Expected ; on CUDA kernel - c++

Question

I am having a very weird error while trying to create a CUDA kernel to execute a for loop:
#include <stdlib.h>
#include <stdio.h>
#include <thrust/reduce.h>
#include <cuda.h>
int main(int argc, char** argv)
{
float *arrayA;
cudaMalloc((void**)&arrayA, 4096 * 4096 * sizeof(float));
float *arrayB;
cudaMalloc((void**)&arrayB, 4096 * 4096 * sizeof(float));
__global__ void loopKernel(float* arrayA, float* arrayB)
{
int i = threadIdx.x + blockDim.x*blockIdx.x;
if (i < m)
{
//do stuf
}
}
loopKernel << 8, 256 >> (arrayA, arrayB);
}
the error is on the opening { for the kernel (line 14):
error: expected a ";"
it seems really odd as I get the same error on Visual Studio and linux terminal, so it is not an OS issue.
The file is also .cu so there is no way it's being sent to the wrong compiler.
Any help will be appreciated.

Answer 1

A __global__ function definition (i.e. kernel definition) is not something you do within the body of another function. We don't typically do this in C or C++ either (a C/C++ function definition is not usually placed within the body of another function definition).
Place your kernel definitions at global scope (i.e. outside the body of any other function definition, including main).
Something like this:
#include <stdlib.h>
#include <stdio.h>
#include <thrust/reduce.h>
#include <cuda.h>
__global__ void loopKernel(float* arrayA, float* arrayB)
{
int i = threadIdx.x + blockDim.x*blockIdx.x;
if (i < m)
{
arrayA[i] = 0.f;
arrayB[(n - 1)*m + i] = 0.f;
}
}
int main(int argc, char** argv)
{
float *arrayA;
cudaMalloc((void**)&arrayA, 4096 * 4096 * sizeof(float));
float *arrayB;
cudaMalloc((void**)&arrayB, 4096 * 4096 * sizeof(float));
loopKernel << 8, 256 >> (arrayA, arrayB);
}
There are various other issues with the posted code:
It provided no definition for m or n.
The kernel calling syntax is wrong, instead of <<...>> it should be <<<...>>>
For these types of basic issues, its probably better to study a simple (correct) code like the vectorAdd sample code.