AlexVlx wrote: > Looks like this is still using magic globals rather than intrinsics.
I replied to this a few days ago, but either I didn't press the right button or messed something else up, apologies. At any rate, yes, for now it is still using magic globals since we've had this working for some time internally and also for AMDGCN SPIR-V uses we can rely on pass ordering to ensure no optimisations muck things up. Unless you consider it crucial to switch to intrinsics from the get-go, I would rather postpone that for later, after adding the full-fat diagnostics, possibly as part of generalising the functionality to other targets, if they are interested. This is mostly to minimise risk / go with something that we know works (and piping intrinsics through SPIR-V is slightly more tricky), and because I'm rather certain we can do the switch transparently i.e. users of the feature wouldn't know anything happened. https://github.com/llvm/llvm-project/pull/134016 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
