如何用可变参数函数覆盖 C++ 类中的运算符? [英] How to overwrite operator in C++ class with a variadic function?
问题描述
C++ 新手这里:我想创建一个模板类来创建不同数据类型和 d 维度的张量，其中 d 由形状指定.例如，形状为 (2, 3, 5) 的张量有 3 个维度，包含 24 个元素.我使用一维向量存储所有数据元素，并希望使用形状信息访问元素以查找元素.
C++ newbie here: I want to create a template class to create tensors of different data types and d dimensions, where d is specified by a shape. For example, a tensor with shape (2, 3, 5) has 3 dimensions holding 24 elements. I store all data elements using a 1d vector and want to access elements using the shape information to find the elements.
我想覆盖 () 操作符来访问元素.由于维度可能会有所不同，() 运算符的输入参数数量也会有所不同.从技术上讲，我可以使用向量作为输入参数，但 C++ 似乎也支持可变参数函数.但是，我无法将头环绕.
I would like to overwrite the () operator to access the elements. Since the dimensions can vary, so can the number of input parameters for the () operator. Technically, I can use a vector as input parameter but C++ also seems to support variadic functions. However, I cannot wrap my head around it.
到目前为止我所拥有的:
What I have so far:
#ifndef TENSOR_HPP
#define TENSOR_HPP
#include <vector>
#include <numeric>
#include <algorithm>
#include <stdexcept>
#include <iostream>
#include <stdarg.h>
template <typename T> class Tensor {
private:
std::vector<T> m_data;
std::vector<std::size_t> m_shape;
std::size_t m_size;
public:
// Constructors
Tensor(std::vector<T> data, std::vector<std::size_t> shape);
// Destructor
~Tensor();
// Access the individual elements
T& operator()(std::size_t&... d_args);
};
template <typename T> Tensor<T>::Tensor(std::vector<T> data, std::vector<std::size_t> shape) {
// Calculate number of data values based on shape
m_size = std::accumulate(std::begin(shape), std::end(shape), 1, std::multiplies<std::size_t>());
// Check if calculated number of values match the actual number
if (data.size() != m_size) {
throw std::length_error("Tensor shape does not match the number of data values");
}
// All good from here
m_data = data;
m_shape = shape;
}
template <typename T> T& Tensor<T>::operator() (std::size_t&... d_args) {
// Return something to avoid warning
return m_data[0];
};
template <typename T> Tensor<T>::~Tensor() {
//delete[] m_values;
};
#endif
不，当我执行以下操作时:
No when I do the following:
std::vector<float> data = {1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24};
std::vector<std::size_t> shape = {2, 3, 4};
Tensor<float> tensor(data, shape);
tensor(2,0,3); // <-- What I would like to do
// Possible workaround with vector which I would like to avoid
// std::vector<std::size_t> index = {2,0,3};
// tensor(index);
我收到错误:
tensor2.hpp:27:33: error: expansion pattern ‘std::size_t&’ {aka ‘long unsigned int&’} contains no parameter packs
使用可变参数函数覆盖 () 运算符的正确方法是什么?
What is the correct approach to overwrite the () operator using a variadic function?
推荐答案
通过提供形状"作为模板参数，你可以这样做:
By providing "shape" as template parameter, you might do:
// Helper for folding to specific type
template <std::size_t, typename T> using always_type = T;
// Your Tensor class
template <typename T, std::size_t... Dims>
class MultiArray
{
public:
explicit MultiArray(std::vector<T> data) : values(std::move(data))
{
assert(values.size() == (1 * ... * Dims));
}
const T& get(const std::array<std::size_t, sizeof...(Dims)>& indexes) const
{
return values[computeIndex(indexes)];
}
T& get(const std::array<std::size_t, sizeof...(Dims)>& indexes)
{
return values[computeIndex(indexes)];
}
const T& get(always_type<Dims, std::size_t>... indexes) const
{
return get({{indexes...}});
}
T& get(always_type<Dims, std::size_t>... indexes)
{
return get({{indexes...}});
}
static std::size_t computeIndex(const std::array<std::size_t, sizeof...(Dims)>& indexes)
{
constexpr std::array<std::size_t, sizeof...(Dims)> dimensions{{Dims...}};
size_t index = 0;
size_t mul = 1;
for (size_t i = dimensions.size(); i != 0; --i) {
assert(indexes[i - 1] < dimensions[i - 1]);
index += indexes[i - 1] * mul;
mul *= dimensions[i - 1];
}
assert(index < (1 * ... * Dims));
return index;
}
static std::array<std::size_t, sizeof...(Dims)> computeIndexes(std::size_t index)
{
assert(index < (1 * ... * Dims));
constexpr std::array<std::size_t, sizeof...(Dims)> dimensions{{Dims...}};
std::array<std::size_t, sizeof...(Dims)> res;
std::size_t mul = (1 * ... * Dims);
for (std::size_t i = 0; i != dimensions.size(); ++i) {
mul /= dimensions[i];
res[i] = index / mul;
assert(res[i] < dimensions[i]);
index -= res[i] * mul;
}
return res;
}
private:
std::vector<T> values; // possibly: std::array<T, (1 * ... * Dims)>
};
用法类似于
std::vector<float> data = {1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24};
MultiArray<float, 2, 3, 4> tensor(data);
std::cout << tensor.get(1, 0, 3); // 16
这篇关于如何用可变参数函数覆盖 C++ 类中的运算符?的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！
