It is slow in Microsoft

[AlbertoCasasOrtiz]

It could be because of the used compiler: Visual Studio 2019.

[AlbertoCasasOrtiz]

News: Previously I though that the problem was the use of Eigen, whose inline functions can cause incompatibilities with msvc compiler. After checking the code again, looks like Eigen is not the problem.

Measuring times in the code I've found that most of the time is spent in the function AppendChunk (When creating the sparse matrix). Looks like the problem is the Microsoft implementation of malloc and realloc.

Link 1: Explanation of why realloc function could be slower in microsoft that in Linux/Mac.
Link 2: Microsoft developer explaining why malloc can be slower in microsoft.
Link 3: Another explanation in stackoverflow, more detailed.

[AlbertoCasasOrtiz]

After searching for alternatives, looks like the problem is the compiler implementation of those functions. We have to wait till another compiler is supported like MinGW.

Refer to issue #60.

[jasondk]

This doesn't sound like a satisfactory explanation to me. I'll take a look at AppendChunk(), but it seems to me we could pre-allocate larger chunks of memory to use and release the unused portion at the end (or re-allocate additional memory, if needed, a small number of times by allocating with a large chunk size). I assume the only reason we are iteratively allocating memory is because the sparsity pattern is unknown until we compute the matrix. I suspect an algorithmic improvement here could be useful...