RFR: 8305959: x86: Improve itable_stub [v6]

[Aleksey Shipilev]

On Mon, 5 Jun 2023 10:10:22 GMT, Boris Ulasevich <bulasevich at openjdk.org> wrote:

>> Async profiler shows that applications spend up to 10% in itable_stubs.
>> 
>> The current inefficiency of itable stubs is as follows. The generated itable_stub scans itable twice: first it checks if the object class is a subtype of the resolved_class, and then it finds the holder_class that implements the method. I suggest doing this in one pass: with a first loop over itable, check pointer equality to both holder_class and resolved_class. Once we have finished searching for resolved_class, continue searching for holder_class in a separate loop if it has not yet been found.
>> 
>> This approach gives 1-10% improvement on the synthetic benchmarks and 3% improvement on Naive Bayes benchmark from the Renaissance Benchmark Suite (Intel Xeon X5675).
>
> Boris Ulasevich has updated the pull request incrementally with one additional commit since the last revision:
> 
>   push/pop(temp_get) -> push/pop(rdx)

I measured the last PR on `c6.8xlarge`:


Benchmark                        Mode  Cnt   Score   Error  Units

# Baseline
InterfaceCalls.test1stInt2Types  avgt   12  1.582 ±  0.098  ns/op
InterfaceCalls.test1stInt3Types  avgt   12  6.218 ±  0.001  ns/op
InterfaceCalls.test1stInt5Types  avgt   12  6.220 ±  0.004  ns/op
InterfaceCalls.test2ndInt2Types  avgt   12  2.215 ±  0.004  ns/op
InterfaceCalls.test2ndInt3Types  avgt   12  7.590 ±  0.008  ns/op
InterfaceCalls.test2ndInt5Types  avgt   12  7.591 ±  0.004  ns/op
InterfaceCalls.testIfaceCall     avgt   12  6.238 ±  0.006  ns/op
InterfaceCalls.testIfaceExtCall  avgt   12  8.389 ±  0.500  ns/op
InterfaceCalls.testMonomorphic   avgt   12  1.035 ±  0.001  ns/op

# Patched
InterfaceCalls.test1stInt2Types  avgt   12  1.476 ±  0.001  ns/op ;  +7.2%
InterfaceCalls.test1stInt3Types  avgt   12  5.848 ±  0.012  ns/op ;  +6.3%
InterfaceCalls.test1stInt5Types  avgt   12  5.842 ±  0.009  ns/op ;  +6.4%
InterfaceCalls.test2ndInt2Types  avgt   12  2.213 ±  0.001  ns/op ; <wash>
InterfaceCalls.test2ndInt3Types  avgt   12  6.548 ±  0.002  ns/op ; +15.9%
InterfaceCalls.test2ndInt5Types  avgt   12  6.549 ±  0.003  ns/op ; +15.9%
InterfaceCalls.testIfaceCall     avgt   12  5.872 ±  0.007  ns/op ;  +6.2%
InterfaceCalls.testIfaceExtCall  avgt   12  6.589 ±  0.008  ns/op ; +27.3% (high noise)
InterfaceCalls.testMonomorphic   avgt   12  1.035 ±  0.001  ns/op ; <wash>

-------------

PR Comment: https://git.openjdk.org/jdk/pull/13460#issuecomment-1577320318