Hi,

Runtime performances are a major issue in the project I am working on, and I wondered if falling back to the Intel MKL (for example) for some operations might not give me slightly better results (along with ICPC optimizations for instance)
The dev branch of eigen now allows a quite gracious fallback to MKL, but only for dynamic and large matrices, and the benchmarks also test dynamic matrices and vectors only.
The changes in my project would not necessarily be straightforward if I want to use MKL instead of/along eigen, so my question is (before I do anything useless):
- Is there any particular reason why the benchmarks don't compare fixed-size matrices operations ? (similar results ? In which case eigen seems significantly faster for small sizes...) I am particularly interested in small fixed-size matrices, such as 3x3 matrices and 3x1 vectors.

Thanks in advance,
Regards.

Don't bother, Intel MKL is very slow for small matrices.