Add a namespace for ATen mode (#9894)

Summary:
Pull Request resolved: #9894

## Context

As titled. This is an effort trying to solve a big pain point for ATen mode users: duplicate symbols and duplicate operators.

A typical duplicate symbol issue looks like:

```
ld.lld: error: duplicate symbol: executorch::runtime::Method::get_num_external_constants()
>>> defined at __stripped__/method.cpp.pic.stripped.o:(executorch::runtime::Method::get_num_external_constants()) in archive buck-out/v2/gen/fbcode/712c6d0a4cb497c7/executorch/runtime/executor/__program_no_prim_ops_aten__/libprogram_no_prim_ops_aten.stripped.pic.a
>>> defined at __stripped__/method.cpp.pic.stripped.o:(.text._ZN10executorch7runtime6Method26get_num_external_constantsEv+0x0) in archive buck-out/v2/gen/fbcode/712c6d0a4cb497c7/executorch/runtime/executor/__program_no_prim_ops__/libprogram_no_prim_ops.stripped.pic.a
```
[User post](https://fb.workplace.com/groups/pytorch.edge.users/permalink/1727735561430063/)

This is caused by user depending on both `program_no_prim_ops` and `program_no_prim_ops_aten`.

The issue happens because both libraries define symbols like: `executorch::runtime::Method` and they transitively depend on different definitions of `Tensor` and other types, see `exec_aten.h`.

The other common issue is re-registering operators:

```
buck2 test  //arvr/libraries/wristband/tsn/transformers:TorchstreamTransformer  -- --print-passing-details
File changed: fbsource//xplat/executorch/build/fb/clients.bzl
File changed: fbsource//xplat/executorch
File changed: fbcode//executorch/build/fb/clients.bzl
16 additional file change events
⚠ Listing failed: fbsource//arvr/libraries/wristband/tsn/transformers:TorchstreamTransformerTestFbcode
Failed to list tests. Expected exit code 0 but received: ExitStatus(unix_wait_status(134))
STDOUT:
STDERR:E 00:00:00.000543 executorch:operator_registry.cpp:86] Re-registering aten::sym_size.int, from /data/sandcastle/boxes/fbsource/buck-out/v2/gen/fbsource/cfdc20bd56300721/arvr/libraries/wristband/tsn/transformers/__TorchstreamTransformerTestFbcode__/./__TorchstreamTransformerTestFbcode__shared_libs_symlink_tre$
E 00:00:00.000572 executorch:operator_registry.cpp:87] key: (null), is_fallback: true
F 00:00:00.000576 executorch:operator_registry.cpp:111] In function register_kernels(), assert failed (false): Kernel registration failed with error 18, see error log for details.
```
[User post](https://fb.workplace.com/groups/pytorch.edge.users/permalink/1691696305033989/)
[User post 2](https://fb.workplace.com/groups/pytorch.edge.users/permalink/1510414646495490/)

This is worse than duplicate symbols because it's a runtime error. This happens because a user depends on a kernel library built with ATen tensors and a kernel library built with ET tensor at the same time. For example, if a C++ binary depends on `//executorch/kernels/prim_ops:prim_ops_registry` and `//executorch/kernels/prim_ops:prim_ops_registry_aten` then this will happen.

## My proposal

Add a new namespace to the symbols in ATen mode.

`executorch::runtime::Method` --> `executorch::runtime::aten::Method`

This way a C++ binary is allowed to depend on an ET library with ATen mode enabled and an ET library with ATen mode disabled.

This is not BC breaking for OSS users, since ATen mode was never exposed.

Reviewed By: iseeyuan

Differential Revision: D72440313

Author: larryliu0820

codegen/templates/RegisterCodegenUnboxedKernels.cpp

@@ -22,8 +22,8 @@
 // JIT op registry instead of c10 dispatcher. JIT op registry only takes boxed
 // kernels, so we are calling unboxing functions in UnboxingFunctions.h to cast
 // arguments into C++ types (instead of IValue) and delegate to unboxed kernels.
-using KernelSpan =
-    ::executorch::runtime::Span<const ::executorch::runtime::Kernel>;
+using KernelSpan = ::executorch::runtime::Span<
+    const ::executorch::ET_RUNTIME_NAMESPACE::Kernel>;
 namespace torch {
 namespace executor {
 namespace function {

