[PATCH] D109707: [HIP] [AlwaysInliner] Disable AlwaysInliner to eliminate undefined symbols

[Anshil Gandhi via Phabricator via llvm-commits]

gandhi21299 created this revision.
gandhi21299 added reviewers: yaxunl, aeubanks.
Herald added subscribers: foad, kerbowa, hiraditya, tpr, nhaehnle, jvesely, arsenm.
gandhi21299 requested review of this revision.
Herald added projects: clang, LLVM.
Herald added subscribers: llvm-commits, cfe-commits.

By default clang emits complete contructors as alias of base constructors if they are the same. The backend is supposed to emit symbols for the alias, otherwise it causes undefined symbols. @yaxunl observed that this issue is related to the llvm options `-amdgpu-early-inline-all=true` and `-amdgpu-function-calls=false`. Disabling the AlwaysInliner fixes this issue.


Repository:
  rG LLVM Github Monorepo

https://reviews.llvm.org/D109707

Files:
  clang/lib/Driver/ToolChains/Clang.cpp
  clang/test/CodeGen/amdgpu-alias-undef-symbols.hip
  llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp


Index: llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
===================================================================
--- llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
+++ llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
@@ -689,8 +689,6 @@
         if (InternalizeSymbols) {
           PM.addPass(GlobalDCEPass());
         }
-        if (EarlyInlineAll && !EnableFunctionCalls)
-          PM.addPass(AMDGPUAlwaysInlinePass());
       });
 
   PB.registerCGSCCOptimizerLateEPCallback(
Index: clang/test/CodeGen/amdgpu-alias-undef-symbols.hip
===================================================================
--- /dev/null
+++ clang/test/CodeGen/amdgpu-alias-undef-symbols.hip
@@ -0,0 +1,14 @@
+// RUN: %clang -c --offload-arch=gfx906 --cuda-device-only -emit-llvm -S -o - %s
+//  -fgpu-rdc -O3 -mllvm -amdgpu-early-inline-all=true -mllvm -amdgpu-function-calls=false
+//  FileCheck %s
+
+// CHECK: %struct.B = type { i8 }
+struct B {
+
+  // CHECK: @_ZN1BC1Ei = hidden unnamed_addr alias void (%struct.B*, i32), void (%struct.B*, i32)* @_ZN1BC2Ei
+  __device__ B(int x);
+};
+
+__device__ B::B(int x) {
+
+}
Index: clang/lib/Driver/ToolChains/Clang.cpp
===================================================================
--- clang/lib/Driver/ToolChains/Clang.cpp
+++ clang/lib/Driver/ToolChains/Clang.cpp
@@ -5066,7 +5066,7 @@
   // Enable -mconstructor-aliases except on darwin, where we have to work around
   // a linker bug (see <rdar://problem/7651567>), and CUDA/AMDGPU device code,
   // where aliases aren't supported.
-  if (!RawTriple.isOSDarwin() && !RawTriple.isNVPTX() && !RawTriple.isAMDGPU())
+  if (!RawTriple.isOSDarwin() && !RawTriple.isNVPTX())
     CmdArgs.push_back("-mconstructor-aliases");
 
   // Darwin's kernel doesn't support guard variables; just die if we


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D109707.372294.patch
Type: text/x-patch
Size: 1791 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20210913/1bee6b1c/attachment-0001.bin>