vm server option generating different LIR nodes
thomas.wuerthinger at oracle.com
Wed May 8 22:45:46 PDT 2013
the gathered profiling information might be different between the "--vm server" and the "--vm graal" option. In general, linking to the generated "Client Compiler Visualizer" or "IdealGraphVisualizer" files can be very helpful.
On May 6, 2013, at 5:04 PM, "Venkatachalam, Vasanth" <Vasanth.Venkatachalam at amd.com> wrote:
> I am debugging a test case for my HSAIL backend that is giving the wrong result when run with -vm server option.
> Using the Client Compiler Visualizer, I noticed that the nodes generated are different with the -vm server option, even before the "After LIR generation" phase.
> Is there a reason why the nodes would be different when running with -vm server?
> The nodes shown for the "After linear scan order" phase (right before LIR) look suspicious to me in that it looks like there's a dangling else block, so that the exit out of a loop isn't hitting some cleanup code that needs to be executed. I see this issue propagated down to the LIR, and since my codegen is matching up with the LIR, it inherits the dangling else which seems to be the culprit here.
More information about the graal-dev