search

scijs / contributing

Contribution guide lines
10 stars 0 forks source link

Faster BLAS functionality #1

Open mikolalysenko opened 9 years ago

mikolalysenko commented 9 years ago

Currently, we have ndgemm which implements the corresponding BLAS function but it isn't very fast yet. This code could use some attention/optimizations, or alternatively a rewrite with some better approach.

rreusser commented 9 years ago

A benchmark suite, perhaps? In the spirit of modularization, maybe the right answer is to keep it test-driven, modularized, and benchmarked so that the first goal is to get things implemented, after which the components can be swapped with more robust, optimized implementations.

mikolalysenko commented 9 years ago

That would be great!

rreusser commented 9 years ago

Also, it's not totally clear to me whether the goal should be to move toward porting existing GSL/BLAS-like code via emscripten or whether the goal should be plain vanilla implementations. Oh wait, this is happening now :smile:

mikolalysenko commented 9 years ago

Yeah, but the proposal ended up getting rejected due to lack of slots. The jQuery org only got like 3 out of a pool of something like 200 proposals, and so their core projects ended up getting priority over SciJS :(

So probably not gonna happen unfortunately...

rreusser commented 9 years ago

Ah, I see. Yeah, wasn't sure if it was accepted but was at least surprised/pleased to see such a recent date. It means there's interest and motivation. I'm more than willing to still try to make things happen one way or another. I promise I'll stop with the talk ASAP and start coding some algorithms…