Switch to ParallelTestRunner.#79
Conversation
maleadt
force-pushed
the
tb/paralleltestrunner
branch
from
5f1c92f to
a83e1dc
Compare
christiangnrd
force-pushed
the
tb/paralleltestrunner
branch
from
0fc44bd to
4207091
Compare
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Drop the functional-backend autodetection: the CPU back-end is always tested, and GPU back-ends are opt-in via CLI flags (--cuda, --metal, etc.), in which case load or initialization failures propagate instead of being swallowed. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
christiangnrd
force-pushed
the
tb/paralleltestrunner
branch
from
4207091 to
27a0a21
Compare
|
I think I got it to a "ready" state. I reverted back to |
I really wanted to avoid that. AFAICT the only conflict was the SPIRV to LLVM translator, which I've made statically linked in JuliaPackaging/Yggdrasil#13945, so shouldn't cause conflicts between oneAPI.jl and OpenCL.jl anymore (and I'm also looking into switching oneAPI.jl to the LLVM SPIR-V back-end). |
I agree it's not ideal. Another option is that we require the backends to be manually added and take care of it in the CI files. Or we wait a bit until the conflicts are gone? |
They should be gone already. I've pushed a commit to try out. |
|
Looks good! |
2 participants
No description provided.