[ExecuTorch] Separate extension.Module Namespaces from Aten and non-Aten

Separate `extension.Module` Namespaces to be `executorch::extension::module` and `executorch::extension::module_aten`, otherwise in the future there will be issues like P1799454769.

It's similar to what we already did for bundled_program in #10307

Differential Revision: [D73903870](https://our.internmc.facebook.com/intern/diff/D73903870/)

ghstack-source-id: 281114516
Pull Request resolved: #10576

Author: zhenyan-zhang-meta

docs/source/extension-module.md

@@ -13,6 +13,7 @@ Let's see how we can run the `SimpleConv` model generated from the [Exporting to
 #include <executorch/extension/tensor/tensor.h>
 
 using namespace ::executorch::extension;
+using ::executorch::extension::ET_MODULE_NAMESPACE::Module;
 
 // Create a Module.
 Module module("/path/to/model.pte");
@@ -221,6 +222,7 @@ Use [ExecuTorch Dump](etdump.md) to trace model execution. Create an `ETDumpGen`
 #include <executorch/devtools/etdump/etdump_flatcc.h>
 
 using namespace ::executorch::extension;
+using ::executorch::extension::ET_MODULE_NAMESPACE::Module;
 
 Module module("/path/to/model.pte", Module::LoadMode::MmapUseMlock, std::make_unique<ETDumpGen>());
 

docs/source/extension-tensor.md

@@ -11,6 +11,7 @@ Imagine you’re working with a [`Module`](extension-module.md) interface, and y
 
 using namespace executorch::aten;
 using namespace executorch::extension;
+using ::executorch::extension::ET_MODULE_NAMESPACE::Module;
 
 SizesType sizes[] = {2, 3};
 DimOrderType dim_order[] = {0, 1};
@@ -42,6 +43,7 @@ Here’s how you can use it:
 #include <executorch/extension/tensor/tensor.h>
 
 using namespace executorch::extension;
+using ::executorch::extension::ET_MODULE_NAMESPACE::Module;
 
 auto tensor = make_tensor_ptr(
     {2, 3},                                // sizes

docs/source/getting-started.md

@@ -194,6 +194,7 @@ Loading and running a model using the high-level API can be done as follows:
 #include <executorch/extension/tensor/tensor.h>
 
 using namespace ::executorch::extension;
+using ::executorch::extension::ET_MODULE_NAMESPACE::Module;
 
 // Load the model.
 Module module("/path/to/model.pte");