devtools/bundled_program/bundled_program.cpp

@@ -27,9 +27,9 @@ using executorch::aten::ArrayRef;
 using executorch::aten::Half;
 using executorch::aten::ScalarType;
 using executorch::aten::Tensor;
+using ::executorch::ET_RUNTIME_NAMESPACE::Method;
 using ::executorch::runtime::Error;
 using ::executorch::runtime::EValue;
-using ::executorch::runtime::Method;
 using ::executorch::runtime::Result;
 
 namespace executorch {
@@ -332,8 +332,9 @@ ET_NODISCARD Error load_bundled_input(
         static_cast<uint32_t>(status));
   }
 
-  ::executorch::runtime::internal::event_tracer_set_bundled_input_index(
-      method.get_event_tracer(), testset_idx);
+  ::executorch::ET_RUNTIME_NAMESPACE::internal::
+      event_tracer_set_bundled_input_index(
+          method.get_event_tracer(), testset_idx);
 
   return Error::Ok;
 }

devtools/bundled_program/bundled_program.h

@@ -18,7 +18,7 @@ namespace bundled_program {
  * An opaque pointer to a serialized bundled program.
  */
 using SerializedBundledProgram = const void;
-
+using ::executorch::ET_RUNTIME_NAMESPACE::Method;
 /**
  * Load testset_idx-th bundled input of method_idx-th Method test in
  * bundled_program_ptr to given Method.
@@ -31,7 +31,7 @@ using SerializedBundledProgram = const void;
  * execution.
  */
 ET_NODISCARD ::executorch::runtime::Error load_bundled_input(
-    ::executorch::runtime::Method& method,
+    Method& method,
     SerializedBundledProgram* bundled_program_ptr,
     size_t testset_idx);
 
@@ -49,7 +49,7 @@ ET_NODISCARD ::executorch::runtime::Error load_bundled_input(
  * execution.
  */
 ET_NODISCARD ::executorch::runtime::Error verify_method_outputs(
-    ::executorch::runtime::Method& method,
+    Method& method,
     SerializedBundledProgram* bundled_program_ptr,
     size_t testset_idx,
     double rtol = 1e-5,
@@ -106,7 +106,7 @@ using serialized_bundled_program =
     ::executorch::bundled_program::SerializedBundledProgram;
 
 ET_NODISCARD inline ::executorch::runtime::Error LoadBundledInput(
-    ::executorch::runtime::Method& method,
+    Method& method,
     serialized_bundled_program* bundled_program_ptr,
     size_t testset_idx) {
   return ::executorch::bundled_program::load_bundled_input(
@@ -115,7 +115,7 @@ ET_NODISCARD inline ::executorch::runtime::Error LoadBundledInput(
 
 ET_NODISCARD inline ::executorch::runtime::Error
 VerifyResultWithBundledExpectedOutput(
-    ::executorch::runtime::Method& method,
+    Method& method,
     serialized_bundled_program* bundled_program_ptr,
     size_t testset_idx,
     double rtol = 1e-5,

examples/portable/custom_ops/custom_ops_2_out.cpp

@@ -13,7 +13,7 @@ namespace native {
 
 using executorch::aten::ScalarType;
 using executorch::aten::Tensor;
-using executorch::runtime::KernelRuntimeContext;
+using executorch::ET_RUNTIME_NAMESPACE::KernelRuntimeContext;
 
 namespace {
 void check_preconditions(const Tensor& in, Tensor& out) {

extension/flat_tensor/flat_tensor_data_map.cpp

@@ -25,8 +25,8 @@ using executorch::runtime::Result;
 using executorch::runtime::Span;
 
 using executorch::aten::ScalarType;
+using executorch::ET_RUNTIME_NAMESPACE::TensorLayout;
 using executorch::runtime::DataLoader;
-using executorch::runtime::TensorLayout;
 
 namespace executorch {
 namespace extension {

extension/flat_tensor/flat_tensor_data_map.h

@@ -32,7 +32,8 @@ namespace extension {
 /**
  * A NamedDataMap implementation for FlatTensor-serialized data.
  */
-class FlatTensorDataMap final : public executorch::runtime::NamedDataMap {
+class FlatTensorDataMap final
+    : public executorch::ET_RUNTIME_NAMESPACE::NamedDataMap {
  public:
   /**
    * Creates a new DataMap that wraps FlatTensor data.
@@ -51,7 +52,8 @@ class FlatTensorDataMap final : public executorch::runtime::NamedDataMap {
    * @return Error::NotFound if the key is not present.
    */
   ET_NODISCARD
-  executorch::runtime::Result<const executorch::runtime::TensorLayout>
+  executorch::runtime::Result<
+      const executorch::ET_RUNTIME_NAMESPACE::TensorLayout>
   get_metadata(const char* key) const override;
 
