Thyre wrote: I agree that these nested events can become complicated for a tool to deal with. On the host side (which we're adding here), I think this is fine. The device tracing records, not yet implemented in LLVM, are more complicated.
If the "slow path" is replaced by something not calling `omp_target_memcpy`, the `target_data_op` callbacks will vanish. https://github.com/llvm/llvm-project/pull/194168 _______________________________________________ cfe-commits mailing list [email protected] https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
