Dear Kim,
A magic instruction is a special no-operation instruction. When Simics executes
such an instruction, it triggers a Core_Magic_Instruction hap and calls all the
callbacks functions registered on this hap. Because a magic instruction is a
no-operation instruction, adding a magic instruction to a benchmark does not
change the behavior of the benchmark.
Magic instructions are useful. As an example, you can add a magic instruction
to the location in your benchmark where you want to take the first checkpoint
(i.e., after initialization and warmup).
[Pseudo Code *BEGIN*]
// initialization
// warm up
MAGIC(1);
// the rest of the benchmark
[Pseudo code *END*]
Then you can add a Core_Magic_Instruction hap to stop Simics simulation
whenever the magic instruction is executed.
[Simics script *BEGIN*]
@def hap_callback(user_arg, cpu, arg):
if arg == 1:
SIM_break_simulation("Magic Instruction")
@SIM_hap_add_callback("Core_Magic_Instruction", hap_callback, None)
[Simics script *END*]
You should compile the benchmark with the magic instruction and register the
hap callback with the Simics script. If you run the benchmark in Simics, the
simulation will stop whenever the magic instruction get executed. You can
easily the the first checkpoint.
Regards,
Pejman
________________________________________
From: Eun-Sung Kim [[email protected]]
Sent: Monday, January 24, 2011 9:39 PM
To: simflex
Subject: Magic Breakpoint and checkpoint
Hi all,
I inserted a magic breakpoint in a SPEC benchmark, and compiled the benchmark
in a targeted architecture(sun-solaris). I made a Simics script file for Flexus
simulation(Thanks to Mahmood Naderan for importing binary images to Simics). As
an example of preparing the test application in "Flexus Getting Started Guide",
'filter' test application works well, but my work does not. My script file is
mostly same as an prepare-test-app-v9.simics. In simulated machine on Simics,
the benchmark program is run and finished, then there is no further processing.
It still stays in simulated machine(solaris).
I think it is due to a problem of magic breakpoint injection in SPEC benchmark.
Could you show me an example of how and where you can inject a Magic breakpoint
into SPEC benchmark(ex. gzip)?
Thanks,
--Kim
