[clang] [TBAA] Only emit pointer tbaa metedata for record types. (PR #116991)
Florian Hahn via cfe-commits
cfe-commits at lists.llvm.org
Thu Nov 21 10:48:43 PST 2024
================
@@ -205,14 +205,27 @@ llvm::MDNode *CodeGenTBAA::getTypeInfoHelper(const Type *Ty) {
llvm::MDNode *AnyPtr = createScalarTypeNode("any pointer", getChar(), Size);
if (!CodeGenOpts.PointerTBAA)
return AnyPtr;
- // Compute the depth of the pointer and generate a tag of the form "p<depth>
- // <base type tag>".
+ // C++ [basic.lval]p11 permits objects to accessed through an l-value of
+ // similar type. Two types are similar under C++ [conv.qual]p2 if the
+ // decomposition of the types into pointers, member pointers, and arrays has
+ // the same structure when ignoring cv-qualifiers at each level of the
+ // decomposition. Meanwhile, C makes T(*)[] and T(*)[N] compatible, which
+ // would really complicate any attempt to distinguish pointers to arrays by
+ // their bounds. It's simpler, and much easier to explain to users, to
+ // simply treat all pointers to arrays as pointers to their element type for
+ // aliasing purposes. So when creating a TBAA tag for a pointer type, we
+ // recursively ignore both qualifiers and array types when decomposing the
+ // pointee type. The only meaningful remaining structure is the number of
+ // pointer types we encountered along the way, so we just produce the tag
+ // "p<depth> <base type tag>". If we do find a member pointer type, for now
+ // we just conservatively bail out with AnyPtr (below) rather than trying to
+ // create a tag that honors the similar-type rules while still
+ // distinguishing different kinds of member pointer.
unsigned PtrDepth = 0;
do {
PtrDepth++;
- Ty = Ty->getPointeeType().getTypePtr();
- } while (Ty->isPointerType());
- Ty = Context.getBaseElementType(QualType(Ty, 0)).getTypePtr();
+ Ty = Ty->getPointeeType()->getBaseElementTypeUnsafe();
+ } while (Ty->isPointerType() || Ty->isArrayType());
----------------
fhahn wrote:
Removed, thanks.
https://github.com/llvm/llvm-project/pull/116991
More information about the cfe-commits
mailing list