[CIR] Upstream handling for BaseToDerived casts

[andykaylor]

Upstream handling for BaseToDerived casts, adding the cir.base_class_addr operation and lowering to LLVM IR.

[llvmbot]

@llvm/pr-subscribers-clangir

@llvm/pr-subscribers-clang

Author: Andy Kaylor (andykaylor)

Changes

Upstream handling for BaseToDerived casts, adding the cir.base_class_addr operation and lowering to LLVM IR.

---

Full diff: https://github.com/llvm/llvm-project/pull/167769.diff

8 Files Affected:

• (modified) clang/include/clang/CIR/Dialect/IR/CIROps.td (+48)
• (modified) clang/lib/CIR/CodeGen/CIRGenBuilder.h (+13)
• (modified) clang/lib/CIR/CodeGen/CIRGenClass.cpp (+19)
• (modified) clang/lib/CIR/CodeGen/CIRGenExpr.cpp (+16-1)
• (modified) clang/lib/CIR/CodeGen/CIRGenExprScalar.cpp (+14-1)
• (modified) clang/lib/CIR/CodeGen/CIRGenFunction.h (+5)
• (modified) clang/lib/CIR/Lowering/DirectToLLVM/LowerToLLVM.cpp (+35)
• (added) clang/test/CIR/CodeGen/base-to-derived.cpp (+97)

diff --git a/clang/include/clang/CIR/Dialect/IR/CIROps.td b/clang/include/clang/CIR/Dialect/IR/CIROps.td
index 16258513239d9..4c62f8a36c9ae 100644
--- a/clang/include/clang/CIR/Dialect/IR/CIROps.td
+++ b/clang/include/clang/CIR/Dialect/IR/CIROps.td
@@ -3369,6 +3369,54 @@ def CIR_BaseClassAddrOp : CIR_Op<"base_class_addr"> {
   }];
 }
 
+//===----------------------------------------------------------------------===//
+// DerivedClassAddrOp
+//===----------------------------------------------------------------------===//
+
+def CIR_DerivedClassAddrOp : CIR_Op<"derived_class_addr"> {
+  let summary = "Get the derived class address for a class/struct";
+  let description = [{
+    The `cir.derived_class_addr` operaration gets the address of a particular
+    derived class given a non-virtual base class pointer. The offset in bytes
+    of the base class must be passed in, similar to `cir.base_class_addr`, but
+    going into the other direction. This means lowering to a negative offset.
+
+    The operation contains a flag for whether or not the operand may be nullptr.
+    That depends on the context and cannot be known by the operation, and that
+    information affects how the operation is lowered.
+
+    Example:
+    ```c++
+    class A {};
+    class B : public A {};
+
+    B *getAsB(A *a) {
+      return static_cast<B*>(a);
+    }
+    ```
+
+    leads to
+    ```mlir
+      %2 = cir.load %0 : !cir.ptr<!cir.ptr<!rec_A>>, !cir.ptr<!rec_A>
+      %3 = cir.base_class_addr %2 : !cir.ptr<!rec_B> [0] -> !cir.ptr<!rec_A>
+    ```
+  }];
+
+  // The validity of the relationship of derived and base cannot yet be
+  // verified, currently not worth adding a verifier.
+  let arguments = (ins
+    Arg<CIR_PointerType, "base class pointer", [MemRead]>:$base_addr,
+    IndexAttr:$offset, UnitAttr:$assume_not_null);
+
+  let results = (outs Res<CIR_PointerType, "">:$derived_addr);
+
+  let assemblyFormat = [{
+      $base_addr `:` qualified(type($base_addr))
+      (`nonnull` $assume_not_null^)?
+      ` ` `[` $offset `]` `->` qualified(type($derived_addr)) attr-dict
+  }];
+}
+
 //===----------------------------------------------------------------------===//
 // ComplexCreateOp
 //===----------------------------------------------------------------------===//
diff --git a/clang/lib/CIR/CodeGen/CIRGenBuilder.h b/clang/lib/CIR/CodeGen/CIRGenBuilder.h
index e5066fac19185..675babc229701 100644
--- a/clang/lib/CIR/CodeGen/CIRGenBuilder.h
+++ b/clang/lib/CIR/CodeGen/CIRGenBuilder.h
@@ -403,6 +403,19 @@ class CIRGenBuilderTy : public cir::CIRBaseBuilderTy {
     return Address(baseAddr, destType, addr.getAlignment());
   }
 
