It’s really hard to design APIs that: 1. Encourage fine-grained multicore parallelism 2. Don’t force any threading model on end users or create threads behind the user’s back 3. Are simple and easy to use
-
Show this thread
-
Replying to @pcwalton
Honestly is data parallelism on the CPU ever *really* what you want? :)
2 replies 0 retweets 0 likes -
Replying to @Gok
Well, in my case, yes, because (a) I want to pipeline between CPU and GPU; (b) I have to work on older GPUs or bad drivers that don’t expose working compute shader.
2 replies 0 retweets 0 likes
I do want to move most of it to the GPU eventually though as an opt-in for devices where that will result in a speedup.
9:04 PM - 2 Apr 2019
0 replies
0 retweets
0 likes
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.