Jessica Clarke
Jessica Clarke
That doesn't help you when the model, as it should for clarity, accesses the register directly by its language-level register, which has zero ties to any notion of CSR address...
Your model might want sstatus but someone else’s might want mstatus. Canonicalising makes things simpler to compare, and avoids confusion around whether state is aliasing or separate. I would argue...
How was the original PR tested if we’re finding this now?
> > How was the original PR tested if we’re finding this now? > > We ran it on the tests in this repo, which I thought does include RV32...
The current trace output it not a stable interface, changing it is fine
I don't see why riscof should care. The log format is not something that should be baked into it, for that you want a very clearly-defined machine-readable trace, not the...
(isn't there a trace document in the works for precisely that kind of thing?)
There are also a couple of problems with the patch as it stands: 1. The remapping is kept separate from the actual implementation, so it's easy to add a new...
> Does it make sense for the latter to be something more like: > > mapping clause assembly = C_LI(imm, rd) > if rd == zreg > "c.li" ^ spc()...
Note that LLVM will treat it as `c.li zero, imm`, i.e. the "obvious" mapping. I assume binutils is similar.