JDK-4750176 : [1.4.0_02] Regression on s9u2_06 of -5.74% in specJBB2k (-d64), -8.72% (-server)
  • Type: Bug
  • Component: hotspot
  • Sub-Component: compiler
  • Affected Version: solaris_9u2
  • Priority: P1
  • Status: Closed
  • Resolution: Duplicate
  • OS: solaris_9
  • CPU: sparc
  • Submitted: 2002-09-19
  • Updated: 2013-11-02
  • Resolved: 2002-09-30
Related Reports
Duplicate :  
Description
see comments

Comments
PUBLIC COMMENTS see comments
10-06-2004

EVALUATION JPSE team will need to handle this one. It appears an update release may be the suspect. More testing is being done. ###@###.### 2002-09-25 We've identified a large part for the performance regression. Here is what's explained by Chuck Rasbold: This is why: The fix for 4697245 triggers a bailout by having ciTypeFlow indicate a trap at bci 0. The fix for 4684993 removes a ciTypeFlow problem that could put a trap at an invoke bytecode, possibly at bci 0. If you have the 4697245 fix but not the 4684993 fix (like SpecJBB does in those three methods) a method will fail to compile and never re-attempt. We never had this problem in hopper since the 4684993 fix went in first. ###@###.### 2002-09-25 I've verified the binary that Mingyao sent me and this completely eliminates the performance regression. I've run 6 runs of jbb and across all warehouses I'm seeing performance differences in the 1/2% range, non statistically significant (i.e. it looks like the runs come from the same jvm). ###@###.### 2002-09-27 Close as a duplicate of 4684993, which was integrated into 1.4.0_03. ###@###.### 2002-09-30
27-09-2002