   /**

extension/flat_tensor/targets.bzl

@@ -1,24 +1,26 @@
 load("@fbsource//xplat/executorch/build:runtime_wrapper.bzl", "runtime")
 
 def define_common_targets():
-    runtime.cxx_library(
-        name = "flat_tensor_data_map",
-        srcs = [
-            "flat_tensor_data_map.cpp",
-        ],
-        exported_headers = ["flat_tensor_data_map.h"],
-        deps = [
-            "//executorch/runtime/core:core",
-            "//executorch/runtime/core:evalue",
-            "//executorch/runtime/core:named_data_map",
-            "//executorch/runtime/core/exec_aten:lib",
-            "//executorch/runtime/core/exec_aten/util:tensor_util",
-        ],
-        exported_deps = [
-            "//executorch/extension/flat_tensor/serialize:flat_tensor_header",
-            "//executorch/extension/flat_tensor/serialize:generated_headers",
-        ],
-        visibility = [
-            "//executorch/...",
-        ],
-    )
+    for aten_mode in [True, False]:
+        aten_suffix = "_aten" if aten_mode else ""
+        runtime.cxx_library(
+            name = "flat_tensor_data_map" + aten_suffix,
+            srcs = [
+                "flat_tensor_data_map.cpp",
+            ],
+            exported_headers = ["flat_tensor_data_map.h"],
+            deps = [
+                "//executorch/runtime/core:core",
+                "//executorch/runtime/core:evalue",
+                "//executorch/runtime/core:named_data_map" + aten_suffix,
+                "//executorch/runtime/core/exec_aten:lib" + aten_suffix,
+                "//executorch/runtime/core/exec_aten/util:tensor_util",
+            ],
+            exported_deps = [
+                "//executorch/extension/flat_tensor/serialize:flat_tensor_header",
+                "//executorch/extension/flat_tensor/serialize:generated_headers",
+            ],
+            visibility = [
+                "//executorch/...",
+            ],
+        )

extension/module/module.cpp

@@ -37,6 +37,9 @@
 namespace executorch {
 namespace extension {
 
+using ET_RUNTIME_NAMESPACE::MethodMeta;
+using ET_RUNTIME_NAMESPACE::Program;
+
 namespace {
 runtime::Result<std::unique_ptr<runtime::DataLoader>> load_file(
     const std::string& file_path,
@@ -113,7 +116,7 @@ Module::Module(
 }
 
 Module::Module(
-    std::shared_ptr<runtime::Program> program,
+    std::shared_ptr<Program> program,
     std::unique_ptr<runtime::MemoryAllocator> memory_allocator,
     std::unique_ptr<runtime::MemoryAllocator> temp_allocator,
     std::unique_ptr<runtime::EventTracer> event_tracer,
@@ -131,7 +134,7 @@ Module::Module(
   runtime::runtime_init();
 }
 
-runtime::Error Module::load(const runtime::Program::Verification verification) {
+runtime::Error Module::load(const Program::Verification verification) {
   if (!is_loaded()) {
     // Load the program
     if (!data_loader_) {
@@ -156,10 +159,10 @@ runtime::Error Module::load(const runtime::Program::Verification verification) {
     }
     // else: either the map itself was provided or we have no data map, either
     // way no work to do.
-    auto program = ET_UNWRAP_UNIQUE(
-        runtime::Program::load(data_loader_.get(), verification));
-    program_ = std::shared_ptr<runtime::Program>(
-        program.release(), [](runtime::Program* pointer) { delete pointer; });
+    auto program =
+        ET_UNWRAP_UNIQUE(Program::load(data_loader_.get(), verification));
+    program_ = std::shared_ptr<Program>(
+        program.release(), [](Program* pointer) { delete pointer; });
   }
   return runtime::Error::Ok;
 }
@@ -224,7 +227,7 @@ runtime::Error Module::load_method(
   return runtime::Error::Ok;
 }
 
