Fork me on GitHub
#uncomplicate
<
2016-05-10
>
carocad09:05:11

hey @blueberry , might I ask you how is it possible for neanderthal to be so much faster than the equivalent JBLAS? I mean they use the same ATLAS technology on the background right? then what makes them so different? o.O

blueberry09:05:28

@carocad Every cook has some secrets to his recipe simple_smile

blueberry09:05:42

But I don't - I just researched in detail various aspects of calling native libraries and working with them. Also wrote by hand some things that people usually use a (bad) code generator for.

carocad11:05:31

@blueberry well my deepest congratulations to the chef 😉 very tasty !! I thought that pure Java was as fast as Clojure could get but I realize that I was gracefully wrong 😄

blueberry13:05:50

@carocad Thanks. One delicious course that might also be to your liking is that the next release (probably in a week or two) comes with a GPU engine that works on all major GPUs (AMD, Nvidia, Intel, AND Mac) and is more than 1000x faster than *optimized* java for large matrix multiplication.

carocad17:05:16

@blueberry wow !! awesome. Just out of curiosity: are you planning to support the core.matrix API?

blueberry18:05:34

I do not need that, so I guess that depends on the clojurians that do.

carocad21:05:27

I thought so. Even so it is still an awesome project 😄. I opened an issue in clatrix repo to let them know about your advances. It will up to them to adopt your library.