jhuber6 wrote: > Why not stick with `--offload-arch` for generic use? I'm not really sure > about the motivation.
This is the main problem I have with a lot of the offloading infrastructure. As far as I see it `--offload-arch=` is intended to behave like the `-mcpu` value but it's pulling double duty by also using *known* CPU names to *infer* the toolchain triple. We can still *do* that, I'm not planning on changing it, but I think it makes the infrastructure much easier to build on if we have a generic way to create an offloading toolchain in the same vein of `--target=`. I've had qualms with this since we do things like `--offload-arch=amdgcnspirv,gfx942` which implicitly creates two separate toolchains with incompatible triples. I'm open to suggestions however. https://github.com/llvm/llvm-project/pull/146594 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
