From 95000fdb9e98a88b923fa7aa4bdeffbc618a323c Mon Sep 17 00:00:00 2001 From: Anshil Gandhi <95053726+gandhi56@users.noreply.github.com> Date: Wed, 19 Feb 2025 21:47:22 -0700 Subject: [PATCH] [CUDA] Increment VTable index for device thunks (#124989) Currently, the clang frontend incorrectly emits the callee instead of the thunk for the callee in the VTable. This is the case because the thunk index is not incremented when their callees cannot be emitted. This patch fixes the bug. --- clang/lib/CodeGen/CGVTables.cpp | 13 +++++++++---- .../test/CodeGenCUDA/increment-index-for-thunks.cu | 4 ++-- 2 files changed, 11 insertions(+), 6 deletions(-) diff --git a/clang/lib/CodeGen/CGVTables.cpp b/clang/lib/CodeGen/CGVTables.cpp index 7faf6821a6cd..c9108938bca5 100644 --- a/clang/lib/CodeGen/CGVTables.cpp +++ b/clang/lib/CodeGen/CGVTables.cpp @@ -771,6 +771,10 @@ 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 @@ -782,9 +786,12 @@ void CodeGenVTables::addVTableComponent(ConstantArrayBuilder &builder, CGM.getLangOpts().CUDAIsDevice ? MD->hasAttr() : (MD->hasAttr() || !MD->hasAttr()); - if (!CanEmitMethod) + if (!CanEmitMethod) { + if (IsThunk) + nextVTableThunkIndex++; return builder.add( llvm::ConstantExpr::getNullValue(CGM.GlobalsInt8PtrTy)); + } // Method is acceptable, continue processing as usual. } @@ -830,9 +837,7 @@ 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++; diff --git a/clang/test/CodeGenCUDA/increment-index-for-thunks.cu b/clang/test/CodeGenCUDA/increment-index-for-thunks.cu index af764198f5ea..48dbf6ef82b7 100644 --- a/clang/test/CodeGenCUDA/increment-index-for-thunks.cu +++ b/clang/test/CodeGenCUDA/increment-index-for-thunks.cu @@ -3,10 +3,10 @@ // 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 @_ZN1C2f1Ev to ptr addrspace(1))] }, comdat, align 8 +// 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) @_ZN1C2f1Ev 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