Ben Vanik

(note that there's a spectrum between "generic" and "host" - we can decide we want avx2 as a baseline in our generic config instead of emitting scalar code, etc -...

Oof - are these self-hosted runners on a potato? 2hr builds feels very high. We could reduce the surface area - disable expensive compilation targets (riscv, arm, etc) to start...

Ouch, yeah, that'll do it :P Can you attach a bigger disk and put the build there?

note that this may edge into non-default path territory - downloading/compiling cuda stuff for one particular board vs the more common arm systems is a big addition to the default...

Yeah, it's even trickier because we check for `NOT IREE_CUDA_AVAILABLE OR CMAKE_CROSSCOMPILING` for changing the default - so cross-compiling for jetson won't have CUDA I'm going to suggest such users...

The only thing IREE_CUDA_AVAILABLE does is change the default flags for compiler and runtime backend inclusion. I don't see why we'd want to change the default for all source builds...

I have no idea why ROCM got enabled by default on arm64 - that's silly :) We should have ROCM track what the default for CUDA is and then require...

I think regardless of choice we need to _not_ tie defaults to what the CI/bots/packaging systems do - those should _never_ use defaults. The big red flag here for me...

Sorry, sent without the last sentence: we can enable a lightweight rocm by default in source builds if everyone thinks that's best - I don't, but I'm just one vote...

The folder has been busted for 12mo it seems as it was never updated to support resource attrs. You may be hitting it now because something is folding a constant...