On 02/02/12 15:47, Geoffrey Blake wrote: > Hey all, > > Now that the CheckerCPU is back from the dead, I know that it would be > extremely useful for me to create a few regression tests for it to > make sure it doesn't fall back into disrepair. I'm not quite sure how > to go about this because currently the checker is a compile time > option. The gem5 wiki shows how to make a new regression test but not > for this particular case were the sim would have to recompiled first > before running the test. Any pointers on this? Or should I follow > Gabe's lead and make the USE_CHECKER compile time variable a runtime > variable instead? This would simplify the regression process, but > would add overhead in the O3 cpu as it would have to check on every > commit whether or not to call the checker. Thoughts? > > Geoff > _______________________________________________ > gem5-dev mailing list > [email protected] > http://m5sim.org/mailman/listinfo/gem5-dev
This is just my opinion, and on top of that I haven't really looked into exactly what the facts of the situation are, but I think something like this wouldn't be a bad idea. If you can boil down the points of interaction with the checker to just one or two ifs, the overhead relative to the rest of O3 would be very small. I would suggest that rather than check USE_CHECKER or UseChecker or whatever at run time, you use the presence or absence of a checker as your on/off switch. If the checker pointer is NULL, don't bother to exercise those one or two points of contact. Finding out if the checker is there should also be fast, ideally by just checking a pointer directly. I'm not that committed to this idea, though. It's just my gut reaction, and I thought I'd throw it out there. Gabe _______________________________________________ gem5-dev mailing list [email protected] http://m5sim.org/mailman/listinfo/gem5-dev
