epita/notes-ing2
1
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

gpu

Browse source
This commit is contained in:
JOLIMAITRE Matthieu 2024-03-28 17:58:33 +01:00
parent 8b3bb9c382
commit d976cfaf74
37 changed files with 2669 additions and 371 deletions
Download patch file Download diff file Expand all files Collapse all files

96
gpu/tp2/c/src/Matrix.h
View file

@ -1,14 +1,16 @@
#pragma once
#include <vector>
#include <iostream>
#include <vector>
#define CUDA_CHECK(code) { cuda_check((code), __FILE__, __LINE__); }
inline void cuda_check(cudaError_t code, const char *file, int line) {
if(code != cudaSuccess) {
std::cout << file << ':' << line << ": [CUDA ERROR] " << cudaGetErrorString(code) << std::endl;
std::abort();
}
#define CUDA_CHECK(code) \
{ cuda_check((code), __FILE__, __LINE__); }
inline void cuda_check(cudaError_t code, const char* file, int line) {
if (code != cudaSuccess) {
std::cout << file << ':' << line << ": [CUDA ERROR] "
<< cudaGetErrorString(code) << std::endl;
std::abort();
}
}
namespace linalg {
@ -16,55 +18,53 @@ namespace linalg {
//
// Generic matrix of type T (int, float, double...)
//
template<typename T>
class Matrix
{
public:
// construct matrix, allocate the 2D pitched memory on the device
__host__ Matrix(int rows, int cols);
template <typename T> class Matrix {
public:
// construct matrix, allocate the 2D pitched memory on the device
__host__ Matrix(int rows, int cols);
// free allocated device memory
__host__ void free();
// free allocated device memory
__host__ void free();
public:
// copy values from host std::vector to device Matrix
// values must be a vector of size rows x cols
// allocation is already done in the constructor
__host__ void to_cuda(const std::vector<T>& values);
public:
// copy values from host std::vector to device Matrix
// values must be a vector of size rows x cols
// allocation is already done in the constructor
__host__ void to_cuda(const std::vector<T>& values);
// copy values from device Matrix to host std::vector
// values may not ne resized
__host__ void to_cpu(std::vector<T>& values) const;
// copy values from device Matrix to host std::vector
// values may not ne resized
__host__ void to_cpu(std::vector<T>& values) const;
public:
// accessor at row i and column j
__device__ const T& operator()(int i, int j) const;
__device__ T& operator()(int i, int j);
public:
// accessor at row i and column j
__device__ const T& operator()(int i, int j) const;
__device__ T& operator()(int i, int j);
public:
__host__ Matrix operator + (const Matrix<T>& other) const;
__host__ Matrix operator - (const Matrix<T>& other) const;
__host__ Matrix operator * (const Matrix<T>& other) const;
__host__ Matrix operator / (const Matrix<T>& other) const;
public:
__host__ Matrix operator+(const Matrix<T>& other) const;
__host__ Matrix operator-(const Matrix<T>& other) const;
__host__ Matrix operator*(const Matrix<T>& other) const;
__host__ Matrix operator/(const Matrix<T>& other) const;
private:
// apply binary functor f on all pairs of elements
// f must provide the following operator
//
// T operator()(T a, T b)
//
// template<typename BinaryFunctor>
// __host__ Matrix apply(const Matrix<T>& other, BinaryFunctor&& f) const;
private:
// apply binary functor f on all pairs of elements
// f must provide the following operator
//
// T operator()(T a, T b)
//
// template<typename BinaryFunctor>
// __host__ Matrix apply(const Matrix<T>& other, BinaryFunctor&& f) const;
public:
__host__ __device__ inline int rows() const {return m_rows;}
__host__ __device__ inline int cols() const {return m_cols;}
public:
__host__ __device__ inline int rows() const { return m_rows; }
__host__ __device__ inline int cols() const { return m_cols; }
private:
T* m_data_ptr; // device pointer
int m_rows;
int m_cols;
size_t m_pitch;
private:
T* m_data_ptr; // device pointer
int m_rows;
int m_cols;
size_t m_pitch;
};
} // namespace linalg