-runtime::Result<runtime::MethodMeta> Module::method_meta(
+runtime::Result<MethodMeta> Module::method_meta(
     const std::string& method_name) {
   ET_CHECK_OK_OR_RETURN_ERROR(load_method(method_name));
   return methods_.at(method_name).method->method_meta();

extension/module/module.h

@@ -19,6 +19,11 @@
 namespace executorch {
 namespace extension {
 
+using ET_RUNTIME_NAMESPACE::Method;
+using ET_RUNTIME_NAMESPACE::MethodMeta;
+using ET_RUNTIME_NAMESPACE::NamedDataMap;
+using ET_RUNTIME_NAMESPACE::Program;
+
 /**
  * A facade class for loading programs and executing methods within them.
  */
@@ -95,7 +100,7 @@ class Module {
    * @param[in] data_map_loader A DataLoader used for loading external weights.
    */
   explicit Module(
-      std::shared_ptr<runtime::Program> program,
+      std::shared_ptr<Program> program,
       std::unique_ptr<runtime::MemoryAllocator> memory_allocator = nullptr,
       std::unique_ptr<runtime::MemoryAllocator> temp_allocator = nullptr,
       std::unique_ptr<runtime::EventTracer> event_tracer = nullptr,
@@ -116,8 +121,8 @@ class Module {
    */
   ET_NODISCARD
   runtime::Error load(
-      const runtime::Program::Verification verification =
-          runtime::Program::Verification::Minimal);
+      const Program::Verification verification =
+          Program::Verification::Minimal);
 
   /**
    * Checks if the program is loaded.
@@ -134,7 +139,7 @@ class Module {
    *
    * @returns Shared pointer to the program or nullptr if it's not yet loaded.
    */
-  inline std::shared_ptr<runtime::Program> program() const {
+  inline std::shared_ptr<Program> program() const {
     return program_;
   }
 
@@ -224,8 +229,7 @@ class Module {
    * @returns A method metadata, or an error if the program or method failed to
    * load.
    */
-  runtime::Result<runtime::MethodMeta> method_meta(
-      const std::string& method_name);
+  runtime::Result<MethodMeta> method_meta(const std::string& method_name);
 
   /**
    * Execute a specific method with the given input values and retrieve the
@@ -473,20 +477,20 @@ class Module {
     std::vector<runtime::Span<uint8_t>> planned_spans;
     std::unique_ptr<runtime::HierarchicalAllocator> planned_memory;
     std::unique_ptr<runtime::MemoryManager> memory_manager;
-    std::unique_ptr<runtime::Method> method;
+    std::unique_ptr<Method> method;
     std::vector<runtime::EValue> inputs;
   };
 
   std::string file_path_;
   std::string data_map_path_;
   LoadMode load_mode_{LoadMode::MmapUseMlock};
-  std::shared_ptr<runtime::Program> program_;
+  std::shared_ptr<Program> program_;
   std::unique_ptr<runtime::DataLoader> data_loader_;
   std::unique_ptr<runtime::MemoryAllocator> memory_allocator_;
   std::unique_ptr<runtime::MemoryAllocator> temp_allocator_;
   std::unique_ptr<runtime::EventTracer> event_tracer_;
   std::unique_ptr<runtime::DataLoader> data_map_loader_;
-  std::unique_ptr<runtime::NamedDataMap> data_map_;
+  std::unique_ptr<NamedDataMap> data_map_;
 
  protected:
   std::unordered_map<std::string, MethodHolder> methods_;

extension/module/targets.bzl

@@ -25,7 +25,7 @@ def define_common_targets():
                 "//executorch/extension/memory_allocator:malloc_memory_allocator",
                 "//executorch/extension/data_loader:file_data_loader",
                 "//executorch/extension/data_loader:mmap_data_loader",
-                "//executorch/extension/flat_tensor:flat_tensor_data_map",
+                "//executorch/extension/flat_tensor:flat_tensor_data_map" + aten_suffix,
             ],
             exported_deps = [
                 "//executorch/runtime/executor:program" + aten_suffix,

