[SYCL] Fixes to issues reported by a static analyzer run

sycl-jit/common/include/Kernel.h

@@ -380,7 +380,7 @@ struct FrozenPropertyValue {
   FrozenPropertyValue(std::string_view Name, uint32_t Value)
       : Name{Name}, IsUIntValue{true}, UIntValue{Value}, Bytes{0} {}
   FrozenPropertyValue(std::string_view Name, const uint8_t *Ptr, size_t Size)
-      : Name{Name}, IsUIntValue{false}, Bytes{Size} {
+      : Name{Name}, IsUIntValue{false}, UIntValue{0}, Bytes{Size} {
     std::memcpy(Bytes.begin(), Ptr, Size);
   }
 };

sycl-jit/jit-compiler/lib/translation/KernelTranslation.cpp

@@ -133,8 +133,13 @@ KernelTranslator::loadKernels(llvm::LLVMContext &LLVMCtx,
         // read last. This could cause problems if different modules contain
         // definitions with the same name, but different body/content.
         // Check that this is not problematic.
-        Linker::linkModules(*Result, std::move(NewMod),
-                            Linker::Flags::OverrideFromSrc);
+        const bool HasErrors = Linker::linkModules(
+            *Result, std::move(NewMod), Linker::Flags::OverrideFromSrc);
+        if (HasErrors) {
+          return createStringError(inconvertibleErrorCode(),
+                                   "Failed to link modules");
+        }
+
         if (AddressBits != BinInfo.AddressBits) {
           return createStringError(
               inconvertibleErrorCode(),
@@ -300,9 +305,9 @@ KernelTranslator::translateToPTX(SYCLKernelInfo &KernelInfo, llvm::Module &Mod,
   }
 
   // FIXME: Check whether we can provide more accurate target information here
-  auto *TargetMachine = Target->createTargetMachine(
+  std::unique_ptr<TargetMachine> TargetMachine(Target->createTargetMachine(
       TargetTriple, CPU, Features, {}, llvm::Reloc::PIC_, std::nullopt,
-      llvm::CodeGenOptLevel::Default);
+      llvm::CodeGenOptLevel::Default));
 
   llvm::legacy::PassManager PM;
 
@@ -380,9 +385,9 @@ KernelTranslator::translateToAMDGCN(SYCLKernelInfo &KernelInfo,
   }
 
   // FIXME: Check whether we can provide more accurate target information here
-  auto *TargetMachine = Target->createTargetMachine(
+  std::unique_ptr<TargetMachine> TargetMachine(Target->createTargetMachine(
       TargetTriple, CPU, Features, {}, llvm::Reloc::PIC_, std::nullopt,
-      llvm::CodeGenOptLevel::Default);
+      llvm::CodeGenOptLevel::Default));
 
   std::string AMDObj;
   {

sycl-jit/passes/target/TargetFusionInfo.cpp

@@ -458,6 +458,7 @@ class NVPTXAMDGCNTargetFusionInfoBase : public TargetFusionInfoImpl {
     SmallPtrSet<Constant *, 8> DeletedFuncs{Funcs.begin(), Funcs.end()};
     SmallVector<MDNode *> ValidKernels;
     auto *OldAnnotations = LLVMMod->getNamedMetadata(MDName);
+    assert(OldAnnotations && "Failed to retrieve old annotations");
     for (auto *Op : OldAnnotations->operands()) {
       if (auto *TOp = dyn_cast<MDTuple>(Op)) {
         if (auto *COp = dyn_cast_if_present<ConstantAsMetadata>(