4 ms·

I'm trying very hard to follow along but this quickly devolves into language that isn't easy to follow. When he starts going into combining "bitonic globs" is w

I'm trying very hard to follow along but this quickly devolves into language that isn't easy to follow. When he starts going into combining "bitonic globs" is when I start having trouble focusing. For someone that's familiar with parallel programming topics and language, is this talk easy to follow? I'm trying to figure out if the fault lies with myself or the presenter. The vibe I'm getting is that showing slides full of code in an obscure language isn't a good teaching method, but I also don't have the prerequisite background here so my opinion isn't useful. I also don't do work that involves these classes of problems.

diskcat 11y ago
I could see the gist of the procedure i.e. representing it as globs which allow you to break up the data into 2 so you can run it in binary tree like steps. But the code was pretty useless. I didn't know that language and within the short time span that it was presented I couldn't understand the code. Clearly this would be better presented as a paper so you could have time to go through it but I think if you ignore the actual code completely I still understood (or at least i think i do) the talk.
bjornsing 11y ago
> For someone that's familiar with parallel programming topics and language, is this talk easy to follow? I'm trying to figure out if the fault lies with myself or the presenter. I think it was easy to follow if you focused on the aspects of general interest (= what he was doing in principle, e.g. divide-and-conquer). But sure, if you wanted to understand exactly how the merging of "bitonic globs" worked, then no, that's not easy to follow. I remember thinking when he started talking about merging the "bitonic blobs" that "Yea, yea, sure, you can divide-and-conquer, but it's just about as much work combining two of those partial solutions as solving the original problem! There's got to a be a better way...", and I basically stopped listening to the details.

diskcat 11y ago

I could see the gist of the procedure i.e. representing it as globs which allow you to break up the data into 2 so you can run it in binary tree like steps. But the code was pretty useless. I didn't know that language and within the short time span that it was presented I couldn't understand the code. Clearly this would be better presented as a paper so you could have time to go through it but I think if you ignore the actual code completely I still understood (or at least i think i do) the talk.

bjornsing 11y ago

> For someone that's familiar with parallel programming topics and language, is this talk easy to follow? I'm trying to figure out if the fault lies with myself or the presenter. I think it was easy to follow if you focused on the aspects of general interest (= what he was doing in principle, e.g. divide-and-conquer). But sure, if you wanted to understand exactly how the merging of "bitonic globs" worked, then no, that's not easy to follow. I remember thinking when he started talking about merging the "bitonic blobs" that "Yea, yea, sure, you can divide-and-conquer, but it's just about as much work combining two of those partial solutions as solving the original problem! There's got to a be a better way...", and I basically stopped listening to the details.