[chm]

Yes, and having worked with Fortran code for years, we used the routines to calculate... and Python to analyze the data. A colleague even scrapped old Fortran routines in favor of Python + Numpy, as he could be more productive without a noticeable slowdown in his work, as the data analysis part was the bottleneck. So I reiterate my point: I like the project, I like the idea, but we need more information regarding the performance of the routines. Otherwise, using e.g. Python full-stack could be much less of a hassle.

[gaius]

NumPy is FORTRAN under the covers, e.g. BLAS IIRC.

[greglindahl]

BLAS are generally implemented in assembly these days, although they do have Fortran interfaces.

[gnufx]

The important kernels (particularly GEMM) are written in assembler, but you can see OpenBLAS, for instance, importing netlib LAPACK.