cfallin commented on issue #5727:
(Rebased to unstack from #5726; there's no need to couple these actually)
jameysharp commented on issue #5727:
Can we generate an
InstructionData::deep_clone
method in the meta crate by looking at which formats have value-list operands? I'm worried this won't get updated as new formats are added.The same issue applies to
BlockCall
, right?Aside from bugs like #5726 the egraphs code currently can never hit any of these cases, right? I think all the instructions that use value-lists (just block terminators and calls I think) are impure and so are not currently duplicated.
cfallin commented on issue #5727:
Can we generate an InstructionData::deep_clone method in the meta crate by looking at which formats have value-list operands? I'm worried this won't get updated as new formats are added.
Yep, that's the better approach; I'll build that out.
The same issue applies to BlockCall, right?
Ah, yes, indeed, I'll need to fix that too!
Aside from bugs like #5726 the egraphs code currently can never hit any of these cases, right? I think all the instructions that use value-lists (just block terminators and calls I think) are impure and so are not currently duplicated.
Yep, there's no functional impact today; all pure insts happen to have fixed args.
cfallin commented on issue #5727:
Updated, thanks!
Last updated: Nov 22 2024 at 17:03 UTC