Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

I don't think these are realistic. It does far worse than MKL in practice on Intel CPUs in my tests, especially if arrays are not huge.


MKL has a JIT similar to libxsmm for small sizes.


That's for very small matrices, but I forget the dimensions. It was only because of libxsmm that MKL got it. (You can use libxsmm to dispatch to OpenBLAS or BLIS for non-tiny cases.) BLIS has "small"-dimension support in many cases; I don't remember what they call it, but you can see changelog items about it.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: