Fix various errors found with UBSan (microsoft#8184)

* `SPIRVOptions.h` - one of the fields of this struct (signaturePacking)
wasn't getting initialized before the struct's assignment operator was
used. This resulted in the uninitialized fields being copied, which is
UB. Although a more surgical fix is possible, it seemed better to just
ensure all fields are initialized, setting a good precedent if any new
fields are added.
* `FrontendOptions.h` - uninitialized field
* `APInt.cpp` - left shift of a signed integer
* `Triple.h` - uninitialized field
* `VirtualFileSystem.h` - uninitialized field
* `Lookup.h` - uninitialized field
* `MemoryBuffer.cpp` - avoid memcpy from a null dest (UB even if size is
0)

Methodology: had copilot CLI get a baseline running the tests under
ubsan, categorize the failures and fix them, and verify that gcc-14 and
clang 18 are now ubsan clan.

---------

Co-authored-by: Copilot <[email protected]>
Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>

Author: 3 people

include/dxc/Support/SPIRVOptions.h

@@ -39,40 +39,40 @@ enum class SpirvLayoutRule {
 
 struct SpirvCodeGenOptions {
   /// Disable legalization and optimization and emit raw SPIR-V
-  bool codeGenHighLevel;
-  bool debugInfoFile;
-  bool debugInfoLine;
-  bool debugInfoSource;
-  bool debugInfoTool;
-  bool debugInfoRich;
+  bool codeGenHighLevel = false;
+  bool debugInfoFile = false;
+  bool debugInfoLine = false;
+  bool debugInfoSource = false;
+  bool debugInfoTool = false;
+  bool debugInfoRich = false;
   /// Use NonSemantic.Vulkan.DebugInfo.100 debug info instead of
   /// OpenCL.DebugInfo.100
-  bool debugInfoVulkan;
-  bool defaultRowMajor;
-  bool disableValidation;
-  bool enable16BitTypes;
-  bool finiteMathOnly;
-  bool enableReflect;
-  bool invertY; // Additive inverse
-  bool invertW; // Multiplicative inverse
-  bool noWarnEmulatedFeatures;
-  bool noWarnIgnoredFeatures;
-  bool preserveBindings;
-  bool preserveInterface;
-  bool useDxLayout;
-  bool useGlLayout;
-  bool useLegacyBufferMatrixOrder;
-  bool useScalarLayout;
-  bool flattenResourceArrays;
-  bool reduceLoadSize;
-  bool autoShiftBindings;
-  bool supportNonzeroBaseInstance;
-  bool supportNonzeroBaseVertex;
-  bool fixFuncCallArguments;
-  bool enableMaximalReconvergence;
-  bool useVulkanMemoryModel;
-  bool useUnknownImageFormat;
-  bool IEEEStrict;
+  bool debugInfoVulkan = false;
+  bool defaultRowMajor = false;
+  bool disableValidation = false;
+  bool enable16BitTypes = false;
+  bool finiteMathOnly = false;
+  bool enableReflect = false;
+  bool invertY = false; // Additive inverse
+  bool invertW = false; // Multiplicative inverse
+  bool noWarnEmulatedFeatures = false;
+  bool noWarnIgnoredFeatures = false;
+  bool preserveBindings = false;
+  bool preserveInterface = false;
+  bool useDxLayout = false;
+  bool useGlLayout = false;
+  bool useLegacyBufferMatrixOrder = false;
+  bool useScalarLayout = false;
+  bool flattenResourceArrays = false;
+  bool reduceLoadSize = false;
+  bool autoShiftBindings = false;
+  bool supportNonzeroBaseInstance = false;
+  bool supportNonzeroBaseVertex = false;
+  bool fixFuncCallArguments = false;
+  bool enableMaximalReconvergence = false;
+  bool useVulkanMemoryModel = false;
+  bool useUnknownImageFormat = false;
+  bool IEEEStrict = false;
   /// Maximum length in words for the OpString literal containing the shader
   /// source for DebugSource and DebugSourceContinued. If the source code length
   /// is larger than this number, we will use DebugSourceContinued instructions
@@ -81,14 +81,14 @@ struct SpirvCodeGenOptions {
   /// limitation of a single SPIR-V instruction size (0xFFFF) - 2 operand words
   /// for OpString. Currently a smaller value is only used to test
   /// DebugSourceContinued generation.
-  uint32_t debugSourceLen;
-  SpirvLayoutRule cBufferLayoutRule;
-  SpirvLayoutRule sBufferLayoutRule;
-  SpirvLayoutRule tBufferLayoutRule;
-  SpirvLayoutRule ampPayloadLayoutRule;
+  uint32_t debugSourceLen = 0;
+  SpirvLayoutRule cBufferLayoutRule = SpirvLayoutRule::Void;
+  SpirvLayoutRule sBufferLayoutRule = SpirvLayoutRule::Void;
+  SpirvLayoutRule tBufferLayoutRule = SpirvLayoutRule::Void;
+  SpirvLayoutRule ampPayloadLayoutRule = SpirvLayoutRule::Void;
   llvm::StringRef stageIoOrder;
   llvm::StringRef targetEnv;
-  uint32_t maxId;
+  uint32_t maxId = 0;
   llvm::SmallVector<int32_t, 4> bShift;
   llvm::SmallVector<int32_t, 4> sShift;
   llvm::SmallVector<int32_t, 4> tShift;
@@ -109,9 +109,11 @@ struct SpirvCodeGenOptions {
   std::optional<BindingInfo> samplerHeapBinding;
   std::optional<BindingInfo> counterHeapBinding;
 
