Thyre wrote: > Depending on the tool workflow, this wrong information might lead to memory > leaks and double-frees. To handle this nested sequence of data-op callbacks, > we would need a stack for the `HostOpId`. If we know, that it will never be > more than two levels, we could cheat and make it a static-sized thread-local > stack.
I fully agree that this can cause issues. For this particular case, https://github.com/llvm/llvm-project/pull/200202 should resolve the nested `transfer_to_device` event. Having a stack for the `HostOpId` is probably the safest option going forward though, since we might encounter such a case again in the future. https://github.com/llvm/llvm-project/pull/194168 _______________________________________________ cfe-commits mailing list [email protected] https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
