diff --git a/taichi/runtime/llvm/llvm_context_pass.h b/taichi/runtime/llvm/llvm_context_pass.h
index d48303dbc6670..4eb8e4477d87f 100644
--- a/taichi/runtime/llvm/llvm_context_pass.h
+++ b/taichi/runtime/llvm/llvm_context_pass.h
@@ -9,6 +9,7 @@
 #include "llvm/Transforms/IPO/PassManagerBuilder.h"
 #include "llvm/IR/Instructions.h"
 #include "llvm/IR/IRBuilder.h"
+#include "llvm/Support/SourceMgr.h"
 
 #if defined(TI_WITH_AMDGPU)
 #include "taichi/rhi/amdgpu/amdgpu_context.h"
@@ -19,7 +20,7 @@ namespace lang {
 using namespace llvm;
 #if defined(TI_WITH_AMDGPU)
 struct AMDGPUConvertAllocaInstAddressSpacePass : public FunctionPass {
-  static char ID;
+  static inline char ID{0};
   AMDGPUConvertAllocaInstAddressSpacePass() : FunctionPass(ID) {
   }
   bool runOnFunction(llvm::Function &f) override {
@@ -38,7 +39,7 @@ struct AMDGPUConvertAllocaInstAddressSpacePass : public FunctionPass {
       }
       for (auto &allocainst : alloca_inst_vec) {
         auto alloca_type = allocainst->getAllocatedType();
-        IRBuilder<> builder(allocainst);
+        llvm::IRBuilder<> builder(allocainst);
         auto *new_alloca = builder.CreateAlloca(alloca_type, (unsigned)5);
         auto new_type = llvm::PointerType::get(alloca_type, (unsigned)0);
         new_alloca->setAlignment(Align(allocainst->getAlign().value()));
@@ -52,7 +53,7 @@ struct AMDGPUConvertAllocaInstAddressSpacePass : public FunctionPass {
 };
 
 struct AMDGPUConvertFuncParamAddressSpacePass : public ModulePass {
-  static char ID;
+  static inline char ID{0};
   AMDGPUConvertFuncParamAddressSpacePass() : ModulePass(ID) {
   }
   bool runOnModule(llvm::Module &M) override {
@@ -127,8 +128,6 @@ struct AMDGPUConvertFuncParamAddressSpacePass : public ModulePass {
   }
 };
 
-char AMDGPUConvertAllocaInstAddressSpacePass::ID = 0;
-char AMDGPUConvertFuncParamAddressSpacePass::ID = 0;
 #endif
 
 }  // namespace lang
diff --git a/tests/cpp/backends/amdgpu_device_test.cpp b/tests/cpp/backends/amdgpu_device_test.cpp
index 20fc49a0f5e7c..f5c629ae6968a 100644
--- a/tests/cpp/backends/amdgpu_device_test.cpp
+++ b/tests/cpp/backends/amdgpu_device_test.cpp
@@ -5,10 +5,22 @@
 #include "taichi/rhi/amdgpu/amdgpu_driver.h"
 #include "taichi/rhi/amdgpu/amdgpu_context.h"
 #include "taichi/rhi/amdgpu/amdgpu_device.h"
+#include "taichi/runtime/llvm/llvm_context_pass.h"
+
+#include <llvm/IR/Instructions.h>
+#include <llvm/IR/IRBuilder.h>
+#include <llvm/IR/LLVMContext.h>
+#include <llvm/IR/LegacyPassManager.h>
+#include <llvm/IR/Function.h>
+#include <llvm/IR/Module.h>
+#include <llvm/Pass.h>
+#include <llvm/IRReader/IRReader.h>
+
 #include "tests/cpp/program/test_program.h"
 
 namespace taichi {
 namespace lang {
+
 TEST(AMDGPU, CreateDeviceAndAlloc) {
   std::unique_ptr<amdgpu::AmdgpuDevice> device =
       std::make_unique<amdgpu::AmdgpuDevice>();
@@ -87,6 +99,94 @@ TEST(AMDGPU, CreateContextAndGetMemInfo) {
   EXPECT_GE(free_size, 0);
 }
 
+TEST(AMDGPU, ConvertAllocaInstAddressSpacePass) {
+  const std::string program =
+      "define dso_local void @runtime_add(double* %0, double* %1, double* %2) "
+      "#4 "
+      "{\n"
+      "%4 = alloca double*, align 8\n"
+      "%5 = alloca double*, align 8\n"
+      "%6 = alloca double*, align 8\n"
+      "store double* %0, double** %4, align 8\n"
+      "store double* %1, double** %5, align 8\n"
+      "store double* %2, double** %6, align 8\n"
+      "%7 = load double*, double** %4, align 8\n"
+      "%8 = load double, double* %7, align 8\n"
+      "%9 = load double*, double** %5, align 8\n"
+      "%10 = load double, double* %9, align 8\n"
+      "%11 = fadd contract double %8, %10\n"
+      "%12 = load double*, double** %6, align 8\n"
+      "store double %11, double* %12, align 8\n"
+      "ret void\n"
+      "}\n";
+  llvm::LLVMContext llvm_context;
+  llvm::SMDiagnostic diagnostic_err;
+  std::unique_ptr<llvm::Module> llvm_module = llvm::parseIR(
+      llvm::MemoryBuffer::getMemBuffer(program)->getMemBufferRef(),
+      diagnostic_err, llvm_context);
+  llvm::legacy::FunctionPassManager function_pass_manager(llvm_module.get());
+  function_pass_manager.add(new AMDGPUConvertAllocaInstAddressSpacePass());
+  function_pass_manager.doInitialization();
+  for (auto func = llvm_module->begin(); func != llvm_module->end(); ++func) {
+    function_pass_manager.run(*func);
+  }
+  function_pass_manager.doFinalization();
+  auto func = llvm_module->getFunction("runtime_add");
+  for (auto &bb : *func) {
+    for (llvm::Instruction &inst : bb) {
+      auto alloca_inst = llvm::dyn_cast<AllocaInst>(&inst);
+      if (!alloca_inst)
+        continue;
+      EXPECT_EQ(alloca_inst->getAddressSpace(), 5);
+    }
+    int cast_num = 0;
+    for (llvm::Instruction &inst : bb) {
+      auto cast_inst = llvm::dyn_cast<AddrSpaceCastInst>(&inst);
+      if (!cast_inst)
+        continue;
+      cast_num++;
+    }
+    EXPECT_EQ(cast_num, 3);
+  }
+}
+
+TEST(AMDGPU, ConvertFuncParamAddressSpacePass) {
+  const std::string program =
+      "define dso_local void @runtime_add(double* %0, double* %1, double* %2) "
+      "#4 "
+      "{\n"
+      "%4 = alloca double*, align 8\n"
+      "%5 = alloca double*, align 8\n"
+      "%6 = alloca double*, align 8\n"
+      "store double* %0, double** %4, align 8\n"
+      "store double* %1, double** %5, align 8\n"
+      "store double* %2, double** %6, align 8\n"
+      "%7 = load double*, double** %4, align 8\n"
+      "%8 = load double, double* %7, align 8\n"
+      "%9 = load double*, double** %5, align 8\n"
+      "%10 = load double, double* %9, align 8\n"
+      "%11 = fadd contract double %8, %10\n"
+      "%12 = load double*, double** %6, align 8\n"
+      "store double %11, double* %12, align 8\n"
+      "ret void\n"
+      "}\n";
+  llvm::LLVMContext llvm_context;
+  llvm::SMDiagnostic diagnostic_err;
+  std::unique_ptr<llvm::Module> llvm_module = llvm::parseIR(
+      llvm::MemoryBuffer::getMemBuffer(program)->getMemBufferRef(),
+      diagnostic_err, llvm_context);
+  llvm::legacy::PassManager module_pass_manager;
+  module_pass_manager.add(new AMDGPUConvertFuncParamAddressSpacePass());
+  module_pass_manager.run(*llvm_module);
+  auto func = llvm_module->getFunction("runtime_add");
+  for (llvm::Function::arg_iterator I = func->arg_begin(), E = func->arg_end();
+       I != E; ++I) {
+    if (I->getType()->getTypeID() == llvm::Type::PointerTyID) {
+      EXPECT_EQ(I->getType()->getPointerAddressSpace(), 1);
+    }
+  }
+}
+
 }  // namespace lang
 }  // namespace taichi
 #endif