Hi Dan,
Thanks so much for your time. I verified both of the things you mentioned
(that I had setup the Gems/Simics-3 as per the instructions on the wiki,
and that my checkpoint was single-processor). I also tried a couple of
other things, and noticed the following:
a) independent of cache-coherence protocol (I tried both the default
MOSI_SMP_bcast and also MOESI_CMP_directory), I *can* get Ruby to
load, initialize, and dump proper statistics as long as I do not load
Opal at any point
b) if I load Opal, Ruby dies on init
Do you know any good method I could use to attach a debugger to the
proceedings and try and give more precise info? Is it as simple as
building Opal/Ruby with debug info and gdb'ing Simics?
Thanks again,
Adam
> Before I fly off the hinge and assume the worst (like I did with the
> last gems-users segfault), can you verify the following for me?
> 1) your checkpoint has only one processor (I've done it before)
> 2) you modified the files in the Gems 1.3 release appropriately for
> Simics 3.0. These should include:
> $GEMS/common/Makefile.simics_version -- set the version string to 3.0
> $GEMS/common/Makefile.common -- set the CC for your chosen compiler,
> update the path for SIMICS_INCLUDE_ROOT to point to your Simics 3.0
> installation's include/ directory. $GEMS/ruby/module/Makefile -- set the
> runtime library path (not necessary if using native compiler), and use
> the Simics-3.0-appropriate definition of GEMS_ROOT.
> $GEMS/opal/module/Makefile
> -- Use the Simics-3.0-appropriate definition of GEMS_ROOT.
> Regards,
> Dan
|