RFR 8229898 [lworld] Enable C1 by default

David Simms david.simms at oracle.com
Wed Sep 11 09:20:03 UTC 2019


Patch itself looks good.

Excellent, great work Ioi, on the whole C1 effort. So enabling C1 by default is a major step for the moving the prototype towards production code. Obviously there is still work remaining, but until now the performance characteristics of the prototype have been hard to compare directly to mainline. Your efforts with getting C1 up and running, go a long way towards a comparable build.

Once the repo shenanigans are resolved (not sure just now, "hg verify" before push for paranoia)...go ahead and push

/David Simms.

----- Original Message -----
From: ioi.lam at oracle.com
To: valhalla-dev at openjdk.java.net
Sent: Tuesday, September 10, 2019 5:51:36 PM GMT +01:00 Amsterdam / Berlin / Bern / Rome / Stockholm / Vienna
Subject: RFR 8229898 [lworld] Enable C1 by default

https://bugs.openjdk.java.net/browse/JDK-8229898
http://cr.openjdk.java.net/~iklam/valhalla/8229898-enable-c1-default.v01/

We have been testing with -XX:+EnableValhallaC1 in mach5 for a couple of 
weeks and don't see any failures. So this patch is to enable C1 by 
default (same as the usual JDK)

(I'll push after a new LW2 update release is done).

+ By default both C1 and C2 are enabled, and tiered compilation is enabled
+ The EnableValhallaC1 flag is removed
   + To use C1 only: -XX:TieredStopAtLevel=1 -XX:+TieredCompilation
   + To use C2 only: -XX:TieredStopAtLevel=4 -XX:-TieredCompilation

I adjusted the test cases that are specific to C1/C2 testing to use the 
flags the new way.

I also fixed a bug in ValueTypeTest::parseOutput to properly parse the 
-XX:+PrintCompilation output (with tiered compilation, an extra number 
is printed to indicate the tier level). Previously, (due to this parsing 
bug) we disabled VERIFY_IR when C1 is enabled.

@Tobias -- I had to add this. I think when C1 is enabled, the number of 
counted loops could be different

  541       if (TieredCompilation) {
  542           // FIXME: TestLWorld.test88 fails with "expected 4 to 
equal 2"
  543           continue;
  544       }

Thanks
- Ioi


More information about the valhalla-dev mailing list