William Moses

Found the error, we mistakenly run `LowerExcHandlers` before and after Enzyme, changed this to only run after Enzyme.

Okay, going through exceptions is now okay if the exception is not taken. But taking it causes an illegal instruction (and I'm not sure why, IR below): ``` mod =...

https://github.com/EnzymeAD/Enzyme/issues/597

This now hits: ``` ERROR: LoadError: UndefRefError: access to undefined reference Stacktrace: [1] unsafe_convert @ /mnt/Data/git/Enzyme.jl/src/compiler.jl:4493 [inlined] [2] macro expansion @ /mnt/Data/git/Enzyme.jl/src/compiler.jl:4736 [inlined] [3] enzyme_call(::Ptr{Nothing}, ::Type{Enzyme.Compiler.ForwardModeThunk}, ::Type{Val{0x0000000000000001}}, ::Val{true}, ::Type{Tuple{Const{Core.MethodInstance}, Const{Symbol}}},...

Now duplicate of https://github.com/EnzymeAD/Enzyme.jl/issues/293

Yeah, in this regard we'd want a version that keeps the non-zero values from the sparse array being zero'd in the backing array (but set to 0).

Copying my comment from slack here to preserve it: The same structure constraint for sparsity is a bit more rigorous because specifically the data layout itself between the two should...

Another option here, could be to define custom rules for sparse get/set -- obviating the need for this here.

Could we replace zeros with a placeholder value (say nan) then filter through the backing array and replace?

Can you retry on latest main? I see a Julia-level error. ``` Cannot deduce type of store store i64 0, i64 addrspace(13)* %63, align 8, !dbg !140, !tbaa !132 Caused...