Chapter3-EasyCudaProject

Chapter 2: Simple CUDA Vector Addition

Overview

In this chapter, we'll create a basic CUDA program that performs vector addition using the GPU. The program consists of a CUDA kernel that adds corresponding elements from two input vectors and stores the result in an output vector.

Code Explanation

Let's go through the code step by step:

1. Kernel Function (vectorAdd):

   • The vectorAdd function is the heart of our CUDA program. It runs on the GPU and performs the vector addition.
   • It takes four arguments:
     • A: Pointer to the first input vector.
     • B: Pointer to the second input vector.
     • C: Pointer to the output vector (where the result will be stored).
     • size: Size of the vectors (number of elements).
   • Inside the kernel, each thread computes the sum of corresponding elements from A and B and stores the result in C.

2. Main Function:

   • The main function sets up the host (CPU) and device (GPU) memory, initializes input vectors, launches the kernel, and retrieves the result.
   • Key steps in the main function:
     • Allocate memory for vectors (h_A, h_B, and h_C) on the host.
     • Initialize input vectors (h_A and h_B) with sample values.
     • Allocate memory for vectors (d_A, d_B, and d_C) on the device (GPU).
     • Copy data from host to device using cudaMemcpy.
     • Launch the vectorAdd kernel with appropriate block and grid dimensions.
     • Copy the result back from the device to the host.
     • Print the result (output vector h_C).

3. Memory Allocation and Transfer:

   • We allocate memory for vectors on both the host and the device.
   • cudaMalloc allocates memory on the device.
   • cudaMemcpy transfers data between host and device.

4. Kernel Launch:

   • We launch the vectorAdd kernel using <<<numBlocks, threadsPerBlock>>> syntax.
   • numBlocks and threadsPerBlock determine the grid and block dimensions.

5. Clean Up:

   • We free the allocated device memory using cudaFree.
   • We also delete the host vectors (h_A, h_B, and h_C) to avoid memory leaks.

How to Compile and Run

1. Compile the Code:

   • Open your terminal or command prompt.
   • Navigate to the folder containing vector_addition.cu.
   • Compile the code using nvcc (NVIDIA CUDA Compiler):

     nvcc vector_addition.cu -o vector_addition
     

   

2. Run the Executable:

   • Execute the compiled binary:

     ./vector_addition
     

   • You'll see the result of vector addition printed to the console.