[clang] [HIP][Clang][CodeGen] Handle hip bin symbols properly. (PR #107458)

[via cfe-commits]

https://github.com/jofrn updated https://github.com/llvm/llvm-project/pull/107458

>From d36c7e03866a7722c71c60d7e5f3e3c3ece24a60 Mon Sep 17 00:00:00 2001
From: jofernau <Joe.Fernau at amd.com>
Date: Tue, 10 Sep 2024 14:24:27 -0400
Subject: [PATCH] [HIP][Clang][CodeGen] Handle hip bin symbols properly.

Remove '_' in fatbin symbol suffix when missing TU hash ID. Internalize gpubin symbol so that it is not unresolved at link-time.
---
 clang/lib/CodeGen/CGCUDANV.cpp        | 19 +++++++++++--------
 clang/test/CodeGenCUDA/device-stub.cu |  2 +-
 2 files changed, 12 insertions(+), 9 deletions(-)

diff --git a/clang/lib/CodeGen/CGCUDANV.cpp b/clang/lib/CodeGen/CGCUDANV.cpp
index 59c5927717933d..1e8128442131b3 100644
--- a/clang/lib/CodeGen/CGCUDANV.cpp
+++ b/clang/lib/CodeGen/CGCUDANV.cpp
@@ -840,8 +840,10 @@ llvm::Function *CGNVCUDARuntime::makeModuleCtorFunction() {
       FatBinStr = new llvm::GlobalVariable(
           CGM.getModule(), CGM.Int8Ty,
           /*isConstant=*/true, llvm::GlobalValue::ExternalLinkage, nullptr,
-          "__hip_fatbin_" + CGM.getContext().getCUIDHash(), nullptr,
-          llvm::GlobalVariable::NotThreadLocal);
+          "__hip_fatbin" +
+	    (CGM.getLangOpts().CUID.empty() ? ""
+	      : "_" + CGM.getContext().getCUIDHash()),
+	  nullptr, llvm::GlobalVariable::NotThreadLocal);
       cast<llvm::GlobalVariable>(FatBinStr)->setSection(FatbinConstantName);
     }
 
@@ -894,8 +896,8 @@ llvm::Function *CGNVCUDARuntime::makeModuleCtorFunction() {
   // thread safety of the loaded program. Therefore we can assume sequential
   // execution of constructor functions here.
   if (IsHIP) {
-    auto Linkage = CudaGpuBinary ? llvm::GlobalValue::InternalLinkage
-                                 : llvm::GlobalValue::ExternalLinkage;
+    auto Linkage = RelocatableDeviceCode ?
+      llvm::GlobalValue::ExternalLinkage : llvm::GlobalValue::InternalLinkage;
     llvm::BasicBlock *IfBlock =
         llvm::BasicBlock::Create(Context, "if", ModuleCtorFunc);
     llvm::BasicBlock *ExitBlock =
@@ -905,10 +907,11 @@ llvm::Function *CGNVCUDARuntime::makeModuleCtorFunction() {
     GpuBinaryHandle = new llvm::GlobalVariable(
         TheModule, PtrTy, /*isConstant=*/false, Linkage,
         /*Initializer=*/
-        CudaGpuBinary ? llvm::ConstantPointerNull::get(PtrTy) : nullptr,
-        CudaGpuBinary
-            ? "__hip_gpubin_handle"
-            : "__hip_gpubin_handle_" + CGM.getContext().getCUIDHash());
+        !RelocatableDeviceCode ? llvm::ConstantPointerNull::get(PtrTy) : nullptr,
+        "__hip_gpubin_handle" +
+          (CGM.getLangOpts().CUID.empty()
+            ? ""
+            : "_" + CGM.getContext().getCUIDHash()));
     GpuBinaryHandle->setAlignment(CGM.getPointerAlign().getAsAlign());
     // Prevent the weak symbol in different shared libraries being merged.
     if (Linkage != llvm::GlobalValue::InternalLinkage)
diff --git a/clang/test/CodeGenCUDA/device-stub.cu b/clang/test/CodeGenCUDA/device-stub.cu
index 60304647bd4c54..c7f6e0f1168618 100644
--- a/clang/test/CodeGenCUDA/device-stub.cu
+++ b/clang/test/CodeGenCUDA/device-stub.cu
@@ -175,7 +175,7 @@ __device__ void device_use() {
 // HIP-SAME: section ".hipFatBinSegment"
 // * variable to save GPU binary handle after initialization
 // CUDANORDC: @__[[PREFIX]]_gpubin_handle = internal global ptr null
-// HIPNEF: @__[[PREFIX]]_gpubin_handle_{{[0-9a-f]+}} = external hidden global ptr, align 8
+// HIPNEF: @__[[PREFIX]]_gpubin_handle_{{[0-9a-f]+}} = internal global null, align 8
 // * constant unnamed string with NVModuleID
 // CUDARDC: [[MODULE_ID_GLOBAL:@.*]] = private constant
 // CUDARDC-SAME: c"[[MODULE_ID:.+]]\00", section "__nv_module_id", align 32