From: Ian R. <ian...@ma...> - 2005-09-20 13:51:30
|
Hi, still debugging from the IR verifier. I'm trying to work out when in generating the initial HIR it is possible for an invokeinterface bytecode to have created an instanceof test followed by an int_ifcmp. The problem is that they are being created and sharing a register operand: VERIFY: Initial HIR In block BB72 for instruction instanceof_notnull t308i(I) = Lspec/benchmarks/_202_jess/jess/_times;, l178a(Lspec/benchmarks/_202_jess/jess/Userfunction;), <unused> the back link in the def of operand t308i(I) is invalid and references int_ifcmp t309v(GUARD) = t308i(I), 0, ==, LABEL88, Probability: 0.00999999 I'm determining the faulting bytecode from the bytecode index in a trace. The IRGen debugging options don't throw up anything useful. Thanks, Ian Rogers |