C++ Eigen::Map class questions

Question

The Eigen::Map documentation here (https://eigen.tuxfamily.org/dox/classEigen_1_1Map.html) mentions the following: This class represents a matrix or vector expression mapping an existing array of data. It can be used to let Eigen interface without any overhead with non-Eigen data structures, such as plain C arrays or structures from other libraries.

• What does without any overhead mean? Does this mean that if I have an Eigen::VectorXd and I want to map it to a std::vector<double> or vice versa, then since the data type is the same, we can use the same memory location for both objects?

In order to test this, I ran the following code snippet.

Eigen::VectorXd eigenVector (5);
eigenVector << 1, 1, 1, 1, 1;

std::vector<double> stdVector1 {eVector.data(), eVector.data() + eVector.size()};

std::vector<double> stdVector2 (5);
Eigen::VectorXd::Map(&stdVector2[0], eVector.size()) = eVector;

Then I printed the addresses of eigenVector, stdVector1 and stdVector2. All of them are different.

• What is exactly going on behind the scenes with the last line above? : Eigen::VectorXd::Map(&stdVector2[0], eVector.size()) = eVector; Is the presence of an '=' leading to a copy operation to another memory location? Is there another way to use Eigen::Map that would be more efficient here?

• Similar question for the other way round: Mapping a std::vector<double> to Eigen::VectorXd. I used the following line for this - Eigen::Map<Eigen::VectorXd> eVector2(&stdVector1[0], stdVector1.size()); Same story here, eVector2 is a different memory location than stdVector1.

• Given the above, is Eigen::Map even useful for interfacing with raw data types? To me, it is equivalent to using something like std::transform instead.

Answer 1

I would not count exactly on the highlighted statement. The Eigen and std-vectors sure have the same contigouus memory layout. But of course you need a little overhead, such as storing the size of the Eigen vector. And already by this you''ll have different memory addresses.

Let's consider your code in detail:

Eigen::VectorXd eigenVector (5);
eigenVector << 1, 1, 1, 1, 1;

std::vector<double> stdVector1 {eVector.data(), eVector.data() + eVector.size()};

Here, you set up a new vector stdVector1 and copy the content of the eigenVector to it. There is no way that the .data() members could point to the same memory address-.

std::vector<double> stdVector2 (5);
Eigen::VectorXd::Map(&stdVector2[0], eVector.size()) = eVector;

Here you construct a vector stdVector2 which again has an own memory range to hold five doubles.

By Eigen::VectorXd::Map(&stdVector2[0], eVector.size()) you construct a temporary Map-object and let its data refer to the data of stdVector2. But in the next step, with the assignment operator, you reset the reference to the data of eVector. Notably, you do not change any parameter of stdVector2.

Summarizing, all memory locations of your vector are different.