Request for reviews (XS): 7048030: is_scavengable changes causing compiler to embed more constants
Tom Rodriguez
tom.rodriguez at oracle.com
Tue May 24 18:17:13 PDT 2011
On May 24, 2011, at 6:00 PM, John Rose wrote:
> If this affects the processing of 292 data, there may be significant performance regressions. The key uses of embedded non-perms are invokedynamic CallSites, which requires TypeOopPtr::make_from_constant to succeed for user-provided CallSite objects.
>
> But, I think this will continue to work if ScavengeRootsInCode is not zero. Do you agree?
It should restore the logic to how it used to work prior to 7041789.
http://hg.openjdk.java.net/jdk7/hotspot-gc/hotspot/raw-diff/2aa9ddbb9e60/src/share/vm/gc_interface/collectedHeap.hpp
@@ -284,11 +291,7 @@ class CollectedHeap : public CHeapObj {
// An object is scavengable if its location may move during a scavenge.
// (A scavenge is a GC which is not a full GC.)
- // Currently, this just means it is not perm (and not null).
- // This could change if we rethink what's in perm-gen.
- bool is_scavengable(const void *p) const {
- return !is_in_permanent_or_null(p);
- }
+ virtual bool is_scavengable(const void *p) = 0;
292 ignores should_be_constant anyway.
tom
>
> -- John
>
> P.S. FTR, some of history behind is_scavengable and is_perm is in this discussion, relating to 6863023:
> http://mail.openjdk.java.net/pipermail/hotspot-gc-dev/2009-September/001261.html
>
> On May 24, 2011, at 5:19 PM, Vladimir Kozlov wrote:
>
>> http://cr.openjdk.java.net/~kvn/7048030/webrev
>>
>> Fixed 7048030: is_scavengable changes causing compiler to embed more constants
>>
>> This is the cause of 7047300 problem to appear. The fix for 7041789 corrected the meaning of is_scavengable to really mean what it says. Unfortunately there were some places that were really using it as a proxy for !is_perm. In particular in ciObject::can_be_constant and should_be_constant we will now embed constants that we didn't used as long as they are in tenured. The code should be changed from !is_scavengable to is_perm.
>>
>> Verified with failing test from 7047300.
>
More information about the hotspot-compiler-dev
mailing list