[CIR][CIRGen][TBAA] Initial TBAA support for union, enum and bitint

clang/include/clang/CIR/MissingFeatures.h

@@ -60,6 +60,8 @@ struct MissingFeatures {
   static bool tbaa() { return false; }
   static bool tbaaStruct() { return false; }
   static bool tbaaTagForStruct() { return false; }
+  static bool tbaaTagForEnum() { return false; }
+  static bool tbaaTagForBitInt() { return false; }
   static bool tbaaVTablePtr() { return false; }
   static bool tbaaIncompleteType() { return false; }
   static bool tbaaMergeTBAAInfo() { return false; }

clang/lib/CIR/CodeGen/CIRGenTBAA.cpp

@@ -5,6 +5,7 @@
 #include "mlir/Interfaces/DataLayoutInterfaces.h"
 #include "clang/AST/ASTContext.h"
 #include "clang/AST/RecordLayout.h"
+#include "clang/AST/Type.h"
 #include "clang/CIR/Dialect/IR/CIRTypes.h"
 #include "clang/CIR/MissingFeatures.h"
 #include "llvm/Support/ErrorHandling.h"
@@ -61,6 +62,58 @@ static bool isValidBaseType(clang::QualType qty) {
   return false;
 }
 
+cir::TBAAAttr CIRGenTBAA::getTypeInfoHelper(clang::QualType qty) {
+  const clang::Type *ty = astContext.getCanonicalType(qty).getTypePtr();
+  // Handle builtin types.
+  if (mlir::isa<BuiltinType>(ty)) {
+    return cir::TBAAScalarAttr::get(mlirContext, types.ConvertType(qty));
+  }
+  // C++1z [basic.lval]p10: "If a program attempts to access the stored value of
+  // an object through a glvalue of other than one of the following types the
+  // behavior is undefined: [...] a char, unsigned char, or std::byte type."
+  if (ty->isStdByteType())
+    return getChar();
+
+  // Handle pointers and references.
+  //
+  // C has a very strict rule for pointer aliasing. C23 6.7.6.1p2:
+  //     For two pointer types to be compatible, both shall be identically
+  //     qualified and both shall be pointers to compatible types.
+  //
+  // This rule is impractically strict; we want to at least ignore CVR
+  // qualifiers. Distinguishing by CVR qualifiers would make it UB to
+  // e.g. cast a `char **` to `const char * const *` and dereference it,
+  // which is too common and useful to invalidate. C++'s similar types
+  // rule permits qualifier differences in these nested positions; in fact,
+  // C++ even allows that cast as an implicit conversion.
+  //
+  // Other qualifiers could theoretically be distinguished, especially if
+  // they involve a significant representation difference.  We don't
+  // currently do so, however.
+  if (ty->isPointerType() || ty->isReferenceType()) {
+    if (!codeGenOpts.PointerTBAA) {
+      return cir::TBAAScalarAttr::get(mlirContext, types.ConvertType(qty));
+    }
+    llvm_unreachable("NYI");
+  }
+  // Accesses to arrays are accesses to objects of their element types.
+  if (codeGenOpts.NewStructPathTBAA && ty->isArrayType()) {
+    llvm_unreachable("NYI");
+  }
+  // Enum types are distinct types. In C++ they have "underlying types",
+  // however they aren't related for TBAA.
+  if (const EnumType *ety = dyn_cast<EnumType>(ty)) {
+    assert(!cir::MissingFeatures::tbaaTagForEnum());
+    return tbaa_NYI(mlirContext);
+  }
+  if (const auto *eit = dyn_cast<BitIntType>(ty)) {
+    assert(!cir::MissingFeatures::tbaaTagForBitInt());
+    return tbaa_NYI(mlirContext);
+  }
+  // For now, handle any other kind of type conservatively.
+  return getChar();
+}
+
 cir::TBAAAttr CIRGenTBAA::getTypeInfo(clang::QualType qty) {
   // At -O0 or relaxed aliasing, TBAA is not emitted for regular types.
   if (codeGenOpts.OptimizationLevel == 0 || codeGenOpts.RelaxedAliasing) {
@@ -93,7 +146,7 @@ cir::TBAAAttr CIRGenTBAA::getTypeInfo(clang::QualType qty) {
   // cache, which invalidates all its previously obtained iterators. So we
   // first generate the node for the type and then add that node to the
   // cache.
-  auto typeNode = cir::TBAAScalarAttr::get(mlirContext, types.ConvertType(qty));
+  auto typeNode = getTypeInfoHelper(qty);
   return metadataCache[ty] = typeNode;
 }
 

clang/lib/CIR/CodeGen/CIRGenTBAA.h

@@ -108,6 +108,10 @@ class CIRGenTBAA {
 
   cir::TBAAAttr getChar();
 
