Most of weighted norms are broken

CliMA / ClimateMachine.jl

Climate Machine: an Earth System Model that automatically learns from data

https://clima.github.io/ClimateMachine.jl/latest/

Other

452 stars 78 forks source link

Most of weighted norms are broken #718

Open mwarusz opened 4 years ago

mwarusz commented 4 years ago

All weighted norms are broken when the dims argument is specified. It seems that mapreduce doesn't like lazy broadcasts when this argument is specified, on either CPU or GPU.

Even without specifying dims, only the l_2 norm works on GPU. Others (l_1, l_3, etc.) fail with errors similar to https://github.com/JuliaGPU/CuArrays.jl/issues/495.

This is on 1.3.1

blallen commented 4 years ago

Was #717 enough to close this or do we need to also close out #590?

charleskawczynski commented 3 years ago

I think the CUDA issue needs to be resolved