Google has hired R. M. Larsen, who wrote PROPACK[1], and uses it internally. A heck of a lot of computation is still powered by Fortran. Ever used SciPy / NumP

Google has hired R. M. Larsen, who wrote PROPACK[1], and uses it internally. A heck of a lot of computation is still powered by Fortran. Ever used SciPy / NumPy? Deep down, FORTRAN powers it, via BLAS/LAPACK[2] A few years back in Cadence, we were still running a lot of FORTRAN code for computational lithography. Not only it's fast (and parallelizable), there are many algorithms that had been fine-tuned over the many years that you can reuse. Say, we needed to pick a gradient descent implementation for an algorithm that used it. Guess what language the L-BFGS[3] implementation we ended up using was written in - no need, it's FORTRAN, from the year 1983. [1] http://sun.stanford.edu/~rmunk/PROPACK/ http://sun.stanford.edu/~rmunk/PROPACK/ [2] https://stackoverflow.com/questions/7496547/does-python-scipy-need-blas https://stackoverflow.com/questions/7496547/does-python-scip... [3] https://en.wikipedia.org/wiki/Limited-memory_BFGS https://en.wikipedia.org/wiki/Limited-memory_BFGS