Re: [Gems-users] Ruby_cycles differ everytime and simics crash

["Min Xu (Hsu)" <xu@xxxxxxxxxxx>]

On Tue, 04 Oct 2005 Ahn, Minseon wrote :
> Hi, there.
> 
> I'm currently running Ruby (GEMS 1.1) and Simics 2.2.17. However, I don't know
> why Ruby_cycles is different everytime I run the same application with the
> same parameters of the application, the same simics configuration file and
> same parameters of Ruby. I tried to eliminate all the randomness in the
> simulators. I'm running Splash2, for example, fft, lu, radix, and ocean.

Ruby randomize every memory access latency a little bit to deal with workload
variabilities (see our paper in Computer Feb. 2003).

> 
> Also, when running it, someimes simics crashed. GEMS site says "nabling
> DATA_BLOCK when running Ruby w/ Simics causes a crash." But after compiling
> Ruby with enabling it and running, Simics crashed with this message:
> 
> failed assertion 'SIMICS_check_memory_value(m_proc,
> mem_trans->s.physical_address, buffer, mem_trans->s.size)' at fn void
> SimicsProcessor::observeMemoryAccess(memory_transaction_t*) in
> simics/SimicsProcessor.C:340
> failed assertion 'SIMICS_check_memory_value(m_proc,
> mem_trans->s.physical_address, buffer, mem_trans->s.size)' at fn void
> SimicsProcessor::observeMemoryAccess(memory_transaction_t*) in
> simics/SimicsProcessor.C:340

I don't think enabling DATA_BLOCK with simics is fully functioning. The
assertion you have seen is expected. More work is needed to make it work.
Right now, you perhaps don't want to turn DATA_BLOCK on.