extension/pybindings/pybindings.cpp

@@ -85,26 +85,26 @@ void et_pal_emit_log_message(
 
 namespace py = pybind11;
 using executorch::bundled_program::verify_method_outputs;
+using ::executorch::ET_RUNTIME_NAMESPACE::BackendInterface;
+using ::executorch::ET_RUNTIME_NAMESPACE::get_backend_class;
+using ::executorch::ET_RUNTIME_NAMESPACE::get_backend_name;
+using ::executorch::ET_RUNTIME_NAMESPACE::get_num_registered_backends;
+using ::executorch::ET_RUNTIME_NAMESPACE::get_registered_kernels;
+using ::executorch::ET_RUNTIME_NAMESPACE::Kernel;
+using ::executorch::ET_RUNTIME_NAMESPACE::Method;
+using ::executorch::ET_RUNTIME_NAMESPACE::Program;
 using ::executorch::extension::BufferDataLoader;
 using ::executorch::extension::MallocMemoryAllocator;
 using ::executorch::extension::MmapDataLoader;
 using ::executorch::runtime::ArrayRef;
-using ::executorch::runtime::BackendInterface;
 using ::executorch::runtime::DataLoader;
 using ::executorch::runtime::Error;
 using ::executorch::runtime::EValue;
 using ::executorch::runtime::EventTracerDebugLogLevel;
-using ::executorch::runtime::get_backend_class;
-using ::executorch::runtime::get_backend_name;
-using ::executorch::runtime::get_num_registered_backends;
-using ::executorch::runtime::get_registered_kernels;
 using ::executorch::runtime::HierarchicalAllocator;
-using ::executorch::runtime::Kernel;
 using ::executorch::runtime::MemoryAllocator;
 using ::executorch::runtime::MemoryManager;
-using ::executorch::runtime::Method;
 using ::executorch::runtime::prof_result_t;
-using ::executorch::runtime::Program;
 using ::executorch::runtime::Result;
 using ::executorch::runtime::Span;
 using ::executorch::runtime::Tag;

extension/runner_util/inputs.cpp

@@ -12,12 +12,12 @@
 #include <executorch/runtime/executor/method_meta.h>
 #include <executorch/runtime/platform/log.h>
 
+using executorch::ET_RUNTIME_NAMESPACE::Method;
+using executorch::ET_RUNTIME_NAMESPACE::MethodMeta;
+using executorch::ET_RUNTIME_NAMESPACE::TensorInfo;
 using executorch::runtime::Error;
-using executorch::runtime::Method;
-using executorch::runtime::MethodMeta;
 using executorch::runtime::Result;
 using executorch::runtime::Tag;
-using executorch::runtime::TensorInfo;
 
 namespace executorch {
 namespace extension {

extension/runner_util/inputs.h

@@ -15,6 +15,9 @@
 namespace executorch {
 namespace extension {
 
+using ::executorch::ET_RUNTIME_NAMESPACE::Method;
+using ::executorch::ET_RUNTIME_NAMESPACE::TensorInfo;
+
 /**
  * RAII helper that frees a set of buffers when destroyed. Movable.
  */
@@ -80,7 +83,7 @@ struct PrepareInputTensorsOptions {
  * @returns An error on failure.
  */
 executorch::runtime::Result<BufferCleanup> prepare_input_tensors(
-    executorch::runtime::Method& method,
+    Method& method,
     PrepareInputTensorsOptions options = {});
 
 namespace internal {
@@ -89,8 +92,8 @@ namespace internal {
  * fills it with ones, and sets the input at `input_index`.
  */
 executorch::runtime::Error fill_and_set_input(
-    executorch::runtime::Method& method,
-    executorch::runtime::TensorInfo& tensor_meta,
+    Method& method,
+    TensorInfo& tensor_meta,
     size_t input_index,
     void* data_ptr);
 } // namespace internal

extension/runner_util/inputs_aten.cpp

@@ -15,8 +15,8 @@
 #include <executorch/runtime/executor/method_meta.h>
 
 using executorch::runtime::Error;
-using executorch::runtime::Method;
-using executorch::runtime::TensorInfo;
+using executorch::runtime::aten::Method;
+using executorch::runtime::aten::TensorInfo;
 
 namespace executorch {
 namespace extension {