Page MenuHomec4science
Files F60785995

vector_add.cu
No OneTemporary
Actions

File Metadata

Created
Thu, May 2, 14:23

vector_add.cu
View Options

#include <cuda_runtime.h>
#include <random>
#include <vector>
#include <iostream>
#include <exception>
/**
* TODO: write a kernel that does the vector addition C = A + B with 1 thread
*/
__global__ void vectorAddOneThread(const float *A, const float *B, float *C,
int N) {
}
/**
* TODO: write a kernel that does the vector addition C = A + B with 1 Block
* and 256 threads Hint: When 256 threads are working on one loop how the loop
* changes?
*/
__global__ void vectorAddOneBlock(const float *A, const float *B, float *C,
int N) {
}
/**
* TODO: write a kernel that does the vector addition C = A+B with grid of
* blocks. Each block has 256 threads. Hint: what check do you need to implement
* to avoid invalid memory reference?
*/
__global__ void vectorAdd(const float *A, const float *B, float *C, int N) {
}
/* -------------------------------------------------------------------------- */
void checkResults(std::string test, const float *A, const float *B,
const float *C, int N) {
// Verify that the result vector is correct
for (int i = 0; i < N; ++i) {
if (std::abs(A[i] + B[i] - C[i]) > 1e-5) {
throw std::runtime_error("Result verification failed at element "
+ std::to_string(i) + " for test " + test);
}
}
}
/**
* Host main routine
*/
int main() {
// Print the vector length to be used, and compute its size
int N = 1 << 20; // 1M elements
size_t size_in_bytes = N * sizeof(float);
std::cout << "[Vector addition of " << N << " elements]" << std::endl;
float *d_A{nullptr};
float *d_B{nullptr};
float *d_C{nullptr};
// TODO: allocate d_A, d_B, and d_C
std::mt19937 gen(2006);
std::uniform_real_distribution<> dis(0.f, 1.f);
// Initialize the input vectors
for (int i = 0; i < N; ++i) {
d_A[i] = dis(gen);
d_B[i] = dis(gen);
}
// Launch the Vector Add CUDA Kernel
int threads_per_block = 256;
// TODO: Launch the Vector Add CUDA Kernel with one threads
vectorAddOneThread <<<?, ?>>> (d_A, d_B, d_C, N);
cudaDeviceSynchronize(); // Since kernel launches is async wrt to the host we
// have to syncronize
checkResults("vectorAddOneThread", d_A, d_B, d_C, N);
// TODO: Launch the Vector Add CUDA Kernel with one block and 256 threads
vectorAddOneBlock <<<?, ?>>> (d_A, d_B, d_C, N);
cudaDeviceSynchronize();
checkResults("vectorAddOneBlock", d_A, d_B, d_C, N);
int blocks_per_grid = ?; // TODO: compute the blocks per grid
// TODO: Launch the Vector Add CUDA Kernel with blocksPerGrid and 256 threads
vectorAdd <<<?, ?>>> (d_A, d_B, d_C, N);
cudaDeviceSynchronize();
checkResults("vectorAdd", d_A, d_B, d_C, N);
std::cout << "Test PASSED" << std::endl;
// TODO: Free device global memory
std::cout << "Done" << std::endl;
return 0;
}

Event Timeline

c4science · Help