Fix mul overload #440
Fix mul overload #440
Conversation
| @@ -130,14 +130,18 @@ for (cT, aT, bT) in ( | |||
| if any(Base.Fix2(isa, TracedRArray) ∘ ancestor, (C, A, B)) | |||
| TracedLinearAlgebra.overloaded_mul!(C, A, B, α, β) | |||
| else | |||
| LinearAlgebra._mul!(C, A, B, α, β) | |||
| LinearAlgebra.mul!(C, A, B, α, β) | |||
There was a problem hiding this comment.
this will very likely trigger the illegal instruction #393
There was a problem hiding this comment.
There was an absint behavior that we had a while ago but was accidentally removed in a refactor which I re added with the last cuda stuff.
I hope it means we won't hit it
There was a problem hiding this comment.
I checked it again in the stack PR, the crash persists
There was a problem hiding this comment.
Well darn, okay yeah if CI yells I think we need to make a separate noinline function which is called in the override, which itself calls mul
There was a problem hiding this comment.
lets add a test like https://github.com/EnzymeAD/Reactant.jl/blob/main/test/basic.jl#L864 to verify this
* fix: handle aos for mul * Update ext/ReactantArrayInterfaceExt.jl Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> * revert: incorrect aos_to_soa for C --------- Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
|
@avik-pal it looks like this is hitting some scalar index stuff? |
|
Yeah we need to change aos to anytracedarray |
![github-actions[bot]](https://avatars.githubusercontent.com/in/15368?s=60&v=4){kind=small}
Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
|
Okay @avik-pal tesfs pass so I’m going to merge |
No description provided.