+  Address createDerivedClassAddr(mlir::Location loc, Address addr,
+                                 mlir::Type destType, unsigned offset,
+                                 bool assumeNotNull) {
+    if (destType == addr.getElementType())
+      return addr;
+
+    cir::PointerType ptrTy = getPointerTo(destType);
+    auto derivedAddr =
+        cir::DerivedClassAddrOp::create(*this, loc, ptrTy, addr.getPointer(),
+                                        mlir::APInt(64, offset), assumeNotNull);
+    return Address(derivedAddr, destType, addr.getAlignment());
+  }
+
   mlir::Value createVTTAddrPoint(mlir::Location loc, mlir::Type retTy,
                                  mlir::Value addr, uint64_t offset) {
     return cir::VTTAddrPointOp::create(*this, loc, retTy,
diff --git a/clang/lib/CIR/CodeGen/CIRGenClass.cpp b/clang/lib/CIR/CodeGen/CIRGenClass.cpp
index a8296782ebc40..89c4696b9da94 100644
--- a/clang/lib/CIR/CodeGen/CIRGenClass.cpp
+++ b/clang/lib/CIR/CodeGen/CIRGenClass.cpp
@@ -1110,6 +1110,25 @@ mlir::Value CIRGenFunction::getVTTParameter(GlobalDecl gd, bool forVirtualBase,
   }
 }
 
+Address CIRGenFunction::getAddressOfDerivedClass(
+    mlir::Location loc, Address baseAddr, const CXXRecordDecl *derived,
+    llvm::iterator_range<CastExpr::path_const_iterator> path,
+    bool nullCheckValue) {
+  assert(!path.empty() && "Base path should not be empty!");
+
+  QualType derivedTy = getContext().getCanonicalTagType(derived);
+  mlir::Type derivedValueTy = convertType(derivedTy);
+  CharUnits nonVirtualOffset =
+      cgm.computeNonVirtualBaseClassOffset(derived, path);
+
+  // Note that in OG, no offset (nonVirtualOffset.getQuantity() == 0) means it
+  // just gives the address back. In CIR a `cir.derived_class` is created and
+  // made into a nop later on during lowering.
+  return builder.createDerivedClassAddr(loc, baseAddr, derivedValueTy,
+                                        nonVirtualOffset.getQuantity(),
+                                        /*assumeNotNull=*/!nullCheckValue);
+}
+
 Address CIRGenFunction::getAddressOfBaseClass(
     Address value, const CXXRecordDecl *derived,
     llvm::iterator_range<CastExpr::path_const_iterator> path,
diff --git a/clang/lib/CIR/CodeGen/CIRGenExpr.cpp b/clang/lib/CIR/CodeGen/CIRGenExpr.cpp
index 9bb76894c13f1..dc068c2864cf7 100644
--- a/clang/lib/CIR/CodeGen/CIRGenExpr.cpp
+++ b/clang/lib/CIR/CodeGen/CIRGenExpr.cpp
@@ -1204,7 +1204,6 @@ LValue CIRGenFunction::emitCastLValue(const CastExpr *e) {
   case CK_NonAtomicToAtomic:
   case CK_AtomicToNonAtomic:
   case CK_ToUnion:
-  case CK_BaseToDerived:
   case CK_AddressSpaceConversion:
   case CK_ObjCObjectLValueCast:
   case CK_VectorSplat:
@@ -1220,6 +1219,22 @@ LValue CIRGenFunction::emitCastLValue(const CastExpr *e) {
     return {};
   }
 