-  bool signaturePacking; ///< Whether signature packing is enabled or not
+  bool signaturePacking =
+      false; ///< Whether signature packing is enabled or not
 
-  bool printAll; // Dump SPIR-V module before each pass and after the last one.
+  bool printAll =
+      false; // Dump SPIR-V module before each pass and after the last one.
 
   // String representation of all command line options and input file.
   std::string clOptions;

include/llvm/ADT/Triple.h

@@ -193,7 +193,7 @@ class Triple {
   ArchType Arch;
 
   /// The parsed subarchitecture type.
-  SubArchType SubArch;
+  SubArchType SubArch = NoSubArch;
 
   /// The parsed vendor type.
   VendorType Vendor;

lib/Support/APInt.cpp

@@ -552,8 +552,10 @@ bool APInt::ult(const APInt& RHS) const {
 bool APInt::slt(const APInt& RHS) const {
   assert(BitWidth == RHS.BitWidth && "Bit widths must be same for comparison");
   if (isSingleWord()) {
-    int64_t lhsSext = (int64_t(VAL) << (64-BitWidth)) >> (64-BitWidth);
-    int64_t rhsSext = (int64_t(RHS.VAL) << (64-BitWidth)) >> (64-BitWidth);
+    int64_t lhsSext =
+        int64_t(uint64_t(VAL) << (64 - BitWidth)) >> (64 - BitWidth);
+    int64_t rhsSext =
+        int64_t(uint64_t(RHS.VAL) << (64 - BitWidth)) >> (64 - BitWidth);
     return lhsSext < rhsSext;
   }
 
@@ -1061,8 +1063,8 @@ APInt APInt::ashr(unsigned shiftAmt) const {
       return APInt(BitWidth, 0); // undefined
     else {
       unsigned SignBit = APINT_BITS_PER_WORD - BitWidth;
-      return APInt(BitWidth,
-        (((int64_t(VAL) << SignBit) >> SignBit) >> shiftAmt));
+      return APInt(BitWidth, (((int64_t(uint64_t(VAL) << SignBit) >> SignBit) >>
+                               shiftAmt)));
     }
   }
 

lib/Support/MemoryBuffer.cpp

@@ -126,8 +126,9 @@ MemoryBuffer::getMemBufferCopy(StringRef InputData, const Twine &BufferName) {
       getNewUninitMemBuffer(InputData.size(), BufferName);
   if (!Buf)
     return nullptr;
-  memcpy(const_cast<char*>(Buf->getBufferStart()), InputData.data(),
-         InputData.size());
+  if (InputData.size())
+    memcpy(const_cast<char *>(Buf->getBufferStart()), InputData.data(),
+           InputData.size());
   return Buf;
 }
 

tools/clang/include/clang/Basic/VirtualFileSystem.h

@@ -40,10 +40,13 @@ class Status {
   llvm::sys::fs::perms Perms;
 
 public:
-  bool IsVFSMapped; // FIXME: remove when files support multiple names
+  bool IsVFSMapped = false; // FIXME: remove when files support multiple names
 
 public:
-  Status() : Type(llvm::sys::fs::file_type::status_error) {}
+  Status()
+      : User(0), Group(0), Size(0),
+        Type(llvm::sys::fs::file_type::status_error),
+        Perms(llvm::sys::fs::perms_not_known) {}
   Status(const llvm::sys::fs::file_status &Status);
   Status(StringRef Name, StringRef RealName, llvm::sys::fs::UniqueID UID,
          llvm::sys::TimeValue MTime, uint32_t User, uint32_t Group,

tools/clang/include/clang/Frontend/FrontendOptions.h

@@ -92,7 +92,7 @@ class FrontendInputFile {
   bool IsSystem;
 
 public:
-  FrontendInputFile() : Buffer(nullptr), Kind(IK_None) { }
+  FrontendInputFile() : Buffer(nullptr), Kind(IK_None), IsSystem(false) {}
   FrontendInputFile(StringRef File, InputKind Kind, bool IsSystem = false)
     : File(File.str()), Buffer(nullptr), Kind(Kind), IsSystem(IsSystem) { }
   FrontendInputFile(llvm::MemoryBuffer *buffer, InputKind Kind,

tools/clang/include/clang/Sema/Lookup.h

@@ -654,7 +654,7 @@ class LookupResult {
 
   // Results.
   LookupResultKind ResultKind;
-  AmbiguityKind Ambiguity; // ill-defined unless ambiguous
+  AmbiguityKind Ambiguity = {}; // ill-defined unless ambiguous
   UnresolvedSet<8> Decls;
   CXXBasePaths *Paths;
   CXXRecordDecl *NamingClass;