JVM High Performance Linear Algebra

Hi all,

I have just discovered this platform and I’m very impressed that you have created a very clear business model with an Open Source philosophy.

I am the author of https://github.com/fommil/netlib-java which powers high performance linear algebra on the JVM and allows JVM applications to use system optimised natives (such as the Intel Math Kernel Library). I recently gave a talk on this topic at ScalaX which gives details on the incredible performance difference (linked from the project page).

Netlib-java is the backend of Apache Spark and there is a higher level Java interface via the matrix-toolkits-java project.

Of course, the first thing I looked at in your OG-Platform repository was how you do linear algebra and I was surprised to see that you have ported the functions you need into pure Java and have eliminated the possibility of using native backends.

I’m sending this message simply to raise awareness of netlib-java. From my perspective, it makes sense if we all use the same low-level libraries when doing our calculations as it means performance improvements can happen in the one place.

I also have plans to create additional projects like netlib-java for other functions that can be enhanced by hardware: random number generation is next on the list (along with an APU backend for matrix operations). I’d be delighted to hear from anybody with an interest in this area. If I don’t respond on this thread, the github issue tracker is a perfectly good place to find me.

Of course, this is all free software so you don’t need to contact me to use it (even commercially) :smile: I’m happy to hear about any framework that takes it onboard.

Best regards,

Thanks for letting us know about the project. We also have a similar project in that space, OG-Maths which may be of interest.