I have got the barnes benchmark to run on solaris with gcc+pthread+magic instruction!!! I solve the sched_yield problem by -lrt option, which means load with the realtime library. I edited the barnes's code.C and code_io.C, added begin/end transaction with the lock/unlock pair. I have now one more task which is try this out on the GEMS simulator.
I also got the opal of GEMS simulator to run by adding a link of mf_api.h's link in opal/system diretory.
I attempted to run barnes benchmark in GEMS at home computer, libm.so.2 missing caused ld error. I am going to try to copy the libm.so.2 from the solaris machine to the GEMS simulator and retry at uni lab machine.
I still have problem with the shared-counter program that Kevin sent me. I think the problem is that the barrier implemetation in that benchmark is incompatible with the solaris system we have got, so I am going to have a chat with him sometimes tomorrow.
task for tomorrow 1) run the banes benchmark with LogTM setting first in GEMS 2) run the barne benchmark in SLE setting in GEMS 3) write a script to run the simulation 4) talk to Kevin about the shared-counter barrier problem
Task 1) done but simulation takes long time
Task 4) done. Problem solved.
Subscribe to:
Post Comments (Atom)
No comments:
Post a Comment