This is cool, from someone who has written a lot of optimized, vectorized code. Someone put a lot of effort into this. Looks like it has been awhile since it w

This is cool, from someone who has written a lot of optimized, vectorized code. Someone put a lot of effort into this. Looks like it has been awhile since it was updated. It also only uses SSE? No AVX? Writing it all in assembly is cool, but abusing C++ templates and intrinsics would save you so much time and code, especially once you start targeting multiple instruction sets. No benchmarks against other vectorized libraries? If you need optimized math libraries for Linux, and you are running on Intel hardware, take a look at Intel IPPS and MKL. Both are free and collectively offer more functionality and performance. I suspect these libraries do reasonable things on AMD hardware nowadays too, (at least execute the SSE code path)