CUDA beginner asking for help

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi all,

I just started learning CUDA and I have a very basic program.

#include <stdio.h>
#include <stdlib.h>
#include <cuda.h>

__global__ void testCUDA(int *a, int *b){
	a[0] += b[0];
}

int main(){
	printf("hello world!\n");

	return 0;
}

I hide some parts.

A warning is pointed at the start of “global”: this declaration has no storage class or type specifier.
An error is pointed at the start of “void testCUDA()”: expected a “;”.

I have cuda.h included and double underscore on both sides of “global”.

Can anyone help me out here?

Thanks so much!

Tim

2

CUDA programs need to be compiled by nvcc, not your system’s C or C++ compiler.

Once you do that, you can also leave out the cuda.h include.

3

Hi Tera,

Thanks for your reply.
I’m compiling my code on GPU cluster from a provider and the compiler they provide is icc.
They don’t have much tutorial on this and their only example is CUBLAS.
I tried to modify the way of compiling CUBLAS and made sure it finds cuda.h, and this is how I got the above problems.

4

Perhaps this gives an idea about icc:
[url]Intel compiler support for front-end CUDA compilation - CUDA Programming and Performance - NVIDIA Developer Forums

Regardless, the CUDA toolkit needs to be on the test system you are running… and you need to call nvcc, which is NVIDIA’s compiler… example:

[url]http://code.google.com/p/stanford-cs193g-sp2010/wiki/TutorialHelloWorld[/url]