Synchronization cost regression

Jeff Sturm jsturm@one-point.com
Sat Aug 4 07:39:00 GMT 2001


On Wed, 1 Aug 2001, Boehm, Hans wrote:
> I noticed what apppears to be a regression in the synchronization cost with
> the the trunk.  I recently updated my tree, and some synchronization tests
> run 10-20% slower.  Looking at the X86 assembly code, a likely culprit seems
> to be that _Jv_ThreadSelf is not getting inlined into _Jv_monitorEnter and
> _Jv_MonitorExit.  Unfortunately, I no longer have the old tree around to
> easily verify that this was happening before.

Inlining is achieved at -O and higher with the `inline' keyword.  I don't
see a problem, however my backend is slightly out of date.

Somebody else mentioned that the c++ inliner heuristics have recently been
tweaked.  That shouldn't affect an explicit inline.  But if this behavior
persists it might be a good idea to file a PR.

Jeff



More information about the Java mailing list