[13] RFR (S): 8059241: C2: Excessive RemoveUseless passes during incremental inlining

[Vladimir Ivanov]

Thanks, Aleksey.

I did some measurements with earlier version [1] and didn't notice any 
regressions:
                           BEFORE                 AFTER
Box2DBench            50.439 ±  5.644       43.549 ±  2.520  ms/op
CryptoBench            6.972 ±  0.137        7.019 ±  0.163  ms/op
DeltaBlueBench       630.607 ±  6.784      644.160 ± 11.296  us/op
EarleyBoyerBench      20.284 ±  0.415       20.735 ±  0.378  ms/op
GbemuBench            58.731 ±  2.852       59.245 ±  3.936  ms/op
NavierStokesBench      6.606 ±  0.042        6.628 ±  0.057  ms/op
PdfJSBench            83.063 ±  4.480       78.909 ±  2.268  ms/op
RaytraceBench       3583.622 ± 40.646     3578.562 ± 45.962  us/op
RegexpBench           70.060 ±  2.538       72.189 ±  2.651  ms/op
RichardsBench        233.414 ±  2.990      233.116 ±  5.641  us/op
SplayBench           956.145 ± 37.547      917.133 ± 37.245  us/op

Let me know about your findings.

Best regards,
Vladimir Ivanov

[1] http://hg.openjdk.java.net/jdk/jdk/rev/80b55cf3a804

On 25/01/2019 04:17, Aleksey Shipilev wrote:
> On 1/25/19 4:18 AM, Vladimir Ivanov wrote:
>> http://cr.openjdk.java.net/~vlivanov/8059241/webrev.03/
>>
>> I ran Octane w/ -XX:+CITime and observed significant reduction in "Incremental Inline" times (~50%
>> off on IGVN, 80-95% off on "Prune Useless").
> 
> Yes, I see "Prune Useless" going down with -XX:+CITime with this patch. However, I suspect there are
> performance regressions, and I am not able to verify them until this one is fixed:
>    https://bugs.openjdk.java.net/browse/JDK-8217782
> 
> -Aleksey
>