================
@@ -3928,6 +3928,11 @@ static StringRef getCanonicalArchString(Compilation &C,
return StringRef();
}
}
+ } else if (Triple.isSPIRV() && ArchStr.starts_with("igca_") &&
----------------
tahonermann wrote:
I don't think that solves a real problem. I'm not particularly worried about
users accidentally passing `--offload-arch=igcaXXX` for an AMD target.
We should have a definitive way of identifying targets that are appropriate for
IGCA. If we don't (and the only reason we don't is because we have so far
refused to make use of the vendor and OS components to identify Intel devices
for reasons that remain unclear to me), then I think we're better off not
trying to filter targets that are definitively not Intel rather than hoping
that we managed to identify them all.
I think the right answer here would be to require the vendor component to be
"intel" when using `--offload-arch=igcaXXX`. That will require further
discussion though.
https://github.com/llvm/llvm-project/pull/227468
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
