RFR(S): 8207848: [lworld] Compilation fails with "must have enough JVMS stack to execute invokevirtual"
tobias.hartmann at oracle.com
Fri Jul 20 09:53:48 UTC 2018
On 19.07.2018 16:47, Roland Westrelin wrote:
> Would it be feasible to adjust the bci and sp only once, on entry of the
> first call to ValueTypeNode::make_from_oop() then you wouldn't have to
> pass trap_bci around?
Yes, we could do that but the problem is that we also need to push NULL before the uncommon trap if
we continue execution at the next bytecode (because the NULL result needs to be available on the
stack for re-execution, see line 556 and 585). Since we need to do that right before the uncommon
trap and only if trap_bci != -1, we would still need to pass trap_bci to load().
More information about the valhalla-dev