cfallin commented on issue #5046:
It's interesting that this was hit only when egraph-based optimization is on -- I suspect because of the different way that constants are rematerialized...
In any case, I do think that disallowing
iconst.i128
makes the most sense: the type is kind of a lie, because the instruction can't actually carry a 128-bit payload. Given that it's not fully general, producers need to do something else to emit an arbitrary 128-bit constant anyway, so we might as well remove the partial support and relieve the backends of some complexity.
afonso360 commented on issue #5046:
Closing this since I'm working on removing
iconst.i128
instead
Last updated: Nov 22 2024 at 16:03 UTC