+  // An internal helper function to generate metadata used
+  // to describe accesses to objects of the given type.
+  cir::TBAAAttr getTypeInfoHelper(clang::QualType qty);
+
 public:
   CIRGenTBAA(mlir::MLIRContext *mlirContext, clang::ASTContext &astContext,
              CIRGenTypes &types, mlir::ModuleOp moduleOp,

clang/test/CIR/CodeGen/tbaa-bitinit.c

@@ -0,0 +1,13 @@
+// RUN: %clang_cc1 -triple x86_64-unknown-linux-gnu -fclangir -emit-cir %s -o %t.cir -O1
+// RUN: FileCheck --check-prefix=CIR --input-file=%t.cir %s
+
+
+// CIR: #tbaa[[tbaa_NYI:.*]] = #cir.tbaa
+
+_BitInt(33) a;
+_BitInt(31) b;
+void c() {
+  // CIR: %{{.*}} = cir.load %{{.*}} : !cir.ptr<!cir.int<s, 33>>, !cir.int<s, 33> tbaa(#tbaa[[tbaa_NYI]])
+  // CIR: cir.store %{{.*}}, %{{.*}} : !cir.int<s, 31>, !cir.ptr<!cir.int<s, 31>> tbaa(#tbaa[[tbaa_NYI]])
+  b = a;
+}

clang/test/CIR/CodeGen/tbaa-enum.c

@@ -0,0 +1,49 @@
+// RUN: %clang_cc1 -triple x86_64-unknown-linux-gnu -fclangir -emit-cir %s -o %t.cir -O1
+// RUN: FileCheck --check-prefix=CIR --input-file=%t.cir %s
+// RUN: %clang_cc1 -triple x86_64-unknown-linux-gnu -fclangir -emit-llvm %s -o %t.ll -O1 -disable-llvm-passes
+// RUN: FileCheck --check-prefix=LLVM --input-file=%t.ll %s
+// RUN: %clang_cc1 -triple x86_64-unknown-linux-gnu -fclangir -emit-llvm %s -o %t.ll -O1 -disable-llvm-passes -relaxed-aliasing
+// RUN: FileCheck --check-prefix=NO-TBAA --input-file=%t.ll %s
+// RUN: %clang_cc1 -triple x86_64-unknown-linux-gnu -fclangir -emit-llvm %s -o %t.ll -O0 -disable-llvm-passes
+// RUN: FileCheck --check-prefix=NO-TBAA --input-file=%t.ll %s
+
+// NO-TBAA-NOT: !tbaa
+
+// CIR: #tbaa[[tbaa_NYI:.*]] = #cir.tbaa
+// CIR: #tbaa[[INT:.*]] = #cir.tbaa_scalar<type = !u32i>
+// CIR: #tbaa[[INT_PTR:.*]] = #cir.tbaa_scalar<type = !cir.ptr<!u32i>>
+
+typedef unsigned int uint32_t;
+typedef enum {
+  RED_AUTO_32,
+  GREEN_AUTO_32,
+  BLUE_AUTO_32
+} EnumAuto32;
+
+uint32_t g0(EnumAuto32 *E, uint32_t *val) {
+  // CIR-LABEL: cir.func @g0
+  // CIR: %[[C5:.*]] = cir.const #cir.int<5> : !s32i
+  // CIR: %[[U_C5:.*]] = cir.cast(integral, %[[C5]] : !s32i), !u32i
+  // CIR: %[[VAL_PTR:.*]] = cir.load deref %{{.*}} : !cir.ptr<!cir.ptr<!u32i>>, !cir.ptr<!u32i> tbaa(#tbaa[[INT_PTR]])
+  // CIR: cir.store %[[U_C5]], %[[VAL_PTR]] : !u32i, !cir.ptr<!u32i> tbaa(#tbaa[[INT]])
+  // CIR: %[[C0:.*]] = cir.const #cir.int<0> : !s32i
+  // CIR: %[[U_C0:.*]] = cir.cast(integral, %[[C0]] : !s32i), !u32i
+  // CIR: %[[E_PTR:.*]] = cir.load deref %{{.*}} : !cir.ptr<!cir.ptr<!u32i>>, !cir.ptr<!u32i> tbaa(#tbaa[[INT_PTR]])
+  // CIR: cir.store %[[U_C0]], %[[E_PTR]] : !u32i, !cir.ptr<!u32i> tbaa(#tbaa[[tbaa_NYI]])
+  // CIR: %[[RET_PTR:.*]] = cir.load deref %{{.*}} : !cir.ptr<!cir.ptr<!u32i>>, !cir.ptr<!u32i> tbaa(#tbaa[[INT_PTR]])
+  // CIR: %[[RET:.*]] = cir.load %[[RET_PTR]] : !cir.ptr<!u32i>, !u32i tbaa(#tbaa[[INT]])
+  // CIR: cir.store %[[RET]], %{{.*}} : !u32i, !cir.ptr<!u32i>
+
+  // LLVM-LABEL: define{{.*}} i32 @g0(
+  // LLVM: store i32 5, ptr %{{.*}}, align 4, !tbaa [[TAG_i32:!.*]]
+  // LLVM: store i32 0, ptr %{{.*}}, align 4
+  // LLVM: load i32, ptr %{{.*}}, align 4, !tbaa [[TAG_i32]]
+  *val = 5;
+  *E = RED_AUTO_32;
+  return *val;
+}
+
+// LLVM: [[TYPE_char:!.*]] = !{!"omnipotent char", [[TAG_c_tbaa:!.*]],
+// LLVM: [[TAG_c_tbaa]] = !{!"Simple C/C++ TBAA"}
+// LLVM: [[TAG_i32]] = !{[[TYPE_i32:!.*]], [[TYPE_i32]], i64 0}
+// LLVM: [[TYPE_i32]] = !{!"int", [[TYPE_char]], i64 0}

