Thank you, Dan.
I thought setting g_MEMORY_SIZE_BYTES to 2GB would solve the
problem, but it didn’t. The error didn’t change.
However, I got GEMS’ EE/LL TM working with
$GEMS/benchmarks/transactional/prioqueue benchmark.
This hints me that the problem might be on the SPLASH2 benchmark
compilation.
I am not completely sure whether prioqueue worked perfectly, but
this is the stat file I got from the simulation.
Could any of you read this stat file if this is good?
http://www.cs.utah.edu/~bchong/CA/debugging/GEMS-Micro_prioqueue_EETM_04.stats
http://www.cs.utah.edu/~bchong/CA/debugging/GEMS-Micro_prioqueue_LLTM_04.stats
Since prioqueue benchmark is working, there is a high chance
that my SPLASH2 benchmark compilation is wrong.
Could any of you check if this SPLASH2 barnes benchmark is OK?
I've compactly extracted only barnes part.
http://www.cs.utah.edu/~bchong/CA/debugging/SPLASH2_barnes.tbz2
For direct access
http://www.cs.utah.edu/~bchong/CA/debugging/SPLASH2_small/
Or if any of you have one working SPLASH2 benchmark on GEMS TM,
please send me a copy.
BTW, I am using PROTOCOL=CMP_filter_directory for ruby
compilation, do you think this is correct? I thought normal EE/LL TM would need
same ruby protocol as ATMTP’s.
From:
gems-users-bounces@xxxxxxxxxxx [mailto:gems-users-bounces@xxxxxxxxxxx] On
Behalf Of Dan Gibson
Sent: Sunday, March 29, 2009 7:54 AM
To: Gems Users
Subject: Re: [Gems-users] Address assertion error on GEMS LogTM
Based on the contents of the
memory map, set g_MEMORY_SIZE_BYTES to 2147483648.
Regards,
Dan
--
http://www.cs.wisc.edu/~gibson
[esc]:wq!