Valentin Churavy
Valentin Churavy
> So the idea is to decouple the back-ends from KA.jl, instead implementing KernelIntrinsics.jl? What's the advantage; do you envision packages other than KA.jl to build their kernel DSL on...
> Will KA/KI still be a greatest common denominator of the GPU backends The intrinsics proposed here are the greatest common denominator. I could see us adding some more intrinsics...
Replaced by #635
There hasn't been a release of Enzyme.jl in a while, so unlikely? Do you have an MWE?
The three PRs in v0.13.36 are: - #2350 (Fairly straight-forward since it just filters out some pre-header instructions) - #2352 (Show to be unrelated (and I would have been very...
Nothing weird is jumping at me https://github.com/JuliaDiff/DifferentiationInterface.jl/compare/DifferentiationInterface-v0.6.50...DifferentiationInterface-v0.6.52
Regarding it being not reproducing iin the repl: ``` u = copy(u₀) du = copy(du₀) fu = copy(u₀) nlequation = @closure (x, _) -> begin du_ = ifelse.(differential_vars, x, du)...
I suspect this would fail for Forward mode as well. My colleague @efaulhaber yesterday found this: ``` using OrdinaryDiffEqRosenbrock, ADTypes import Enzyme function f(du, u, p, t) du .= u...
@ChrisRackauckas the crux for me is ``` nlequation = @closure (x, _) -> begin du_ = ifelse.(differential_vars, x, du) u_ = ifelse.(differential_vars, u, x) f(du_, u_, p, t) end ```...
Like: ``` using FastClosures, Enzyme t = 0.0 p = [0.04, 3e7, 1e4] u₀ = [1.0, 0, 0] du₀ = [-0.04, 0.04, 0.0] tspan = (0.0, 100000.0) differential_vars = [true,...