William Moses

So if we have it as an option to fill it in from the frontend there'd need to be an Enzyme TypeTree passed in (C++ object). I suppose we could...

I don’t think this would be quite sufficient. This alone would enable the type information at the boundaries whereas to eliminate custom tbaa handling wed also need to enable a...

Ok the primitive C API I just exported should be sufficient for an initial version of this (once we rewrite Enzyme.jl to use the Enzyme c functions to create gradients...

Now duplicate of https://github.com/EnzymeAD/Enzyme.jl/issues/172

Example: ``` ; Function Attrs: willreturn mustprogress define private { {} addrspace(10)*, { { {} addrspace(10)*, {} addrspace(10)* }, { {} addrspace(10)*, {} addrspace(10)* } } } @preprocess_julia_apply_3569_.8([1 x {...

@vchuravy alternatively, would using the GC functions you mentioned earlier remedy. @vtjnash is it possible to have this function extended to also handle (newtype*)malloc(...) somehow

@vchuravy at this point I presume we close this

@vchuravy do you have an example?

@vchuravy closing for now, reopen if this persists.

Adding this as an option actually enables us to handle https://github.com/wsmoses/Enzyme.jl/issues/112 by simply returning the shadow return via the corresponding shadow return arg.