Can i use texture as a function parametter Bug in NVCC

Accelerated Computing CUDA CUDA Programming and Performance
1

We know that in OpenGL, texture is represented by a number. It seems natural to me that texture can be use as a function parameter.

So i did the same thing in CUDA

global void kernel(texture<float, 2> tex);

So that my kernel will give different result with different texture input (I hope so)

There’s no compile error, or linking error but

Segment fault (core dump)

Is this a bug with NVCC.

If i don’t use texture as the input parameter then everything is fine.

2

Textures cannot be kernel parameters. They must be declared globally in the .cu that contains the kernel using them.

If you would like to use a different texture, just do another cudaMemcpyToArray() to the texture array or bind the texture to a different array before calling the kernel.

3

Can you verify that. Because that what i did before. There’s a problem with that one that I experienced. If I want to perform the same function but with different size of texture. I don’t think i can change the size of the texture. I try it one and it did not change the size. That why i want to use another texture