review (S) for 6930043: C2: SIGSEGV in javasoft.sqe.tests.lang.arr017.arr01702.arr01702.loop_forw(II)I
Vladimir.Kozlov at Sun.COM
Fri Mar 12 16:34:01 PST 2010
Fine. But as we discussed you need to verify that invar.clone() does not have any side effects.
Tom Rodriguez wrote:
> That's what I get for doing edits just before generating the webrev. I'd originally deleted the clone since it seems useless. If it's invariant then cloning it doesn't change anything. I chickened out at the last minuted and restored but didn't recompile. I think I'll stick with deleting it. I've regenerated the webrev.
> On Mar 12, 2010, at 3:05 PM, Vladimir Kozlov wrote:
>> You removed lines 2225,2226 but where ctrl node comes from for next line?:
>> 2243 ld_rng = (LoadRangeNode*)invar.clone(ld_rng, ctrl);
>> Tom Rodriguez wrote:
>>> 6930043: C2: SIGSEGV in javasoft.sqe.tests.lang.arr017.arr01702.arr01702.loop_forw(II)I
>>> The new loop predication code is missing logic to test that the
>>> initial value of the index is in range. In many cases will be
>>> eliminated statically. Tested with failing test. Also tested that
>>> this new test doesn't affect the performance improvement we were
>>> seeing with scimark.
More information about the hotspot-compiler-dev