Re: Benchmark Errors
Phillip Stanley-Marbell
(Please post questions directly to the mailing list, so that
other users might be able to answer new questions / benefit
from responses if they run into the same issue. Starting this
year I am attempting to force everyone to use the mailing
list, rather than sending me questions directly. I have
"anonymized" the question and am posting this response to
the mailing list. Thanks!)
It looks like the benchmark is accessing a memory region
(e.g., a memory-mapped register that does not exist in
the machine configuration the benchmark is being executed
on.)
(*) Could you kindly post the full console transcript,
starting from what is printed as the benchmark binaries
are loaded into memory, and also, a log file by e.g.,
"d debug.log" at the simulator prompt)?
(*) Please post a link to the compiled benchmark binaries
you were running (the ".sr" files generated by the benchmark
compilation via sh-coff-gcc via the Makefile) so I can run
the same binaries to reproduce the problem.
(*) I don't see how the same binaries would run on AVRORA,
since the simulators model different instruction set
architectures... perhaps I misunderstand your point. Thanks.
Best regards,
phillip
> I am trying to simulate MiBench Bechmarks (for being a bit light
> weight) on SUNFLOWER. I encounter the following error for all of them.
>
> [ID=0 of 1][PC=0x8009174][3.3E+00V, 6.0E+01MHz] Word access (read)
> at address 0x9312d00
> Sunflower FATAL (node 0) : <Address not in main mem, and not in I/O
> space either !>
>
> I am using your own LARGE.M or SMALL.M files to run the benchmarks
> but after some time it seems that they try to access an area out of
> memory range. I have even tried to increase memory by manyfolds but
> no use. The same benchmarks I have executed on AVRORA (AVR
> SImulator) and they work well. Can you please put some light on this?