RFR (M): (7u60): Backport of 8027476: Improve performance of Stringtable unlink, 8027455: Improve symbol table scan times during gc pauses (original) (raw)

Jon Masamitsu jon.masamitsu at oracle.com
Wed Jan 29 19:55:36 UTC 2014


Thomas,

Changes look good.

Please run some simple test (does enough work to exercise your changes) with -XX:+UseDynamicNumberOfGCThreads .

If there is a problem (I don't expect one but would like it run), please create a CR and assign it to me.

Thanks.

Jon

On 1/29/2014 8:09 AM, Thomas Schatzl wrote:

Hi Mikael,

On Tue, 2014-01-28 at 15:00 +0100, Mikael Gerdin wrote: Thomas,

On Tuesday 28 January 2014 11.57.33 Thomas Schatzl wrote: Hi,

I got the link for the new webrev wrong, taking two from the original These differences are contained in symbolTable.?pp, the rest is verbatim backport iirc.

New webrev: http://cr.openjdk.java.net/~tschatzl/8027476/webrev Correct webrev is at: http://cr.openjdk.java.net/~tschatzl/8027476.7u60/webrev symbolTable.cpp: 103 Symbol* s = entry->literal(); 104 memorytotal += s->objectsize(); << should be (*memorytotal) +=_ _105 (*processed)++;_ _g1ColletedHeap.cpp:_ _5237 guarantee(StringTable::parallelclaimedindex() >= initialstringtablesize, 5238 errmsg("claim value "INT32FORMAT" after unlink less than initial string table size "INT32FORMAT, 5239 StringTable::parallelclaimedindex(), initialstringtablesize)); 5240 guarantee(!processstrings || SymbolTable::parallelclaimedindex() = initialsymboltablesize, 5241 errmsg("claim value "INT32FORMAT" after unlink less than initial symbol table size "INT32FORMAT, 5242 SymbolTable::parallelclaimedindex(), initialsymboltablesize)); 5243 You've removed the !processstrings from the guarantee compared to 9. Also, I noticed that the 9 versions has !processstrings in both guarantees although I think the second one should be !processsymbols. All fixed in the new webrev at http://cr.openjdk.java.net/~tschatzl/8027476.7u60/webrev.1/ Thanks a lot. I will file a bug for 8u20 and 9. Thanks, Thomas



More information about the hotspot-gc-dev mailing list