Won't CompilationEnvironment.getOptimisticType() widen vars that have
not been assigned yet to object type (since their value is undefined)?
If not, why not?
Other than that the patch looks good.
Hannes
Am 2014-03-24 15:12, schrieb Attila Szegedi:
Repeated request to please review this. Since I got no reviews first time
around, I updated it in-place with further development -- see the JIRA issue.
box2d is down to 377 recompilations by having the compiler do safe evaluation
of certain expressions.
http://cr.openjdk.java.net/~attila/8037534/webrev.01/index.html
On Mar 19, 2014, at 5:34 PM, Attila Szegedi <attila.szeg...@oracle.com> wrote:
Actually, the link is
http://cr.openjdk.java.net/~attila/8037534/webrev.01/index.html
On Mar 19, 2014, at 5:33 PM, Attila Szegedi <attila.szeg...@oracle.com> wrote:
Please review JDK-8037534 at
http://cr.openjdk.java.net/~attila/8037534/webrev.00
From the JIRA issue:
"When doing on-demand compilation, we can use the current runtime scope to look up
types of variables in scope; if they're already wider than the current optimistic
assumption, we can use that type and cut down on number of recompilation."
This cuts down warmup time with optimistic compilation. In Octane tests, box2d
now does 1300 deoptimizations instead of 1552 while it warms up, and
earley-boyer is down to 1577 from whopping 3028.
Thanks,
Attila.
