On Mon, Feb 9, 2015 at 3:02 AM, Andreas Sandberg via gem5-dev < [email protected]> wrote:
> > > > On Dec. 12, 2014, 10:24 p.m., Gabe Black wrote: > > > While it's definitely nice to get these into regular C++ instead of > the ISA language, my concern is that these aren't really pseudo > instructions. They are in the sense that they're instructions that wouldn't > exist outside of the simulator, but there's a well defined pseudo inst > concept they don't really fit in. Maybe put them in a .cc and .hh file with > a different name? > > > > Andreas Sandberg wrote: > > I see your point. We could rename them to something like "nonisa" > since they aren't a part of the visible ISA. Would that be a good name? > Another option is to just say that everything in gem5 that isn't a proper > ISA-defined instruction is a pseudo instruction and that m5ops are just a > subset of those. > > Ping. Is renaming the files to nonisa.(cc|hh) an acceptable solution? Or > should we submit as is? I agree with Gabe; to me "pseudoinstruction" in gem5 implies the fake instructions that you can actually put in your program, as in src/sim/pseudo_inst.cc. I guess we also call them m5ops, so maybe we don't need two names. Also, the DecoderFaultInst that's already in the pseudo.* files doesn't really fit that definition either. Abstractly, "pseudoinstruction" to me sounds like something that's not really an instruction, like DecoderFault or FailUnimplemented; I'd argue that the m5ops are more real than those, i.e., they're actual instructions, just not ones defined by the ISA. So "non-isa" sounds more like it's describing m5ops than these internal instruction objects. So overall I'd say: 1. I think Gabe has a point that our naming conventions are inconsistent. 2. In spite of that, given that DecoderFaultInst is already in these pseudo.* files, I don't think the proposed patch makes anything worse, so I don't object to it. 3. Making the naming conventions consistent requires more discussion, but it will also require more work, and it's probably not worth the discussion unless someone is really serious about making it happen. 4. I think introducing new labels like "non-isa" at this point, without committing to the overhaul of #3, is probably not helpful. Steve > > _______________________________________________ gem5-dev mailing list [email protected] http://m5sim.org/mailman/listinfo/gem5-dev