+  case CK_BaseToDerived: {
+    const auto *derivedClassDecl = e->getType()->castAsCXXRecordDecl();
+    LValue lv = emitLValue(e->getSubExpr());
+
+    // Perform the base-to-derived conversion
+    Address derived = getAddressOfDerivedClass(
+        getLoc(e->getSourceRange()), lv.getAddress(), derivedClassDecl,
+        e->path(), /*NullCheckValue=*/false);
+    // C++11 [expr.static.cast]p2: Behavior is undefined if a downcast is
+    // performed and the object is not of the derived type.
+    assert(!cir::MissingFeatures::sanitizers());
+
+    assert(!cir::MissingFeatures::opTBAA());
+    return makeAddrLValue(derived, e->getType(), lv.getBaseInfo());
+  }
+
   case CK_LValueBitCast: {
     // This must be a reinterpret_cast (or c-style equivalent).
     const auto *ce = cast<ExplicitCastExpr>(e);
diff --git a/clang/lib/CIR/CodeGen/CIRGenExprScalar.cpp b/clang/lib/CIR/CodeGen/CIRGenExprScalar.cpp
index 4461875fcf678..f5f59050f8bd4 100644
--- a/clang/lib/CIR/CodeGen/CIRGenExprScalar.cpp
+++ b/clang/lib/CIR/CodeGen/CIRGenExprScalar.cpp
@@ -1933,6 +1933,20 @@ mlir::Value ScalarExprEmitter::VisitCastExpr(CastExpr *ce) {
     return builder.createIntToPtr(middleVal, destCIRTy);
   }
 
+  case CK_BaseToDerived: {
+    const CXXRecordDecl *derivedClassDecl = destTy->getPointeeCXXRecordDecl();
+    assert(derivedClassDecl && "BaseToDerived arg isn't a C++ object pointer!");
+    Address base = cgf.emitPointerWithAlignment(subExpr);
+    Address derived = cgf.getAddressOfDerivedClass(
+        cgf.getLoc(ce->getSourceRange()), base, derivedClassDecl, ce->path(),
+        cgf.shouldNullCheckClassCastValue(ce));
+
+    // C++11 [expr.static.cast]p11: Behavior is undefined if a downcast is
+    // performed and the object is not of the derived type.
+    assert(!cir::MissingFeatures::sanitizers());
+
+    return cgf.getAsNaturalPointerTo(derived, ce->getType()->getPointeeType());
+  }
   case CK_UncheckedDerivedToBase:
   case CK_DerivedToBase: {
     // The EmitPointerWithAlignment path does this fine; just discard
@@ -1940,7 +1954,6 @@ mlir::Value ScalarExprEmitter::VisitCastExpr(CastExpr *ce) {
     return cgf.getAsNaturalPointerTo(cgf.emitPointerWithAlignment(ce),
                                      ce->getType()->getPointeeType());
   }
-
   case CK_Dynamic: {
     Address v = cgf.emitPointerWithAlignment(subExpr);
     const auto *dce = cast<CXXDynamicCastExpr>(ce);
diff --git a/clang/lib/CIR/CodeGen/CIRGenFunction.h b/clang/lib/CIR/CodeGen/CIRGenFunction.h
index cab4a3d4d3b25..8f47b08219301 100644
--- a/clang/lib/CIR/CodeGen/CIRGenFunction.h
+++ b/clang/lib/CIR/CodeGen/CIRGenFunction.h
@@ -816,6 +816,11 @@ class CIRGenFunction : public CIRGenTypeCache {
       llvm::iterator_range<CastExpr::path_const_iterator> path,
       bool nullCheckValue, SourceLocation loc);
 
