64bit precision

not_a_commie · Post by **not_a_commie** » Mon May 21, 2007 10:48 pm

My project (a GIS tool) needs 64bit precision throughout. Has anyone successfully modified Irrlicht to use s64 instead of s32? I saw a post where a guy changed the typedef on s32 to double. I like that idea, but he never got it all working.

Thanks,
Brannon

Ico · Post by **Ico** » Mon May 21, 2007 11:39 pm

Why don't you do your calculations etc. within your own objects and then convert everything for displaying in irrlicht?. (The way most physics library wrapper work.) Create some kind of scale factor to do the conversion (as you won't see a 0.00000000000...1 pixel difference anyway) and to change the precision depending on your detail/zoom level.

hybrid · Post by **hybrid** » Mon May 21, 2007 11:52 pm

You can't change s32 to something different. The colors (4 bytes) depend on it. But you might be able to change f32 to f64.

not_a_commie · Post by **not_a_commie** » Tue May 22, 2007 4:10 pm

So if I were to convert all the f32 to f64, would that mean I could add a transform that would run in double precision? Or is all the data all shipped to the hardware by that point?

hybrid · Post by **hybrid** » Tue May 22, 2007 8:03 pm

You'd have to use an extended meshbuffer format, if it is possible at all. But on the CPU side you can take full advantage of the precision if you really can change all the operations.