elizabethandrews wrote: > Does the change to default the target to `spirv64-unknown-unknown` work for > both cc1 `-fsycl-is-device` and driver `-fsycl-device-only` invocations? If > so, two thoughts:
Yes. The driver invocation sets the triple as spirv64-unknown-unknown and actually also passes `-fsycl-is-device` to the frontend. > * I think a driver test would be appropriate; perhaps one like the existing > `cuda-device-triple.cu` that validates both default and explicit target > selection. I just realized I did not add a test for explicit target selection. I will do so shortly. >* Should we remove the explicit target for most of the tests since they match >the default anyway? > Perhaps we should also update existing tests that specify `spir64` as the > device target to use `spirv64-unknown-unknown` (or rely on the default) for > consistency? Unless we document this and insist on it for future code reviews (which IMO is not required), the consistency is not going to last. I don't think think we need to mandate a specific triple because functionally none of it is wrong. https://github.com/llvm/llvm-project/pull/172366 _______________________________________________ cfe-commits mailing list [email protected] https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
