Open emprice opened 2 years ago
KernelAbstractions.jl cannot generate kernels with broadcast expressions in them.
You might need to help me a little here, @ChrisRackauckas. du
and u
are the same size, so there shouldn't be any broadcasting (from what I usually think of as broadcasting, anyway). And how is that related to the number of trajectories?
At 1 trajectory it has a fallback to just serial.
du and u are the same size, so there shouldn't be any broadcasting (from what I usually think of as broadcasting, anyway)
Based on the DiffEq documentation, I've written this minimal example. If I set
trajectories = 1
, it compiles properly and everything is fine. But whentrajectories = 32
, say, I get errors relating to callingoverdub
from within the kernel. I'm very much a Julia noob, but this seems weird to me. The CUDA debugging settings aren't really helping shed any light on the situation. I've been working on this all day and can't seem to make forward progress. Hoping to get some help here!Example code:
Example error: