Re: [eigen] Parallel matrix multiplication causes heap allocation

[François Fayard <fayard@xxxxxxxxxxxxx>]

> Interesting. And that confirms that we really should update our own benchmarks (and probably run blaze's benchmark suite for comparison, to see if it measures differently).
> Can anyone spare his (up-to-date) CPU for a night?

I think that we should first define what kind of computer should be considered for benchmarking. I would suggest:
- A laptop: core i7, 4 cores, AVX2
- A workstation: Dual Xeon, 2x14 cores, AVX2
- Optional: A Knights Landing
The 14 cores option is the best seller right now (E5-2660, E5-2680 or E5-2690) for clusters.

I assume that it's fairly easy to get a laptop. For the workstation, I have one which is available with CentOS 7.3, and the latest Intel Parallel Studio on it. I don't know how to give access to it though as I am behind a Livebox at work. If you can help me with that, I can give you access to it.
If you are interested there is also a Titan X Pascal on it.

For the KNL, I have access to one of them but the performance is awful compared to what I had this summer. So I would not trust it.

But are you really ready to show that the MKL is giving better performance than the internal BLAS of Eigen?

François