cudaBindTexture trouble in 0.9

Accelerated Computing CUDA CUDA Programming and Performance
1

Hello,

When trying to convert my project to work with version 0.9 of CUDA I have run into problems. It seems wherever I try to use the cudaBindTexture method, it fails to compile. I finally tried putting the following into the simpleTexture sample (which otherwise works fine):

float *devptr; 

	cudaMalloc((void **)&devptr, 256); 

	texture<float, 2, cudaReadModeElementType> texRef;

	cudaBindTexture(texRef, devptr, 256, 0);

And then it fails to compile:

Looking through the sample code I can’t find a single reference to cudaBindTexture (just cudaBindTextureToArray). What am I doing wrong? It worked fine in 0.8. Thanks for any help,

2

I have been having similar troubles. It looks like not a single one of the examples uses linear memory bound to a texture…so no help with examples there (maybe there should be one…). I have also tried using the “low-level API” version of the cudaBindTexture function to no avail.

3

Oh, good. It’s not just me then ;). So I’m guessing this is either a bug of the compiler or an undocummented new feature. NVIDIA folks, any ideas? I’d really like to start taking advantage of asyncronous kernels.

4

There is an error in the documentation you need to use the following function call:

cudaBindTexture(0, texRef, devPtr, size);

The 4th argument in the documentation is actually the first argument in the actual implementation.

The error is on the bottom of page 32 in the programming manual.

Jeroen