[Clang] Freeze padded vectors before storing.

Currently Clang usually leaves padding bits uninitialized, which means
they are undef at the moment.

When expanding stores of vector types to include padding, the padding
lanes will be poison, hence the padding bits will be poison.

This interacts badly with coercion of arguments and return values, where
3 x float vectors will be loaded as i128 integer; poisoning the padding
bits will make the whole value poison.

Author: fhahn

clang/lib/CodeGen/CGExpr.cpp

@@ -2300,6 +2300,9 @@ void CodeGenFunction::EmitStoreOfScalar(llvm::Value *Value, Address Addr,
         SmallVector<int, 16> Mask(NewVecTy->getNumElements(), -1);
         std::iota(Mask.begin(), Mask.begin() + VecTy->getNumElements(), 0);
         Value = Builder.CreateShuffleVector(Value, Mask, "extractVec");
+        // The extra lanes will be poison. Freeze the whole vector to make sure
+        // the padding memory is not poisoned, which may break coercion.
+        Value = Builder.CreateFreeze(Value);
         SrcTy = NewVecTy;
       }
       if (Addr.getElementType() != SrcTy)

clang/test/CodeGen/AArch64/ext-vector-coercion.c

@@ -29,10 +29,11 @@ struct Vec3 {
 // CHECK-NEXT:    [[ADD:%.*]] = fadd <3 x float> [[EXTRACTVEC]], [[EXTRACTVEC2]]
 // CHECK-NEXT:    [[TMP2:%.*]] = getelementptr inbounds nuw [[STRUCT_VEC3]], ptr [[RETVAL]], i32 0, i32 0
 // CHECK-NEXT:    [[EXTRACTVEC3:%.*]] = shufflevector <3 x float> [[ADD]], <3 x float> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 poison>
-// CHECK-NEXT:    store <4 x float> [[EXTRACTVEC3]], ptr [[TMP2]], align 16
+// CHECK-NEXT:    [[TMP3:%.*]] = freeze <4 x float> [[EXTRACTVEC3]]
+// CHECK-NEXT:    store <4 x float> [[TMP3]], ptr [[TMP2]], align 16
 // CHECK-NEXT:    [[COERCE_DIVE4:%.*]] = getelementptr inbounds nuw [[STRUCT_VEC3]], ptr [[RETVAL]], i32 0, i32 0
-// CHECK-NEXT:    [[TMP3:%.*]] = load i128, ptr [[COERCE_DIVE4]], align 16
-// CHECK-NEXT:    ret i128 [[TMP3]]
+// CHECK-NEXT:    [[TMP4:%.*]] = load i128, ptr [[COERCE_DIVE4]], align 16
+// CHECK-NEXT:    ret i128 [[TMP4]]
 //
 struct Vec3 add(struct Vec3 a) {
   struct Vec3 res;