I had a problem on using constant memory

Accelerated Computing CUDA CUDA Programming and Performance
1

I have problems,when I tried to use constant memory and I wrote following code

    __constant__ int M;

    void main(void)
    {
    int M_Host = 5;
    cudaMemcpyToSymbol(&M,&M_Host , sizeof(M));
    Kernel()<<<>>>…
    …
    }

In my kernel, I found M didn’ get any value from “cudaMemcpyToSymbol” and my compilation is successful. Does anybody know what happened??

And I did another tests, if I wrote follows, “M” can work in kernel… I’m confused.

constant int M = 5;

2

A. you need to check the return code of pretty much every cuda… function call. I don’t know what you can do when you get a failure, but this is a very important development step. I have had lots of failures with cudaMemcpyToSymbol and it took a while to figure it out.

B. Be aware that in a .CPP module you are referencing a template and not the base function. Look at the template and make sure you understand what it is doing (for?) to you.

I’m betting the cudaMemcpyToSymbol is failing.

3

Yes, you are right, I got error in cudaMemcpyToSymbol and I solve it according to the error message, thank you : ).

4

You shouldn’t open multiple topics for the same problem

Shouldn’t be like this

// after the headers
___constant__ int M; 

....
// inside the code
cudamemcpytosymbol(M,&M_Host,sizeof(int));

?
When a variable is declared in this manner it is global and it can be access directly from all kernels without needing to use it as an argument.

5

I see. Thanks.