================
@@ -2428,6 +2435,11 @@ static ExprResult BuiltinMaskedScatter(Sema &S, CallExpr
*TheCall) {
QualType PtrTy = PtrArg->getType();
QualType PointeeTy = PtrTy->getPointeeType();
+ if (PtrTy->getPointeeType().isConstQualified())
+ return ExprError(
+ S.Diag(PtrArg->getExprLoc(), diag::err_typecheck_assign_const)
+ << /*read-only*/ 5);
----------------
rjmccall wrote:
The non-memory arguments should all be undergoing lvalue-to-rvalue conversion.
If the builtins are using custom type-checking, you need to do this explicitly.
This will:
- fix the use of ObjC or MSVC property reference expressions in these positions,
- change object references to be non-ODR-used if they're constant-evaluable, and
- remove qualifiers from the argument type.
The memory argument should not undergo lvalue-to-rvalue conversion, of course.
Since the intrinsic has to implement the memory access manually, you need to
actually consider all the qualifiers for the memory argument.
- `restrict`, ObjC, and `__ptrauth` qualifiers do not apply to vector types and
can be ignored.
- `_Atomic` should be disallowed because we don't want to implement it; this
may happen implicitly in Sema because of its representation, but it's something
you should explicitly test.
- The intrinsic appears to be overloaded properly for address spaces, so
address space qualifiers are presumably fine. This should be tested.
- `const` should be allowed only for loads; you've fixed that now, so good.
- Whether `volatile` should be allowed depends on whether we can maintain the
volatile semantics in LLVM IR. There is no `volatile` flag on the intrinsic, so
this really depends on (1) whether LLVM will ever optimize these intrinsics in
the middle-end and (2) whether it ever lowers these intrinsics in a way such
that the back-end could break `volatile` semantics. The safest answer is to
disallow it.
https://github.com/llvm/llvm-project/pull/160185
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
