.19999999

   double con = 6378137.0;
   double tmp2 = 47.6 * con; // tmp2  = 303599321.19999999

   // Create application object
   OgreDemoApplication app;
  
  //From now on the calculation for tmp2 doesn't work any more

...

Best wishes

Ralf

[Doc_QuicknDirty]

Hello,

@Kojack
Billion thanks! You saved my day!!!

Best wishes

Ralf

[betajaen]

You can also try adjusting the floating point model in your project settings.

ALT-F7, C/C++, Code Generation, Floating Point Model.

If I remember correctly Ogre has it on "fast", and your other app is likely on precise.