ClimateMachine.jl

ClimateMachine.jl copied to clipboard

Description

Upgrade to Julia 1.6 🚀. This was all I needed to run the test suite locally!

• [x] Code follows the style guidelines OR N/A.
• [x] Unit tests are included OR N/A.
• [x] Code is exercised in an integration test OR N/A.
• [x] Documentation has been added/updated OR N/A.

Looks like there is an issue with the binary NetCDF JLL? I think we are still waiting on getting 1.6 installed on the cluster but this is promising!

Looks like there is an issue with the binary NetCDF JLL? I think we are still waiting on getting 1.6 installed on the cluster but this is promising!

Yeah! Ah, good point, I didn't try the docs.

bors try

try

Build failed:

• buildkite/climatemachine-ci

ModuleCmd_Load.c(213):ERROR:105: Unable to locate a modulefile for 'julia/1.6'

I’m guessing that is not the right module name (try Julia/1.6.0)

On Thu, Mar 25, 2021 at 7:52 PM Valentin Churavy @.***> wrote:

ModuleCmd_Load.c(213):ERROR:105: Unable to locate a modulefile for 'julia/1.6'

— You are receiving this because your review was requested. Reply to this email directly, view it on GitHub https://github.com/CliMA/ClimateMachine.jl/pull/2146#issuecomment-807871988, or unsubscribe https://github.com/notifications/unsubscribe-auth/AAI2VJXIPXE2HPFDA54BSN3TFPSEZANCNFSM4Z2D3INQ .

bors try

try

Build failed:

• format

It looks like there is an issue combining StaticArrays.ll with StaticNumbers.jl in 1.6+, ex: https://github.com/CliMA/ClimateMachine.jl/blob/master/src/Numerics/DGMethods/remainder.jl#L431

Is StaticNumbers necessary if 1.6 has improved constant propagation when used with KA?

I'll try removing StaticNumbers, but idk the best way to ensure that the performance hasn't changed.

Also getting a segfault: https://buildkite.com/clima/climatemachine-ci/builds/1563#4746f8ba-e1c2-4628-a860-fddb19314844/95-107

bors try

try

Build failed:

• format

It looks like the next thing to try is to bump the version of KA / CUDA / GPUCompiler (ex. GPUCompiler is at 0.8.3 and the latest is 0.10.0 and all the 1.6 fixes were included in the 0.9+ releases).

I suspect the CPU test segfault is an OOB error, running locally to check.

It looks like the advection / diffusion test segfault is not an OOB error (testing locally with --check-bounds=yes with 1.5.4) so something else is triggering that failure.

bors try

try

Build failed:

• buildkite/climatemachine-ci

CUDA.jl needs to be updated along with GPUCompiler.jl

I'll give that a try, thanks!

I just registered KernelAbstractions 0.6, but I would hold off on moving to that. KernelAbstraction 0.5.X should work well on Julia 1.6

bors try

try

Build failed:

• buildkite/climatemachine-ci

bors try

try

Build failed:

• buildkite/climatemachine-docs

It would be nice to bump Cassette as well to https://github.com/JuliaLabs/Cassette.jl/releases/tag/v0.3.5 for the improved stacktraces as pointed out by @vchuravy

It would be nice to bump Cassette as well to https://github.com/JuliaLabs/Cassette.jl/releases/tag/v0.3.5 for the improved stacktraces as pointed out by @vchuravy

It looks like we're already on Cassette 0.3.5

It looks like if I ] add Cassette; up Cassette; rm Cassette, a few deps update, though, most notably GPUArrays (and Compat)

bors try

try

Build failed:

• buildkite/climatemachine-ci

bors try