clang/test/CIR/CodeGen/tbaa-union.c

@@ -0,0 +1,33 @@
+// RUN: %clang_cc1 -triple x86_64-unknown-linux-gnu -fclangir -emit-cir %s -o %t.cir -O1
+// RUN: FileCheck --check-prefix=CIR --input-file=%t.cir %s
+// RUN: %clang_cc1 -triple x86_64-unknown-linux-gnu -fclangir -emit-llvm %s -o %t.ll -O1
+// RUN: FileCheck --check-prefix=LLVM --input-file=%t.ll %s
+// RUN: %clang_cc1 -triple x86_64-unknown-linux-gnu -fclangir -emit-llvm %s -o %t.ll -O1 -relaxed-aliasing
+// RUN: FileCheck --check-prefix=NO-TBAA --input-file=%t.ll %s
+// RUN: %clang_cc1 -triple x86_64-unknown-linux-gnu -fclangir -emit-llvm %s -o %t.ll -O0
+// RUN: FileCheck --check-prefix=NO-TBAA --input-file=%t.ll %s
+
+// NO-TBAA-NOT: !tbaa
+// CIR: #tbaa[[CHAR:.*]] = #cir.tbaa_scalar<type = !cir.int<s, 1>>
+// CIR: #tbaa[[Struct_S_PTR:.*]] = #cir.tbaa_scalar<type = !cir.ptr<!ty_S>>
+typedef struct {
+  union {
+    int a, b;
+  };
+  int c;
+} S;
+
+void foo(S *s) {
+  // CIR-LABEL: cir.func @foo
+  // CIR: %[[C1:.*]] = cir.const #cir.int<1> : !s32i loc(#loc6)
+  // CIR: %{{.*}} = cir.load %{{.*}} : !cir.ptr<!cir.ptr<!ty_S>>, !cir.ptr<!ty_S> tbaa(#tbaa[[Struct_S_PTR]])
+  // CIR: cir.store %[[C1]], %{{.*}} : !s32i, !cir.ptr<!s32i> tbaa(#tbaa[[CHAR]])
+
+  // LLVM-LABEL: void @foo
+  // LLVM: store i32 1, ptr %{{.*}}, align 4, !tbaa ![[TBAA_TAG:.*]]
+  s->a = 1;
+}
+
+// LLVM: ![[TBAA_TAG]] = !{![[CHAR:.*]], ![[CHAR]], i64 0}
+// LLVM: ![[CHAR]] = !{!"omnipotent char", ![[ROOT:.*]], i64 0}
+// LLVM: ![[ROOT]] = !{!"Simple C/C++ TBAA"}