John Rose wrote: > I finished methodHandleWalker.cpp, except for the bytecode compiler, > which I'm not sure we need at this point. > > If you run with +TraceMethodHandles and +Verbose, you'll see zillions > of method handle printouts using the walker.
The current patch does not build. I wonder if you missed something. > > I've spot-checked them and they seem correct, so it's time to think > about using the walker to generate either bytecodes or C2 nodes. Does it have any advantages to use bytecodes? Maybe for C1? Thinking about it, maybe it's easier to generate directly C2 nodes... > > So, Christian and I are a good step closer to full inlining! > > Christian, if we inline with C2 nodes, we'll have to generate oop maps > (safepoints) as of the call into the inlined MH chain; this is similar > to what we are doing with debug info for Object.clone, where sometimes > the operation is restarted instead of continued. At present, it is > always valid to restart an MH chain, because everything but the final > DMH is just casting, conversion, and other argument motion, which > might throw exceptions, but otherwise has no side effects. That is good. Is it possible that future actions in a MH chain can have other side effects? -- Christian _______________________________________________ mlvm-dev mailing list mlvm-dev@openjdk.java.net http://mail.openjdk.java.net/mailman/listinfo/mlvm-dev
