[CUDA] Increment VTable index for device thunks (llvm#124989) (llvm#866)

Author: jharryma

clang/lib/CodeGen/CGVTables.cpp

@@ -772,6 +772,10 @@ void CodeGenVTables::addVTableComponent(ConstantArrayBuilder &builder,
   case VTableComponent::CK_DeletingDtorPointer: {
     GlobalDecl GD = component.getGlobalDecl();
 
+    const bool IsThunk =
+        nextVTableThunkIndex < layout.vtable_thunks().size() &&
+        layout.vtable_thunks()[nextVTableThunkIndex].first == componentIndex;
+
     if (CGM.getLangOpts().CUDA) {
       // Emit NULL for methods we can't codegen on this
       // side. Otherwise we'd end up with vtable with unresolved
@@ -783,9 +787,12 @@ void CodeGenVTables::addVTableComponent(ConstantArrayBuilder &builder,
           CGM.getLangOpts().CUDAIsDevice
               ? MD->hasAttr<CUDADeviceAttr>()
               : (MD->hasAttr<CUDAHostAttr>() || !MD->hasAttr<CUDADeviceAttr>());
-      if (!CanEmitMethod)
+      if (!CanEmitMethod) {
+        if (IsThunk)
+          nextVTableThunkIndex++;
         return builder.add(
             llvm::ConstantExpr::getNullValue(CGM.GlobalsInt8PtrTy));
+      }
       // Method is acceptable, continue processing as usual.
     }
 
@@ -832,9 +839,7 @@ void CodeGenVTables::addVTableComponent(ConstantArrayBuilder &builder,
       fnPtr = DeletedVirtualFn;
 
     // Thunks.
-    } else if (nextVTableThunkIndex < layout.vtable_thunks().size() &&
-               layout.vtable_thunks()[nextVTableThunkIndex].first ==
-                   componentIndex) {
+    } else if (IsThunk) {
       auto &thunkInfo = layout.vtable_thunks()[nextVTableThunkIndex].second;
 
       nextVTableThunkIndex++;

clang/test/CodeGenCUDA/increment-index-for-thunks.cu

@@ -0,0 +1,35 @@
+// RUN: %clang_cc1 -fcuda-is-device -triple amdgcn-amd-amdhsa -target-cpu gfx942 \
+// RUN:   -emit-llvm -xhip %s -o - | FileCheck %s --check-prefix=GCN
+// RUN: %clang_cc1 -fcuda-is-device -triple spirv64-amd-amdhsa \
+// RUN:   -emit-llvm -xhip %s -o - | FileCheck %s --check-prefix=SPIRV
+
+// GCN: @_ZTV1C = linkonce_odr unnamed_addr addrspace(1) constant { [5 x ptr addrspace(1)], [4 x ptr addrspace(1)] } { [5 x ptr addrspace(1)] [ptr addrspace(1) null, ptr addrspace(1) null, ptr addrspace(1) addrspacecast (ptr @_ZN1B2f2Ev to ptr addrspace(1)), ptr addrspace(1) null, ptr addrspace(1) addrspacecast (ptr @_ZN1C2f1Ev to ptr addrspace(1))], [4 x ptr addrspace(1)] [ptr addrspace(1) inttoptr (i64 -8 to ptr addrspace(1)), ptr addrspace(1) null, ptr addrspace(1) null, ptr addrspace(1) addrspacecast (ptr @_ZThn8_N1C2f1Ev to ptr addrspace(1))] }, comdat, align 8
+// GCN: @_ZTV1B = linkonce_odr unnamed_addr addrspace(1) constant { [3 x ptr addrspace(1)] } { [3 x ptr addrspace(1)] [ptr addrspace(1) null, ptr addrspace(1) null, ptr addrspace(1) addrspacecast (ptr @_ZN1B2f2Ev to ptr addrspace(1))] }, comdat, align 8
+// GCN: @_ZTV1A = linkonce_odr unnamed_addr addrspace(1) constant { [4 x ptr addrspace(1)] } { [4 x ptr addrspace(1)] [ptr addrspace(1) null, ptr addrspace(1) null, ptr addrspace(1) null, ptr addrspace(1) addrspacecast (ptr @__cxa_pure_virtual to ptr addrspace(1))] }, comdat, align 8
+// SPIRV: @_ZTV1C = linkonce_odr unnamed_addr addrspace(1) constant { [5 x ptr addrspace(1)], [4 x ptr addrspace(1)] } { [5 x ptr addrspace(1)] [ptr addrspace(1) null, ptr addrspace(1) null, ptr addrspace(1) addrspacecast (ptr addrspace(4) @_ZN1B2f2Ev to ptr addrspace(1)), ptr addrspace(1) null, ptr addrspace(1) addrspacecast (ptr addrspace(4) @_ZN1C2f1Ev to ptr addrspace(1))], [4 x ptr addrspace(1)] [ptr addrspace(1) inttoptr (i64 -8 to ptr addrspace(1)), ptr addrspace(1) null, ptr addrspace(1) null, ptr addrspace(1) addrspacecast (ptr addrspace(4) @_ZThn8_N1C2f1Ev to ptr addrspace(1))] }, comdat, align 8
+// SPIRV: @_ZTV1B = linkonce_odr unnamed_addr addrspace(1) constant { [3 x ptr addrspace(1)] } { [3 x ptr addrspace(1)] [ptr addrspace(1) null, ptr addrspace(1) null, ptr addrspace(1) addrspacecast (ptr addrspace(4) @_ZN1B2f2Ev to ptr addrspace(1))] }, comdat, align 8
+// SPIRV: @_ZTV1A = linkonce_odr unnamed_addr addrspace(1) constant { [4 x ptr addrspace(1)] } { [4 x ptr addrspace(1)] [ptr addrspace(1) null, ptr addrspace(1) null, ptr addrspace(1) null, ptr addrspace(1) addrspacecast (ptr addrspace(4) @__cxa_pure_virtual to ptr addrspace(1))] }, comdat, align 8
+
+struct A {
+  __attribute__((device)) A() { }
+  virtual void neither_device_nor_host_f() = 0 ;
+  __attribute__((device)) virtual void f1() = 0;
+ 
+};
+ 
+struct B {
+  __attribute__((device)) B() { }
+  __attribute__((device)) virtual void f2() { };
+};
+ 
+struct C : public B, public A {
+  __attribute__((device)) C() : B(), A() { }
+ 
+   virtual void neither_device_nor_host_f() override { }
+  __attribute__((device)) virtual void f1() override { }
+ 
+};
+ 
+__attribute__((device)) void test() {
+  C obj;
+}