For another on-going project, I needed to do 64-bits x 64-bits to 128-bits multiply, and I needed it in Python. When doing numerical work with Python, I always use NumPy which is a really awesome library. Unfortunately NumPy is heavily biased towards hardware native types, which makes sense for efficiency …

I recently came across an interesting problem at work: how to efficiently transpose a non-square matrix in-place?

One of my colleague is working on optimizing a six-steps FFT on our manycore processor, the MPPA-256. The six-steps FFT algorithm has a lot of nice properties, especially regarding highly parallel systems, but …