docs/source/llm/getting-started.md

@@ -209,7 +209,7 @@ Create a file called main.cpp with the following contents:
 using executorch::aten::ScalarType;
 using executorch::aten::Tensor;
 using executorch::extension::from_blob;
-using executorch::extension::Module;
+using executorch::extension::ET_MODULE_NAMESPACE::Module;
 using executorch::runtime::EValue;
 using executorch::runtime::Result;
 ```

docs/source/using-executorch-cpp.md

@@ -13,6 +13,7 @@ In addition the Module class, the tensor extension provides an encapsulated inte
 #include <executorch/extension/tensor/tensor.h>
 
 using namespace ::executorch::extension;
+using ::executorch::extension::ET_MODULE_NAMESPACE::Module;
 
 // Load the model.
 Module module("/path/to/model.pte");

examples/llm_manual/main.cpp

@@ -20,7 +20,7 @@
 using executorch::aten::ScalarType;
 using executorch::aten::Tensor;
 using executorch::extension::from_blob;
-using executorch::extension::Module;
+using executorch::extension::ET_MODULE_NAMESPACE::Module;
 using executorch::runtime::EValue;
 using executorch::runtime::Result;
 

examples/models/llama/runner/runner.cpp

@@ -21,7 +21,7 @@
 
 namespace example {
 
-using ::executorch::extension::Module;
+using ::executorch::extension::ET_MODULE_NAMESPACE::Module;
 using ::executorch::runtime::Error;
 using ::executorch::runtime::Result;
 

examples/models/llama/runner/runner.h

@@ -60,7 +60,7 @@ class ET_EXPERIMENTAL Runner : public executorch::extension::llm::IRunner {
   bool shouldStop_{false};
 
   // model
-  std::unique_ptr<::executorch::extension::Module> module_;
+  std::unique_ptr<::executorch::extension::ET_MODULE_NAMESPACE::Module> module_;
   std::string tokenizer_path_;
   std::unique_ptr<::tokenizers::Tokenizer> tokenizer_;
   std::unordered_map<std::string, int64_t> metadata_;

examples/models/llava/runner/llava_image_prefiller.h

@@ -18,7 +18,8 @@ namespace example {
 class ET_EXPERIMENTAL LlavaImagePrefiller
     : public ::executorch::extension::llm::ImagePrefiller {
  public:
-  explicit LlavaImagePrefiller(::executorch::extension::Module* module)
+  explicit LlavaImagePrefiller(
+      ::executorch::extension::ET_MODULE_NAMESPACE::Module* module)
       : ImagePrefiller(module){};
   /**
    * Prefill an LLM Module with the given image input.

examples/models/llava/runner/llava_text_decoder_runner.h

@@ -17,7 +17,8 @@ namespace example {
 class ET_EXPERIMENTAL LlavaTextDecoderRunner
     : public executorch::extension::llm::TextDecoderRunner {
  public:
-  explicit LlavaTextDecoderRunner(executorch::extension::Module* module)
+  explicit LlavaTextDecoderRunner(
+      executorch::extension::ET_MODULE_NAMESPACE::Module* module)
       : TextDecoderRunner(module, true) {}
 
   inline executorch::runtime::Result<executorch::aten::Tensor> step(

examples/models/phi-3-mini/runner.cpp

@@ -16,7 +16,7 @@
 #include <pytorch/tokenizers/llama2c_tokenizer.h>
 
 using executorch::aten::ScalarType;
-using executorch::extension::Module;
+using executorch::extension::ET_MODULE_NAMESPACE::Module;
 using executorch::extension::llm::Sampler;
 using executorch::runtime::Error;
 using tokenizers::Llama2cTokenizer;

examples/models/phi-3-mini/runner.h

@@ -42,7 +42,7 @@ class Runner {
   uint64_t prefill(std::vector<uint64_t>& tokens);
   uint64_t run_model_step(uint64_t token);
 
-  std::unique_ptr<executorch::extension::Module> module_;
+  std::unique_ptr<executorch::extension::ET_MODULE_NAMESPACE::Module> module_;
   std::unique_ptr<tokenizers::Tokenizer> tokenizer_;
   std::unique_ptr<executorch::extension::llm::Sampler> sampler_;
 };

examples/qualcomm/oss_scripts/llama/runner/io_manager.cpp

@@ -12,7 +12,7 @@
 
 using executorch::aten::Tensor;
 using executorch::aten::TensorImpl;
-using executorch::extension::Module;
+using executorch::extension::ET_MODULE_NAMESPACE::Module;
 using executorch::runtime::Error;
 using executorch::runtime::MemoryAllocator;
 using executorch::runtime::MethodMeta;

examples/qualcomm/oss_scripts/llama/runner/io_manager.h

@@ -20,6 +20,8 @@
 #include <executorch/extension/module/module.h>
 #include <executorch/runtime/executor/method_meta.h>
 
+using ::executorch::extension::ET_MODULE_NAMESPACE::Module;
+
 namespace example {
 
 enum EvalMode {
@@ -29,8 +31,8 @@ enum EvalMode {
 };
 class IoMgrBase {
  public:
-  IoMgrBase(
-      std::vector<std::shared_ptr<executorch::extension::Module>>& modules);
+  IoMgrBase(std::vector<std::shared_ptr<
+                executorch::extension::ET_MODULE_NAMESPACE::Module>>& modules);
   virtual ~IoMgrBase();
