[clang] [llvm] [RISCV] Add Smdbltrp and Ssdbltrp extension (PR #111837)
Craig Topper via cfe-commits
cfe-commits at lists.llvm.org
Thu Oct 10 09:48:55 PDT 2024
topperc wrote:
> > > Personally I don't like to add a privileged extension if it doesn't introduce new CSRs/instructions.
> >
> >
> > I'd actually put that on the [agenda](https://docs.google.com/document/d/1G3ocHm2zE6AYTS2N3_3w2UxFnSEyKkcF57siLWe-NVs/edit?tab=t.0) for today's sync-up call already. We did add the ones that are in the profiles. I could see an argument for supporting everything as people can then have a `-march` string that fully describes the target and don't need to go and sort through which options to drop as they don't affect the compiler. But I also see a counter-argument. I'll summarise anything relevant from the sync-up call discussion here.
>
> The conclusion was that nobody present saw a real argument against, feeling "why not" and as I mentioned above, it means people can just match the `-march` string to their CPU data sheet. Did you have a particular reason against adding such extensions?
Our hardware build generates a configuration file with a full -march string based on all extensions present in the RTL which we deliver to our customers. So I would like all extensions to be supported otherwise I have to do it downstream.
https://github.com/llvm/llvm-project/pull/111837
More information about the cfe-commits
mailing list