common

#pragma once
#include <iostream>
#include <cublas_v2.h>

struct launch_params
{
    cublasHandle_t cublasHandle = nullptr;
    cudaStream_t stream = 0;
    int threadsPerBlock = 512;
    bool pinnedMemory = false;
};

template <typename T>
inline void print_matrix1(const T *A, int m, int n)
{
    for(int i = 0; i < m; ++i)
    {
        for(int j = 0; j < n; ++j)
            std::cout << A[j * m + i] << " ";
        std::cout << '\n';
    }
    std::cout << '\n';
}

template <typename T>
struct Id {};

template <typename T>
struct Zero {};

template <typename T>
struct mId {};

template <typename T>
struct AlmostZero {};

// Shortcuts

template <typename T>
constexpr T Id_v = Id<T>::value;

template <typename T>
constexpr T Zero_v = Zero<T>::value;

template <typename T>
constexpr T mId_v = mId<T>::value;

template <typename T>
constexpr T AlmostZero_v = AlmostZero<T>::value;