[Sema] Fix type mismatch error when arguments to elementwise math builtin have different qualifiers, which should be well-formed

[actinks]

Fixes #141397
Element-wise math builtins (e.g. __builtin_elementwise_max/__builtin_elementwise_pow etc.) fail when their arguments have different qualifications, but should be well-formed. The fix is ​​to use hasSameUnqualifiedType to check if the arguments match.

[llvmbot]

@llvm/pr-subscribers-clang

Author: QiYue (QiYueFeiXue)

Changes

Fixes #141397

---

Full diff: https://github.com/llvm/llvm-project/pull/141485.diff

2 Files Affected:

• (modified) clang/lib/Sema/SemaChecking.cpp (+5-5)
• (added) clang/test/SemaCXX/bug141397.cpp (+16)

diff --git a/clang/lib/Sema/SemaChecking.cpp b/clang/lib/Sema/SemaChecking.cpp
index 930e9083365a1..61aeffde338d8 100644
--- a/clang/lib/Sema/SemaChecking.cpp
+++ b/clang/lib/Sema/SemaChecking.cpp
@@ -2907,7 +2907,7 @@ Sema::CheckBuiltinFunctionCall(FunctionDecl *FDecl, unsigned BuiltinID,
       return ExprError();
     }
 
-    if (MagnitudeTy.getCanonicalType() != SignTy.getCanonicalType()) {
+    if (!Context.hasSameUnqualifiedType(MagnitudeTy, SignTy)) {
       return Diag(Sign.get()->getBeginLoc(),
                   diag::err_typecheck_call_different_arg_types)
              << MagnitudeTy << SignTy;
@@ -5265,7 +5265,7 @@ bool Sema::BuiltinComplex(CallExpr *TheCall) {
 
   Expr *Real = TheCall->getArg(0);
   Expr *Imag = TheCall->getArg(1);
-  if (!Context.hasSameType(Real->getType(), Imag->getType())) {
+  if (!Context.hasSameUnqualifiedType(Real->getType(), Imag->getType())) {
     return Diag(Real->getBeginLoc(),
                 diag::err_typecheck_call_different_arg_types)
            << Real->getType() << Imag->getType()
@@ -15568,7 +15568,7 @@ Sema::BuiltinVectorMath(CallExpr *TheCall,
   if (checkMathBuiltinElementType(*this, LocA, TyA, ArgTyRestr, 1))
     return std::nullopt;
 
-  if (TyA.getCanonicalType() != TyB.getCanonicalType()) {
+  if (!Context.hasSameUnqualifiedType(TyA, TyB)) {
     Diag(LocA, diag::err_typecheck_call_different_arg_types) << TyA << TyB;
     return std::nullopt;
   }
@@ -15607,8 +15607,8 @@ bool Sema::BuiltinElementwiseTernaryMath(
   }
 
   for (int I = 1; I < 3; ++I) {
-    if (Args[0]->getType().getCanonicalType() !=
-        Args[I]->getType().getCanonicalType()) {
+    if (!Context.hasSameUnqualifiedType(Args[0]->getType(),
+                                        Args[I]->getType())) {
       return Diag(Args[0]->getBeginLoc(),
                   diag::err_typecheck_call_different_arg_types)
              << Args[0]->getType() << Args[I]->getType();
diff --git a/clang/test/SemaCXX/bug141397.cpp b/clang/test/SemaCXX/bug141397.cpp
new file mode 100644
index 0000000000000..cb9f13a93f955
--- /dev/null
+++ b/clang/test/SemaCXX/bug141397.cpp
@@ -0,0 +1,16 @@
+// RUN: %clang_cc1 %s -fsyntax-only -verify
+// expected-no-diagnostics
+
+// This example uncovered a bug in Sema::BuiltinVectorMath, where we should be
+// using ASTContext::hasSameUnqualifiedType().
+
+typedef float vec3 __attribute__((ext_vector_type(3)));
+
+typedef struct {
+  vec3 b;
+} struc;
+
+vec3 foo(vec3 a,const struc* hi) {
+  vec3 b = __builtin_elementwise_max((vec3)(0.0f), a);
+  return __builtin_elementwise_pow(b, hi->b.yyy);
+}

[cor3ntin]

This change needs a release note.
Please add an entry to clang/docs/ReleaseNotes.rst in the section the most adapted to the change, and referencing any Github issue this change fixes. Thanks!

[shafik]

Thank you for the fix.

Please put more details in your summary. Something along the lines of "builtin functions such as __builtin_elementwise_max .... fail when their arguments have different qualifications but this should be well-formed ... the fix involves using hasSameUnqualifiedType to check..."

It is important to have detailed summaries for the reviewers so they have more context without having to leave the code review and for folks using git log to debug issues when their downstream fork has issues.

[zyn0217]

@QiYueFeiXue I don't think @shafik means to directly copy his words to the PR body/title. Please at least flesh out these dotted parts and make the sentences complete, thanks.

[zyn0217]

Missmach -> mismatch (and same in the test!)

A well-formed -> well-formed

[actinks]

done, thanks!

[shafik]

@QiYueFeiXue I don't think @shafik means to directly copy his words to the PR body/title. Please at least flesh out these dotted parts and make the sentences complete, thanks.

Yes, that is what I meant, thank you for clarifying my intent.

[shafik]

@Acthink-Yang Do you have commit access? If not then let us know and one of us can merge for you.

[actinks]

Hi @shafik

Thanks for checking! I don't have commit access yet.
Could you please help merge my changes?