Warning: Doulbe is not supported. Demoting to float Unexpected warning.

Accelerated Computing CUDA CUDA Programming and Performance
1

When compile a cuda code, I ran into a “Warning: double is not supported. Demoting to float”. But there is not double in my whole program. Need a explanation for this. Thanks in advance.

2

There probably is a double you missed. A common way is in a literal constant you don’t even think about.

float a, b, norm;

....

norm=1.0/(a*a+b*b);

The above will give you the warning because of the literal 1.0 constant. You can use “1.0f” instead to hide the warning.

3

I found this is not due to a double type variable in the kernel, as mater of fact there is not double type in my kernel. It is so because I used a struct type variable in the kernel. And in the struct, there are float type and int type.

4

Thanks. I need test that.

5

Hi, SPWorley, Thanks for your suggestion. It did not show the warning in

float a, b, norm;

....

norm=1.0/(a*a+b*b);

in my case. But use 1.0f is definitely a good suggestion.

The reason for the warning in my case is due to the usage of the struct in the kernel.

typedef struct

{

   float ro[3];

   float d[3];

   float t;

   int index;

} ray;
6

Are you sure,triston? How did you get the reason?

7

I’ve seen this same warning in one of my kernels, which also has no floats in it. I’ve ignored it so far, but maybe I should figure out what is going on. :)