我有下面的例子：

#include <unsupported/Eigen/CXX11/Tensor>
#include <Eigen/Core>

Eigen::Tensor<double, 5> test_tensor(3,3,2,1,1);
test_tensor.setValues({
  {{{{1.1}},{{1.1}}},{{{0}},{{0}}},{{{0}},{{0}}}},
  {{{{0}},  {{0}}},  {{{1}},{{1}}},{{{0}},{{0}}}},
  {{{{0}},  {{0}}},  {{{0}},{{0}}},{{{1}},{{1}}}}
});

//use chip and slice to compute the subtensor sum
template <typename TensorType>
auto tensor_sum(const TensorType& tensor) -> typename TensorType::Scalar {
  using T = typename TensorType::Scalar;
  T sum = 0; //provisory
  for (int i = 0; i < tensor.size(); ++i) {
    sum += tensor.data()[i];
  }
  return sum;
}

Eigen::Tensor<double, 3> field_slice;
for (int l = 0; l < 3; ++l) {
  for (int m = 0; m < 3; ++m) {
    auto field_slice_ = test_tensor.chip(1, m).chip(0, l);
    field_slice = field_slice_.slice(
                    Eigen::array<Eigen::Index, 3>({0, 0, 0}),
                    Eigen::array<Eigen::Index, 3>({2, 1, 1}));
    std::cout << "sum slice " << l << " " << m << " " << tensor_sum(field_slice) << std::endl;
  }
}

// use nested loops to compute subtensor sum
double sum;
for (int l = 0; l < 3; ++l) {
  for (int m = 0; m < 3; ++m) {
    sum = 0;
    for (int i = 0; i < 2; ++i) {
      for (int j = 0; j < 1; ++j) {
        for (int k = 0; k < 1; ++k) {
          sum += test_tensor(l, m, i, j, k);
        }
      }
    }
    std::cout << "sum nested loops " << l << " " << m << " " << sum << std::endl;
  }
}

打印出来

sum slice 0 0 0
sum slice 0 1 0
sum slice 0 2 1.1
sum slice 1 0 1
sum slice 1 1 1
sum slice 1 2 1.1
sum slice 2 0 1
sum slice 2 1 1
sum slice 2 2 1.1
sum nested loops 0 0 2.2
sum nested loops 0 1 0
sum nested loops 0 2 0
sum nested loops 1 0 0
sum nested loops 1 1 2
sum nested loops 1 2 0
sum nested loops 2 0 0
sum nested loops 2 1 0
sum nested loops 2 2 2

为什么结果会不一样？我怀疑chip或slice操作没有按预期的方式工作。通过首先存储切片Tensor然后创建切片来分离这两个步骤也没有改变结果。我如何计算我的子Tensor的和而不使用嵌套循环？