[13] RFR (M): 6986483: CHA: optimize calls through interfaces

[Vladimir Ivanov]

Thanks, Nils.

Additional testing revealed one problematic corner case - Object methods 
on interfaces. The transformation is not valid for them, because it 
eliminates proper receiver subtype check: subtype check against Object 
is a no-op.

Updated version:
   http://cr.openjdk.java.net/~vlivanov/6986483/webrev.02/

The changes in c1_GraphBuilder.cpp and doCall.cpp are trivial 
(cha_monomorphic_target->holder() != Object_klass()), but I extended the 
test with more cases.

Testing: hs-precheckin-comp, tier1-5

Best regards,
Vladimir Ivanov

On 29/01/2019 06:03, Nils Eliasson wrote:
> Hi Vladimir,
> 
> A really good improvement, and I really like the test, excellent coverage.
> 
> Reviewed,
> 
> // Nils
> 
> 
> On 2019-01-25 22:27, Vladimir Ivanov wrote:
>> http://cr.openjdk.java.net/~vlivanov/6986483/webrev.01/
>> https://bugs.openjdk.java.net/browse/JDK-6986483
>>
>> Another candidate for revival. At that time it was reviewed, but 
>> integration was blocked pending another bug fix. Now the fix is in.
>>
>> Quote from original review request [1]:
>>
>> "Proposed change adds CHA support in C2 for interface calls.
>>
>> Consider the following hierarchy:
>>
>>    interface Intf { m(); }
>>    class C implements Intf { public m() { ... } }
>>    class C1 extends C { /* doesn't override m() */ }
>>    ...
>>    class Cn extends C { /* doesn't override m() */ }
>>
>> Call site: invokeinterface Intf.m() ...
>>
>> If Intf were an abstract class, CHA could deduce that Intf::m() can be
>> replaced with C::m(), but it doesn't work for interfaces. Verifier
>> doesn't check interface types in bytecode, so CHA can't assume the
>> receiver implements Intf.
>>
>> CHA in C1 handles such call sites for interfaces with a single
>> implementor. It replaces invokeinterface Intf.m() with invokevirtual
>> C.m() guarded by a subtype check (instanceof C). C2 doesn't do that and
>> this request is about adding that. Type profiling doesn't help here (the
>> call site is usually megamorphic), so C2 can't inline it.
>>
>> The proposed implementation is similar to C1, except that the code
>> deoptimizes when subtype check fails and ICCE is thrown from the
>> interpreter.
>>
>> While working on it, I spotted and fixed a couple of inefficiencies in
>> C1 implementation:
>>
>>    (1) dependency context being used was broader than necessary -
>> resolved instead of declared interface (hence, possibility of
>> unnecessary invalidations);
>>
>>    (2) didn't work for interfaces w/ any default methods: CHA doesn't
>> support default methods at the moment, so what matters is whether
>> Intf::m() is default or not and not whether Intf has *any* concrete 
>> methods."
>>
>>
>> Testing: hs-precheckin-comp, hs-tier1, hs-tier2
>>
>> Best regards,
>> Vladimir Ivanov
>>
>> [1] 
>> https://mail.openjdk.java.net/pipermail/hotspot-compiler-dev/2017-February/025630.html 
>>