+  Address getAddressOfDerivedClass(
+      mlir::Location loc, Address baseAddr, const CXXRecordDecl *derived,
+      llvm::iterator_range<CastExpr::path_const_iterator> path,
+      bool nullCheckValue);
+
   /// Return the VTT parameter that should be passed to a base
   /// constructor/destructor with virtual bases.
   /// FIXME: VTTs are Itanium ABI-specific, so the definition should move
diff --git a/clang/lib/CIR/Lowering/DirectToLLVM/LowerToLLVM.cpp b/clang/lib/CIR/Lowering/DirectToLLVM/LowerToLLVM.cpp
index b4afed7019417..991c18003f055 100644
--- a/clang/lib/CIR/Lowering/DirectToLLVM/LowerToLLVM.cpp
+++ b/clang/lib/CIR/Lowering/DirectToLLVM/LowerToLLVM.cpp
@@ -1336,6 +1336,41 @@ mlir::LogicalResult CIRToLLVMBaseClassAddrOpLowering::matchAndRewrite(
   return mlir::success();
 }
 
+mlir::LogicalResult CIRToLLVMDerivedClassAddrOpLowering::matchAndRewrite(
+    cir::DerivedClassAddrOp derivedClassOp, OpAdaptor adaptor,
+    mlir::ConversionPatternRewriter &rewriter) const {
+  const mlir::Type resultType =
+      getTypeConverter()->convertType(derivedClassOp.getType());
+  mlir::Value baseAddr = adaptor.getBaseAddr();
+  // The offset is set in the operation as an unsigned value, but it must be
+  // applied as a negative offset.
+  int64_t offsetVal = -(adaptor.getOffset().getZExtValue());
+  if (offsetVal == 0) {
+    // If the offset is zero, we can just return the base address,
+    rewriter.replaceOp(derivedClassOp, baseAddr);
+    return mlir::success();
+  }
+  llvm::SmallVector<mlir::LLVM::GEPArg, 1> offset = {offsetVal};
+  mlir::Type byteType = mlir::IntegerType::get(resultType.getContext(), 8,
+                                               mlir::IntegerType::Signless);
+  if (derivedClassOp.getAssumeNotNull()) {
+    rewriter.replaceOpWithNewOp<mlir::LLVM::GEPOp>(
+        derivedClassOp, resultType, byteType, baseAddr, offset,
+        mlir::LLVM::GEPNoWrapFlags::inbounds);
+  } else {
+    mlir::Location loc = derivedClassOp.getLoc();
+    mlir::Value isNull = mlir::LLVM::ICmpOp::create(
+        rewriter, loc, mlir::LLVM::ICmpPredicate::eq, baseAddr,
+        mlir::LLVM::ZeroOp::create(rewriter, loc, baseAddr.getType()));
+    mlir::Value adjusted =
+        mlir::LLVM::GEPOp::create(rewriter, loc, resultType, byteType, baseAddr,
+                                  offset, mlir::LLVM::GEPNoWrapFlags::inbounds);
+    rewriter.replaceOpWithNewOp<mlir::LLVM::SelectOp>(derivedClassOp, isNull,
+                                                      baseAddr, adjusted);
