Hi,


I tried to test the effects of "invocation_count_threshold" with the following command.

rvm -X:vm:measureCompilation=true -X:aos:invocation_count_threshold=1 -jar dacapo-2006-10-MR2.jar -s small eclipse


invocation_count_threshold=1
===== DaCapo eclipse PASSED in 5978 msec =====

Compilation Subsystem Report
Comp #Meths Time bcb/ms mcb/bcb MCKB BCKB
JNI 32 0.43 NA NA 3.8 NA
Base 7650 392.70 1208.02 6.37 3539.6 555.6
Opt 201 673.44 39.14 6.72 104.7 15.6


invocation_count_threshold=100
===== DaCapo eclipse PASSED in 5295 msec =====

Compilation Subsystem Report
Comp #Meths Time bcb/ms mcb/bcb MCKB BCKB
JNI 32 0.45 NA NA 3.8 NA
Base 7571 494.50 1261.62 6.37 3508.4 551.0
Opt 164 522.23 37.85 6.24 80.9 13.0


invocation_count_threshold=10000
===== DaCapo eclipse PASSED in 5321 msec =====

Compilation Subsystem Report
Comp #Meths Time bcb/ms mcb/bcb MCKB BCKB
JNI 32 0.49 NA NA 3.8 NA
Base 7572 431.47 1216.95 6.37 3506.0 550.6
Opt 157 710.36 39.95 6.65 84.8 12.8

invocation_count_threshold=1
===== DaCapo eclipse PASSED in 5078 msec =====

Compilation Subsystem Report
Comp #Meths Time bcb/ms mcb/bcb MCKB BCKB
JNI 32 0.47 NA NA 3.8 NA
Base 7580 361.04 1273.16 6.37 3508.6 550.9
Opt 147 445.70 42.29 6.32 75.8 12.0

However, I didn't observe any obvious differences: the numbers of methods in "Base"
and "Opt" keep in the same level (there are some data fluctuation)
Did you use it correctly? If so, why doesn't affect the performance?


Thanks,
Zhijia