Ben Vanik
Ben Vanik
Yep, sounds great! You can follow the existing enum-like ops for 1. For 2/3 a ResolveTopologyQueriesPass could do that op to start and others in the future - if it...
Simplified to just removing the attribute and using the C `inline` keyword. PTAL!
It's not the bug here, but I'd look further back in the IR for where this is coming from - it's splatting a 0 value into a buffer to then...
(consteval may handle this case but not others in the program if this is a new construct we haven't optimized yet, and running full consteval to do this slows down...
oof, that's bad
Yeah, I don't understand either - we use affine.apply everywhere (which you're finding) as everyone has said "it's just a way to calculate some indexing math" - if it really...
The comment may be dated but the code is still doing that - if you think it's safe to remove (or conditionally do it based on the `op` being inlined)...
Good points - negative indexing and 0-element tensors are a big red flag. I'm quite sure that will cause issues somewhere. I don't think it solves the failed to inline...
This won't be implemented as it is upstream, but the functionality is present for it - someone just needs to wire it up (and it's proven not to be useful...
This is due to the very long tied op chain causing the solver to walk one tied op per iteration and eventually hit the limit. You can try working around...