+  }
+  return mlir::success();
+}
+
 mlir::LogicalResult CIRToLLVMATanOpLowering::matchAndRewrite(
     cir::ATanOp op, OpAdaptor adaptor,
     mlir::ConversionPatternRewriter &rewriter) const {
diff --git a/clang/test/CIR/CodeGen/base-to-derived.cpp b/clang/test/CIR/CodeGen/base-to-derived.cpp
new file mode 100644
index 0000000000000..af9aa0ffd19c1
--- /dev/null
+++ b/clang/test/CIR/CodeGen/base-to-derived.cpp
@@ -0,0 +1,97 @@
+// RUN: %clang_cc1 -triple aarch64-none-linux-android21 -fclangir -emit-cir %s -o %t.cir
+// RUN: FileCheck --check-prefix=CIR --input-file=%t.cir %s
+// RUN: %clang_cc1 -triple aarch64-none-linux-android21 -fclangir -emit-llvm %s -o %t-cir.ll
+// RUN: FileCheck --check-prefix=LLVM --input-file=%t-cir.ll %s
+// RUN: %clang_cc1 -triple aarch64-none-linux-android21 -emit-llvm %s -o %t.ll
+// RUN: FileCheck --check-prefix=OGCG --input-file=%t.ll %s
+
+class A {
+    int a;
+};
+
+class B {
+    int b;
+public:
+    A *getAsA();
+};
+
+class X : public A, public B {
+    int x;
+};
+
+X *castAtoX(A *a) {
+  return static_cast<X*>(a);
+}
+
+// CIR: cir.func {{.*}} @_Z8castAtoXP1A(%[[ARG0:.*]]: !cir.ptr<!rec_A> {{.*}})
+// CIR:   %[[A_ADDR:.*]] = cir.alloca !cir.ptr<!rec_A>, !cir.ptr<!cir.ptr<!rec_A>>, ["a", init]
+// CIR:   cir.store %[[ARG0]], %[[A_ADDR]] : !cir.ptr<!rec_A>, !cir.ptr<!cir.ptr<!rec_A>>
+// CIR:   %[[A:.*]] = cir.load{{.*}} %[[A_ADDR]] : !cir.ptr<!cir.ptr<!rec_A>>, !cir.ptr<!rec_A>
+// CIR:   %[[X:.*]] = cir.derived_class_addr %[[A]] : !cir.ptr<!rec_A> [0] -> !cir.ptr<!rec_X>
+
+// Note: Because the offset is 0, a null check is not needed.
+
+// LLVM: define {{.*}} ptr @_Z8castAtoXP1A(ptr %[[ARG0:.*]])
+// LLVM:   %[[A_ADDR:.*]] = alloca ptr
+// LLVM:   store ptr %[[ARG0]], ptr %[[A_ADDR]]
+// LLVM:   %[[X:.*]] = load ptr, ptr %[[A_ADDR]]
+
+// OGCG: define {{.*}} ptr @_Z8castAtoXP1A(ptr {{.*}} %[[ARG0:.*]])
+// OGCG:   %[[A_ADDR:.*]] = alloca ptr
+// OGCG:   store ptr %[[ARG0]], ptr %[[A_ADDR]]
+// OGCG:   %[[X:.*]] = load ptr, ptr %[[A_ADDR]]
+
+X *castBtoX(B *b) {
+  return static_cast<X*>(b);
+}
+
+// CIR: cir.func {{.*}} @_Z8castBtoXP1B(%[[ARG0:.*]]: !cir.ptr<!rec_B> {{.*}})
+// CIR:   %[[B_ADDR:.*]] = cir.alloca !cir.ptr<!rec_B>, !cir.ptr<!cir.ptr<!rec_B>>, ["b", init]
+// CIR:   cir.store %[[ARG0]], %[[B_ADDR]] : !cir.ptr<!rec_B>, !cir.ptr<!cir.ptr<!rec_B>>
+// CIR:   %[[B:.*]] = cir.load{{.*}} %[[B_ADDR]] : !cir.ptr<!cir.ptr<!rec_B>>, !cir.ptr<!rec_B>
