weliveindetail wrote: Alright. The above change makes it as cheap as it gets with an abstraction: virtual function calls only on init and one function pointer call for each processed statement.
> I was wondering if we can extend that functionality via a callback mechanism How would that look like? I assume the runtime interface will have other functions than `getCall()` in the future right? Then we need to keep it as an entity. What is the difference between a callback and the virtual function call from the first patch? We can change it to provide function pointers instead like in my last patch. It's less idiomatic but a bit cheaper. Of course we could use `std::function` callbacks and capturing lambdas instead, but I am afraid that would be even more expensive than the virtual function proposal. > We also have some pending work (which I was planning on working when time > permits) in that area [D146809](https://reviews.llvm.org/D146809). I was > wondering if that'd interfere with your work... Your patch doesn't actually touch the `RuntimeInterfaceBuilder`, so this should be no problem. https://github.com/llvm/llvm-project/pull/83126 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
