================
@@ -2036,6 +2041,7 @@ void CIRToLLVMFuncOpLowering::lowerFuncAttributes(
attr.getName() == func.getDsoLocalAttrName() ||
attr.getName() == func.getInlineKindAttrName() ||
attr.getName() == func.getSideEffectAttrName() ||
+ attr.getName() == CIRDialect::getNoReturnAttrName() ||
----------------
erichkeane wrote:
Hmm.... I agree this is a little awkward in a few ways unfortunately. IMO,
LLVM-IR seems to be leaning towards most attributes being added/encoded, at
least as much as is in LLVM as these 'builtin' attributes. So I think their
balance for that makes sense.
As far as here, `noreturn` is a bit on the border. It is one that I don't
expect to emit very often/regularly, and I would like us to find a pretty
reasonable balance there at one point. IF we find ourselves looking noreturn
up a lot, I'd be completely in agreement/favor of 'promoting' it though.
I think the extra-attrs is a bit of a waste when that functionality is already
sorta built in to MLIR, as it is here.
https://github.com/llvm/llvm-project/pull/177978
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
