I agree that having both optimized and unoptimized binaries for at least some workloads is useful for regressions, but I'm not convinced it's a huge loss to require different names for them... you'd get some duplication in the input files but that's about it. However, there are enough things that need to be fixed about the regression system that we might as well add this capability in while we're fixing everything else.
I think the 32- vs 64-bit regression should already be taken care of as long as those appear as independent operating systems (or even ISAs)... I'm guessing your problem before was that you weren't making that distinction at the top level. We should also get some regressions that use prefetching; it's unlikely the prefetching code would have gotten so broken if we had. I think the current structure is capable of handling this by creating a new platform (e.g., 'simple-timing-prefetch' in addition to 'simple-timing'), but there may be a better way to support variations like this if we revamp the whole system. Steve On Mon, Feb 16, 2009 at 1:54 AM, Gabe Black <[email protected]> wrote: > I have a 32 bit hello world working, and thinking about making it > part of the regressions reminded me of when I was doing something > similar for SPARC. Back then I didn't find a good way for 32 bit and 64 > bit benchmarks/binaries to exist in the regression system at the same > time since they're always expected to be in the same directory with the > same name. It would be nice to be able to also have, for instance, 32 > bit versions of hello world compiled with and without optimizations. > I've found the optimized binaries are non trivially harder to run > correctly, but they may not cover all the same cases as the unoptimized > ones. If we're doing some sort of fancy regressions overhaul in the > future, it would be nice if this sort of thing could be accommodated > better. > > > Gabe > _______________________________________________ > m5-dev mailing list > [email protected] > http://m5sim.org/mailman/listinfo/m5-dev >
_______________________________________________ m5-dev mailing list [email protected] http://m5sim.org/mailman/listinfo/m5-dev
