Hacker News Reader

You can't parallelize a for-loop the way you can apply a function on every member of a vector. If code can be vectorized in Julia, that is what I'm interested in seeing. These samples are obviously not intended for distributed compute. Thousands of CUDA cores sit idle waiting for this synchronous for-loop to finish.

View on HN

Cython has a construct for parallelizing for loops.

http://docs.cython.org/src/userguide/parallelism.html

I wouldn't want to claim an informed opinion about whether it is a good style or not, but a block with no side effects isn't that far off from a function with no side effects.

@maxerickson I can't reply to you so I'll reply next to you. That looks very interesting. I will have to look into it. Thanks

Reply on news.ycombinator.com