+// CIR:   %[[X:.*]] = cir.derived_class_addr %[[B]] : !cir.ptr<!rec_B> [4] -> !cir.ptr<!rec_X>
+
+// LLVM: define {{.*}} ptr @_Z8castBtoXP1B(ptr %[[ARG0:.*]])
+// LLVM:   %[[B_ADDR:.*]] = alloca ptr, i64 1, align 8
+// LLVM:   store ptr %[[ARG0]], ptr %[[B_ADDR]], align 8
+// LLVM:   %[[B:.*]] = load ptr, ptr %[[B_ADDR]], align 8
+// LLVM:   %[[IS_NULL:.*]] = icmp eq ptr %[[B]], null
+// LLVM:   %[[B_NON_NULL:.*]] = getelementptr inbounds i8, ptr %[[B]], i32 -4
+// LLVM:   %[[X:.*]] = select i1 %[[IS_NULL]], ptr %[[B]], ptr %[[B_NON_NULL]]
+
+// OGCG: define {{.*}} ptr @_Z8castBtoXP1B(ptr {{.*}} %[[ARG0:.*]])
+// OGCG: entry:
+// OGCG:   %[[B_ADDR:.*]] = alloca ptr
+// OGCG:   store ptr %[[ARG0]], ptr %[[B_ADDR]]
+// OGCG:   %[[B:.*]] = load ptr, ptr %[[B_ADDR]]
+// OGCG:   %[[IS_NULL:.*]] = icmp eq ptr %[[B]], null
+// OGCG:   br i1 %[[IS_NULL]], label %[[LABEL_NULL:.*]], label %[[LABEL_NOTNULL:.*]]
+// OGCG: [[LABEL_NOTNULL]]:
+// OGCG:   %[[B_NON_NULL:.*]] = getelementptr inbounds i8, ptr %[[B]], i64 -4
+// OGCG:   br label %[[LABEL_END:.*]]
+// OGCG: [[LABEL_NULL]]:
+// OGCG:   br label %[[LABEL_END:.*]]
+// OGCG: [[LABEL_END]]:
+// OGCG:   %[[X:.*]] = phi ptr [ %[[B_NON_NULL]], %[[LABEL_NOTNULL]] ], [ null, %[[LABEL_NULL]] ]
+
+X &castBReftoXRef(B &b) {
+  return static_cast<X&>(b);
+}
+
+// CIR: cir.func {{.*}} @_Z14castBReftoXRefR1B(%[[ARG0:.*]]: !cir.ptr<!rec_B> {{.*}})
+// CIR:   %[[B_ADDR:.*]] = cir.alloca !cir.ptr<!rec_B>, !cir.ptr<!cir.ptr<!rec_B>>, ["b", init, const]
+// CIR:   cir.store %[[ARG0]], %[[B_ADDR]] : !cir.ptr<!rec_B>, !cir.ptr<!cir.ptr<!rec_B>>
+// CIR:   %[[B:.*]] = cir.load{{.*}} %[[B_ADDR]] : !cir.ptr<!cir.ptr<!rec_B>>, !cir.ptr<!rec_B>
+// CIR:   %[[X:.*]] = cir.derived_class_addr %[[B]] : !cir.ptr<!rec_B> nonnull [4] -> !cir.ptr<!rec_X>
+
+// LLVM: define {{.*}} ptr @_Z14castBReftoXRefR1B(ptr %[[ARG0:.*]])
+// LLVM:   %[[B_ADDR:.*]] = alloca ptr
+// LLVM:   store ptr %[[ARG0]], ptr %[[B_ADDR]]
+// LLVM:   %[[B:.*]] = load ptr, ptr %[[B_ADDR]]
+// LLVM:   %[[X:.*]] = getelementptr inbounds i8, ptr %[[B]], i32 -4
+
+// OGCG: define {{.*}} ptr @_Z14castBReftoXRefR1B(ptr {{.*}} %[[ARG0:.*]])
+// OGCG:   %[[B_ADDR:.*]] = alloca ptr
+// OGCG:   store ptr %[[ARG0]], ptr %[[B_ADDR]]
+// OGCG:   %[[B:.*]] = load ptr, ptr %[[B_ADDR]]
+// OGCG:   %[[X:.*]] = getelementptr inbounds i8, ptr %[[B]], i64 -4

[bcardosolopes]

LGTM after minor nit

[bcardosolopes]

Might be worth moving this to the description and perhaps mention the behavior is undefined in such cases?