What's wrong with cudaMallocPitch()?

Accelerated Computing CUDA CUDA Programming and Performance
1

I have been banging my head against the wall trying to copy 2D matrices to the GPU and back to the host. I am trying to use the cudaMallocPitch() method, but it simply doesn’t work. I mean, it never compiles.

[codebox]///// Inside Kernel.cu /////

int main()

{

    float* devPtr; 

    int pitch; 

    cudaMallocPitch((void**)&devPtr, &pitch, 3 * sizeof(float), 3);

return 0;

}[/codebox]

This code will not compile. This the most stripped down usage of cudaMallocPitch() I have even seen. Upon trying to compile, I get the error message, “error: no instance of overloaded function “cudaMallocPitch” matches the argument list”, which means that I’m sending in the wrong set of parameters than cudaMallocPitch() requires; however, that code I just posted is identical to what you see in the Programming Guide 3.0, page 20.

What gives?

EDIT: And when I changed “int pitch” to “size_t pitch”, it worked fine. I hate not being told the right thing to do. Oh well!

2

Hi

change your cudaMallocPitch to

cudaMallocPitch((float**)&devPtr, (size_t)&pitch, 3 * sizeof(float), 3);

This should solve your problem

3

hi,it should be cudaMallocPitch((float**)&devPtr, (size_t *)&pitch, 3 * sizeof(float), 3);

you lose a * :rolleyes:

4

hi,it should be cudaMallocPitch((float**)&devPtr, (size_t *)&pitch, 3 * sizeof(float), 3);

you lose a * :rolleyes: