CUDA capabilities gist

cuda.cu

#include "cuda_runtime.h"
#include "device_launch_parameters.h"
#include "device_functions.h"

#include <stdio.h>

void print(char*, int[], int);

__global__ void square(int *a) { // our computation logic, will be run on GPU
	a[threadIdx.x] *= a[threadIdx.x];
	__syncthreads(); // synchronizes threads of a block
	                 // needed when number of specified threads exceeds
	                 // the number of actual available ones
}

int main() {
	int arr[] = { 1, 2, 3, 4, 5, 6 };
	int arraySizeOnMemory = sizeof arr;
	int arrayEntriesCount = arraySizeOnMemory / sizeof(int);

	print("arr: ", arr, arrayEntriesCount); // our custom array printer

	int *arrGPU; // handle of GPU allocated memory
	cudaMalloc((void**)&arrGPU, arraySizeOnMemory); // allocate GPU memory for our array
	// copy our array to the allocated GPU memory
	cudaMemcpy(arrGPU, arr, arraySizeOnMemory, cudaMemcpyHostToDevice);

	dim3 dimBlock(arrayEntriesCount, 1); // define GPU computation configs
	dim3 dimGrid(1, 1);
	square<<<dimGrid, dimBlock>>>(arrGPU); // actual GPU processing happens here
	cudaThreadSynchronize(); // above was an async call so we need to ensure it is finished

	cudaMemcpy(arr, arrGPU, arraySizeOnMemory, cudaMemcpyDeviceToHost); // copy back the results
	// free the GPU allocated memory
	cudaFree(arrGPU);

	print("\nResult of GPU processing (x .^ 2): ", arr, arrayEntriesCount);

	getchar(); // don't exit immediately
	return 0;
}

void print(char* str, int arr[], int n) {
	printf("%s{ ", str);
	for (int i = 0; i < n; ++i)
		printf("%d%s", arr[i], n - 1 == i ? "" : ", ");
	printf(" }\n");
}