   virtual void init_io() = 0;
   virtual void reset_io(
@@ -81,13 +83,16 @@ class IoMgrBase {
       std::string,
       std::vector<std::vector<executorch::aten::TensorImpl*>>>
       output_tensors_;
-  std::vector<std::shared_ptr<executorch::extension::Module>> modules_;
+  std::vector<
+      std::shared_ptr<executorch::extension::ET_MODULE_NAMESPACE::Module>>
+      modules_;
 };
 
 class ShiftPointerIoMgr : public IoMgrBase {
  public:
   ShiftPointerIoMgr(
-      std::vector<std::shared_ptr<executorch::extension::Module>>& modules,
+      std::vector<std::shared_ptr<
+          executorch::extension::ET_MODULE_NAMESPACE::Module>>& modules,
       int32_t context_len,
       int32_t prefill_ar_len,
       int32_t prefill_cache_len,
@@ -199,7 +204,8 @@ class ShiftPointerIoMgr : public IoMgrBase {
 class SmartMaskIoMgr : public IoMgrBase {
  public:
   SmartMaskIoMgr(
-      std::vector<std::shared_ptr<executorch::extension::Module>>& modules,
+      std::vector<std::shared_ptr<
+          executorch::extension::ET_MODULE_NAMESPACE::Module>>& modules,
       int32_t context_len,
       int32_t prefill_ar_len,
       int32_t prefill_cache_len,

examples/qualcomm/oss_scripts/llama/runner/runner.cpp

@@ -23,7 +23,7 @@
 #include <sstream>
 
 using executorch::aten::Tensor;
-using executorch::extension::Module;
+using executorch::extension::ET_MODULE_NAMESPACE::Module;
 using executorch::extension::llm::Sampler;
 using executorch::extension::llm::time_in_ms;
 using executorch::runtime::Error;

examples/qualcomm/oss_scripts/llama/runner/runner.h

@@ -101,7 +101,9 @@ class Runner {
   std::unordered_set<uint64_t> eos_id_;
   const int32_t n_bos_;
   const int32_t n_eos_;
-  std::vector<std::shared_ptr<executorch::extension::Module>> modules_;
+  std::vector<
+      std::shared_ptr<executorch::extension::ET_MODULE_NAMESPACE::Module>>
+      modules_;
   std::string tokenizer_path_;
   std::string performance_output_path_;
   float logits_scale_;

examples/qualcomm/qaihub_scripts/llama/runner/io_memory.cpp

@@ -14,7 +14,7 @@
 
 using executorch::aten::Tensor;
 using executorch::aten::TensorImpl;
-using executorch::extension::Module;
+using executorch::extension::ET_MODULE_NAMESPACE::Module;
 using executorch::runtime::Error;
 using executorch::runtime::MethodMeta;
 using executorch::runtime::Result;

examples/qualcomm/qaihub_scripts/llama/runner/io_memory.h

@@ -32,7 +32,9 @@ class Memory {
  public:
   Memory(
       const std::vector<std::string>& pos_embs_path,
-      std::vector<std::shared_ptr<executorch::extension::Module>>& modules);
+      std::vector<
+          std::shared_ptr<executorch::extension::ET_MODULE_NAMESPACE::Module>>&
+          modules);
   virtual ~Memory();
   virtual void prepare_io(
       const std::vector<
@@ -51,15 +53,19 @@ class Memory {
   std::vector<std::vector<executorch::aten::TensorImpl*>> input_tensors_;
   std::vector<std::vector<executorch::aten::TensorImpl*>> output_tensors_;
   std::vector<std::string> pos_embs_path_;
-  std::vector<std::shared_ptr<executorch::extension::Module>> modules_;
+  std::vector<
+      std::shared_ptr<executorch::extension::ET_MODULE_NAMESPACE::Module>>
+      modules_;
   std::vector<std::string> method_names_;
 };
 
 class BertMemory : public Memory {
  public:
   BertMemory(
       const std::vector<std::string>& pos_embs_path,
-      std::vector<std::shared_ptr<executorch::extension::Module>>& modules,
+      std::vector<
+          std::shared_ptr<executorch::extension::ET_MODULE_NAMESPACE::Module>>&
+          modules,
       std::vector<int> shard_layers);
   void prepare_io(const std::vector<executorch::runtime::Result<
                       executorch::runtime::MethodMeta>>& methods_meta) override;
@@ -121,7 +127,9 @@ class KVCachedMemory : public Memory {
  public:
   KVCachedMemory(
       const std::vector<std::string>& pos_embs_path,
-      std::vector<std::shared_ptr<executorch::extension::Module>>& modules,
+      std::vector<
+          std::shared_ptr<executorch::extension::ET_MODULE_NAMESPACE::Module>>&
+          modules,
       std::vector<int> shard_layers);
   void prepare_io(const std::vector<executorch::runtime::Result<
                       executorch::runtime::MethodMeta>>& methods_meta) override;

examples/qualcomm/qaihub_scripts/llama/runner/runner.cpp

@@ -30,7 +30,7 @@
 #endif
 
 using executorch::aten::Tensor;
-using executorch::extension::Module;
+using executorch::extension::ET_MODULE_NAMESPACE::Module;
 using executorch::extension::llm::Sampler;
