beginner's question about texture reference

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi there,

I am beginner to CUDA and I have a question about the compiling error:

“error: no instance of overloaded function “tex3D” matches the argument list” When

I am trying to reference data elements in a 3D texture:

texture<float4, 3, cudaReadModeNormalizedFloat> tex;

float u,v,w;

float4 P = tex3D(tex, u,v,w);

I wrote the above code by following the simpleTexture3D example in SDK:

texture<uchar, 3, cudaReadModeNormalizedFloat> tex;

float voxel = tex3D(tex, u, v, w);

Thanks a lot!

2

It looks correct, but why compiler complains? External Media

3

Is there anyone can help? External Image Thanks!

4

My guess is that its just as the compiler says: there is no overloaded version of tex3D that returns a float4. You have the same problems when trying to adjust the type to just float or something else?

5

Yes, float3 and float2 are not supported for example. But other types like float, char…are ok. In fact, I successfully skipped this problem by using “__ftexfetch”, which is defined in texND(). I don’t know if this is a bug of compiler. :mad:

6

I have used float4’s for 2D texture fetching and everything was fine. What does your channel descriptor look like?

7

The channel descriptor is like this:

cudaChannelFormatDesc channelDesc = cudaCreateChannelDesc();

I figured out the reason of compiler error: the cudaReadModeNormalizedFloat should be replaced by cudaReadModeElementType.