Floating point arithmetic

Accelerated Computing CUDA CUDA Programming and Performance
1

Since my GPU card does not support double precision, I have to use floating point instead of double. Then I encounter the following issues!

  1. A floating point type variable’s values become integers. This variable is the result of some math calculations. I don’t understand why so because it was declared and initialized to be float point. Anything I am not aware of about cuda? Is there any way to specify the precision?

  2. This is a general question. What’s the best way to rewrite the programs from c++ to cuda in which c++ double variables need to be converted into float point type? The straight forward way is to simply re-declare and re-initialize them from double to float point. But this caused above problem for me. any suggestions?

plz help and thanks!

2

Can you post some sample code? FP values spontaneously becoming integers sounds rather strange.

3

Thanks YDD! The code is large and I tried to narrow it down to a more specific question or code. But I couldn’t identify it yet.

The 7 significant digits for floating point might not be accurate enough for my calculation.