JonChesterfield added a comment. There is a potential hazard here for parallel compilation and to a lesser extent testing. (I have just learned that) kfd imposes a process limit which is low enough that we may see hsa_init fail under multiple processes.
@jdoerfert you talked me out of embedding a map from pci.ids to architecture as we'd have to keep it up to date, and it can diverge from how the runtime libraries identify the hardware. I'm starting to think that's the lesser of two evils. Would probably look something like id = cat-somewhere-in-/sys switch(id) { default: return nullptr; case 0x67C0: case 0x67C1: case 0x67C2: return "gfx803"; ... // ~ 100 lines on this theme, picks up new entries when new hardware is released }; Repository: rG LLVM Github Monorepo CHANGES SINCE LAST ACTION https://reviews.llvm.org/D99949/new/ https://reviews.llvm.org/D99949 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits