Fix lookup symbol from cuda module failed

taichi-dev · Nov 22, 2022 · e8eb7f0 · e8eb7f0
1 parent 5191a2c
commit e8eb7f0
Show file tree

Hide file tree

Showing 4 changed files with 25 additions and 24 deletions.
diff --git a/taichi/codegen/cuda/codegen_cuda.cpp b/taichi/codegen/cuda/codegen_cuda.cpp
@@ -618,6 +618,12 @@ FunctionType CUDAModuleToFunctionConverter::convert(
   auto &mod = data.module;
   auto &tasks = data.tasks;
 #ifdef TI_WITH_CUDA
+  for (const auto &task : tasks) {
+    llvm::Function *func = mod->getFunction(task.name);
+    TI_ASSERT(func);
+    tlctx_->mark_function_as_cuda_kernel(func, task.block_dim);
+  }
+
   auto jit = tlctx_->jit.get();
   auto cuda_module =
       jit->add_module(std::move(mod), executor_->get_config()->gpu_max_reg);

diff --git a/taichi/codegen/llvm/codegen_llvm.cpp b/taichi/codegen/llvm/codegen_llvm.cpp
@@ -2638,15 +2638,6 @@ LLVMCompiledTask TaskCodeGenLLVM::run_compilation() {
   emit_to_module();
   eliminate_unused_functions();
 
-  if (config.arch == Arch::cuda) {
-    // CUDA specific metadata
-    for (const auto &task : offloaded_tasks) {
-      llvm::Function *func = module->getFunction(task.name);
-      TI_ASSERT(func);
-      tlctx->mark_function_as_cuda_kernel(func, task.block_dim);
-    }
-  }
-
   return {std::move(offloaded_tasks), std::move(module),
           std::move(used_tree_ids), std::move(struct_for_tls_sizes)};
 }

diff --git a/taichi/runtime/llvm/llvm_context.cpp b/taichi/runtime/llvm/llvm_context.cpp
@@ -663,16 +663,17 @@ void TaichiLLVMContext::insert_nvvm_annotation(llvm::Function *func,
                float addrspace(1)*,
                float addrspace(1)*)* @kernel, !"kernel", i32 1}
   *******************************************************************/
-  auto ctx = get_this_thread_context();
-  llvm::Metadata *md_args[] = {llvm::ValueAsMetadata::get(func),
-                               MDString::get(*ctx, key),
-                               llvm::ValueAsMetadata::get(get_constant(val))};
+  auto *llvm_mod = func->getParent();
+  auto *ctx = &llvm_mod->getContext();
+  llvm::Metadata *md_args[] = {
+      llvm::ValueAsMetadata::get(func), MDString::get(*ctx, key),
+      llvm::ValueAsMetadata::get(llvm::ConstantInt::get(
+          *ctx, llvm::APInt(sizeof(val) * 8, (uint64)val,
+                            std::is_signed_v<decltype(val)>)))};
 
   MDNode *md_node = MDNode::get(*ctx, md_args);
 
-  func->getParent()
-      ->getOrInsertNamedMetadata("nvvm.annotations")
-      ->addOperand(md_node);
+  llvm_mod->getOrInsertNamedMetadata("nvvm.annotations")->addOperand(md_node);
 }
 
 void TaichiLLVMContext::mark_function_as_cuda_kernel(llvm::Function *func,

diff --git a/taichi/runtime/llvm/llvm_context.h b/taichi/runtime/llvm/llvm_context.h
@@ -123,8 +123,6 @@ class TaichiLLVMContext {
       llvm::Module *module,
       std::function<bool(const std::string &)> export_indicator);
 
-  void mark_function_as_cuda_kernel(llvm::Function *func, int block_dim = 0);
-
   void fetch_this_thread_struct_module();
   llvm::Module *get_this_thread_runtime_module();
   llvm::Function *get_runtime_function(const std::string &name);
@@ -139,29 +137,34 @@ class TaichiLLVMContext {
 
   void add_struct_for_func(llvm::Module *module, int tls_size);
 
-  static std::string get_struct_for_func_name(int tls_size);
-
   LLVMCompiledKernel link_compiled_tasks(
       std::vector<std::unique_ptr<LLVMCompiledTask>> data_list);
 
+  static std::string get_struct_for_func_name(int tls_size);
+
+  static void mark_function_as_cuda_kernel(llvm::Function *func,
+                                           int block_dim = 0);
+
  private:
   std::unique_ptr<llvm::Module> clone_module_to_context(
       llvm::Module *module,
       llvm::LLVMContext *target_context);
 
   void link_module_with_cuda_libdevice(std::unique_ptr<llvm::Module> &module);
 
-  static int num_instructions(llvm::Function *func);
-
-  void insert_nvvm_annotation(llvm::Function *func, std::string key, int val);
-
   std::unique_ptr<llvm::Module> clone_module_to_this_thread_context(
       llvm::Module *module);
 
   ThreadLocalData *get_this_thread_data();
 
   void update_runtime_jit_module(std::unique_ptr<llvm::Module> module);
 
+  static int num_instructions(llvm::Function *func);
+
+  static void insert_nvvm_annotation(llvm::Function *func,
+                                     std::string key,
+                                     int val);
+
   std::unordered_map<std::thread::id, std::unique_ptr<ThreadLocalData>>
       per_thread_data_;