Hi @ckk,
I wonder if this is related to issues with when optimisation occurs and what it targets as described in this discussion: Compilation pipeline, compile time and vectorization. Also, the inline="always", functionality will likely lead to the use of additional variables etc to “wire up” the inlined call site, it may be that there’s something in there which prevents an optimisation occurring (e.g. the complexity gets too high).
Hope this helps.