One of my favorite topics lately is Python optimization. A few weeks ago FB put out a release on how they had done fast randomized SVD and in the implementations section they mentinoned using intels kernel math libraries for the BLAS libraries. Largely the result of this is many more functions are F-Contiguous rather than C-contiguous (in linear algebra, fortan obviously wins here). I had done something similar on a much smaller scale about 4 months ago. https://medium.com/@_devbob/from-0-to-warp-speed-b780a2bc36c...

Does anyone else have some good bits or blog entries on these sorts of inner workings and optimizations?