 using executorch::extension::llm::time_in_ms;
 using executorch::runtime::Error;

examples/qualcomm/qaihub_scripts/llama/runner/runner.h

@@ -97,7 +97,9 @@ class Runner {
   const int32_t vocab_size_;
   const int32_t max_seq_len_;
   int32_t eval_mode_;
-  std::vector<std::shared_ptr<executorch::extension::Module>> modules_;
+  std::vector<
+      std::shared_ptr<executorch::extension::ET_MODULE_NAMESPACE::Module>>
+      modules_;
   std::vector<std::string> method_names_;
   std::string tokenizer_path_;
   float temperature_;

examples/qualcomm/qaihub_scripts/stable_diffusion/runner/runner.cpp

@@ -23,8 +23,8 @@
 #include <executorch/runtime/platform/log.h>
 
 using executorch::extension::from_blob;
-using executorch::extension::Module;
 using executorch::extension::TensorPtr;
+using executorch::extension::ET_MODULE_NAMESPACE::Module;
 using executorch::extension::llm::time_in_ms;
 using executorch::runtime::Error;
 using executorch::runtime::MethodMeta;

examples/qualcomm/qaihub_scripts/stable_diffusion/runner/runner.h

@@ -111,7 +111,9 @@ class Runner {
 
  private:
   Stats stats_;
-  std::vector<std::unique_ptr<executorch::extension::Module>> modules_;
+  std::vector<
+      std::unique_ptr<executorch::extension::ET_MODULE_NAMESPACE::Module>>
+      modules_;
   std::vector<std::string> method_names_;
   std::vector<std::vector<uint16_t>> time_emb_list_;
   std::unordered_map<std::string, int32_t> vocab_to_token_map_;

extension/android/jni/jni_layer.cpp

@@ -36,6 +36,7 @@
 
 using namespace executorch::extension;
 using namespace torch::executor;
+using ::executorch::extension::ET_MODULE_NAMESPACE::Module;
 
 namespace executorch::extension {
 class TensorHybrid : public facebook::jni::HybridClass<TensorHybrid> {

extension/apple/ExecuTorch/Exported/ExecuTorchModule.mm

@@ -15,6 +15,7 @@
 
 using namespace executorch::extension;
 using namespace executorch::runtime;
+using namespace executorch::extension::ET_MODULE_NAMESPACE::Module;
 
 static inline EValue toEValue(ExecuTorchValue *value) {
   if (value.isTensor) {

extension/llm/runner/image_prefiller.h

@@ -14,14 +14,17 @@
 #include <executorch/extension/module/module.h>
 #include <executorch/runtime/platform/compiler.h>
 
+using ::executorch::extension::ET_MODULE_NAMESPACE::Module;
+
 namespace executorch {
 namespace extension {
 namespace llm {
 
 // Assuming kv cache and parallel prefill are enabled.
 class ET_EXPERIMENTAL ImagePrefiller {
  public:
-  explicit ImagePrefiller(::executorch::extension::Module* module)
+  explicit ImagePrefiller(
+      ::executorch::extension::ET_MODULE_NAMESPACE::Module* module)
       : module_(module) {}
 
   /**

extension/llm/runner/multimodal_runner.h

@@ -29,6 +29,8 @@
 #include <executorch/extension/module/module.h>
 #include <pytorch/tokenizers/tokenizer.h>
 
+using ::executorch::extension::ET_MODULE_NAMESPACE::Module;
+
 namespace executorch {
 namespace extension {
 namespace llm {

extension/llm/runner/text_decoder_runner.h

@@ -15,6 +15,8 @@
 #include <executorch/extension/tensor/tensor.h>
 #include <executorch/runtime/platform/compiler.h>
 
+using ::executorch::extension::ET_MODULE_NAMESPACE::Module;
+
 namespace executorch {
 namespace extension {
 namespace llm {

extension/module/module.cpp

@@ -36,6 +36,7 @@
 
 namespace executorch {
 namespace extension {
+namespace ET_MODULE_NAMESPACE {
 
 using ET_RUNTIME_NAMESPACE::MethodMeta;
 using ET_RUNTIME_NAMESPACE::Program;
@@ -312,5 +313,6 @@ ET_NODISCARD inline runtime::Result<Method*> Module::get_method(
   return methods_[method_name].method.get();
 }
 
+} // namespace ET_MODULE_NAMESPACE
 } // namespace extension
 } // namespace executorch

extension/module/module.h

@@ -16,8 +16,15 @@
 
 #include <executorch/runtime/executor/program.h>
 
+#ifdef USE_ATEN_LIB
+#define ET_MODULE_NAMESPACE module::aten
+#else // !USE_ATEN_LIB
+#define ET_MODULE_NAMESPACE module
+#endif // USE_ATEN_LIB
+
 namespace executorch {
 namespace extension {
+namespace ET_MODULE_NAMESPACE {
 
 using ET_RUNTIME_NAMESPACE::Method;
 using ET_RUNTIME_NAMESPACE::MethodMeta;
@@ -506,13 +513,14 @@ class Module {
   friend class ExecuTorchJni;
 };
 
+} // namespace ET_MODULE_NAMESPACE
 } // namespace extension
 } // namespace executorch
 
 namespace torch {
 namespace executor {
 // TODO(T197294990): Remove these deprecated aliases once all users have moved
 // to the new `::executorch` namespaces.
-using ::executorch::extension::Module;
+using ::executorch::extension::ET_MODULE_NAMESPACE::Module;
 } // namespace executor
 } // namespace torch