#include <iostream>
#include <hip/hip_runtime.h>

#define HIP(e) \
if (hipError_t err = (e); err != hipSuccess) { \
	std::cerr << __FILE__ << ":" << __LINE__ << " " << err << "\n"; \
	exit(1); \
}

__global__ void set(double* p, size_t n) {
    for (size_t idx = blockIdx.x * blockDim.x + threadIdx.x; idx < n; idx += gridDim.x * blockDim.x) {
	    p[idx] = idx;
    }
}

int main(void) {

	size_t n = 1024 * 1024;
	double *p;
	p = (double*)malloc(sizeof(double)*n);
	HIP(hipDeviceSynchronize());

	hipLaunchKernelGGL(set, dim3(128), dim3(128), 0, 0, p, n);
	HIP(hipDeviceSynchronize());

	return 0;
}