add aten compatibility layer

Summary:
https://docs.google.com/spreadsheets/d/17mEkBHN0XwXN21pG2IxpANX-jd4hNMrX4x9FthYB5Ok/edit#gid=0

added explicit exec_aten layer. This is the abstraction layer around things that need to work in both aten and lean mode is found.

Killing off "lean" in favor of portable.

Reviewed By: mergennachin

Differential Revision: D47383587

fbshipit-source-id: ebd0ea9247960f1bb7763b91ceb31421982efdd4

Author: JacobSzwejbka

backends/qnnpack/QNNPackBackend.cpp

@@ -1,11 +1,11 @@
 #include <executorch/backends/qnnpack/executor/QNNExecutor.h>
 #include <executorch/backends/qnnpack/qnnpack_schema_generated.h>
 #include <executorch/backends/qnnpack/utils/utils.h>
-#include <executorch/core/kernel_types/util/ScalarTypeUtil.h>
-#include <executorch/core/kernel_types/util/tensor_util.h>
 #include <executorch/runtime/backend/backend_registry.h>
 #include <executorch/runtime/core/error.h>
 #include <executorch/runtime/core/evalue.h>
+#include <executorch/runtime/core/exec_aten/util/scalar_type_util.h>
+#include <executorch/runtime/core/exec_aten/util/tensor_util.h>
 #include <executorch/runtime/platform/profiler.h>
 #include <executorch/threadpool/threadpool.h>
 #include <executorch/util/memory_utils.h>

backends/qnnpack/executor/QNNExecutor.h

@@ -5,7 +5,7 @@
 #include <qnnpack_func.h>
 #include <memory>
 
-#include <executorch/core/kernel_types/kernel_types.h>
+#include <executorch/runtime/core/exec_aten/exec_aten.h>
 
 namespace torch {
 namespace executor {

backends/qnnpack/targets.bzl

@@ -50,7 +50,7 @@ def define_common_targets():
             ],
             exported_headers = ["utils/utils.h"],
             deps = [
-                "//executorch/core/kernel_types:kernel_types" + aten_suffix,
+                "//executorch/runtime/core/exec_aten:lib" + aten_suffix,
                 "//executorch/runtime/backend:backend_registry",
             ],
             visibility = [
@@ -79,8 +79,8 @@ def define_common_targets():
             "@EXECUTORCH_CLIENTS",
         ],
         deps = [
-            "//executorch/core/kernel_types/util:scalar_type_util",
-            "//executorch/core/kernel_types/util:tensor_util",
+            "//executorch/runtime/core/exec_aten/util:scalar_type_util",
+            "//executorch/runtime/core/exec_aten/util:tensor_util",
             "//executorch/runtime/backend:backend_registry",
             "//executorch/threadpool:threadpool",
             "//executorch/util:memory_utils",

backends/qnnpack/test/targets.bzl

@@ -18,8 +18,8 @@ def define_common_targets():
         ],
         platforms = [ANDROID, APPLE, CXX],
         deps = [
-            "//executorch/core/kernel_types/testing:tensor_util",
-            "//executorch/core/kernel_types/util:scalar_type_util",
+            "//executorch/runtime/core/exec_aten/testing_util:tensor_util",
+            "//executorch/runtime/core/exec_aten/util:scalar_type_util",
             "//executorch/util:aten_bridge",
             "//executorch/backends/qnnpack:qnnpack_utils",
         ],

backends/qnnpack/test/test_utils.cpp

@@ -2,9 +2,9 @@
 
 #include <gtest/gtest.h>
 
-#include <executorch/core/kernel_types/kernel_types.h>
-#include <executorch/core/kernel_types/testing/TensorFactory.h>
-#include <executorch/core/kernel_types/testing/TensorUtil.h>
+#include <executorch/runtime/core/exec_aten/exec_aten.h>
+#include <executorch/runtime/core/exec_aten/testing_util/tensor_factory.h>
+#include <executorch/runtime/core/exec_aten/testing_util/tensor_util.h>
 #include <executorch/runtime/platform/runtime.h>
 #include <executorch/util/aten_bridge.h>
 #include "executorch/backends/qnnpack/utils/utils.h"

backends/qnnpack/utils/utils.h

@@ -5,8 +5,8 @@
 #include <limits>
 #include <vector>
 
-#include <executorch/core/kernel_types/kernel_types.h>
 #include <executorch/runtime/core/error.h>
+#include <executorch/runtime/core/exec_aten/exec_aten.h>
 
 #ifdef __aarch64__
 #include <arm_neon.h>

backends/xnnpack/runtime/XNNCompiler.cpp

@@ -2,7 +2,7 @@
 
 #include <executorch/backends/xnnpack/runtime/XNNCompiler.h>
 #include <executorch/backends/xnnpack/xnnpack_schema_generated.h>
-#include <executorch/core/kernel_types/util/ScalarTypeUtil.h>
+#include <executorch/runtime/core/exec_aten/util/scalar_type_util.h>
 #include <executorch/threadpool/threadpool.h>
 #include <unordered_map>
 

backends/xnnpack/runtime/XNNExecutor.h

@@ -2,9 +2,9 @@
 #pragma once
 
 #include <executorch/backends/xnnpack/runtime/XNNStatus.h>
-#include <executorch/core/kernel_types/util/tensor_util.h>
 #include <executorch/runtime/backend/backend_registry.h>
 #include <executorch/runtime/core/error.h>
+#include <executorch/runtime/core/exec_aten/util/tensor_util.h>
 
 #include <xnnpack.h>
 #include <map>

backends/xnnpack/targets.bzl

@@ -57,7 +57,7 @@ def define_common_targets():
             "//executorch/backends/qnnpack:qnnpack_utils",  # TODO Use (1) portable for choose_qparams(), (2) xnnpack for quantize_per_tensor()
             "//executorch/threadpool:threadpool",
             "//executorch/util:memory_utils",
-            "//executorch/core/kernel_types/util:tensor_util",
+            "//executorch/runtime/core/exec_aten/util:tensor_util",
         ],
         # XnnpackBackend.cpp needs to compile with executor as whole
         # @lint-ignore BUCKLINT: Avoid `link_whole=True` (https://fburl.com/avoid-link-whole)

codegen/codegen.bzl

@@ -381,7 +381,7 @@ def executorch_generated_lib(
                 "//executorch/codegen:macros",
             ] + deps,
             exported_deps = [
-                "//executorch/core/kernel_types:kernel_types" + aten_suffix,
+                "//executorch/runtime/core/exec_aten:lib" + aten_suffix,
                 "//executorch/runtime/kernel:kernel_runtime_context" + aten_suffix,
             ],
             xplat_deps = xplat_deps,
@@ -426,7 +426,7 @@ def executorch_generated_lib(
             link_whole = True,
             visibility = visibility,
             deps = [
-                "//executorch/core/kernel_types:kernel_types_aten",
+                "//executorch/runtime/core/exec_aten:lib_aten",
                 "//executorch/runtime/core:core",
                 "//executorch/codegen:macros",
             ] + torch_dep + custom_ops_aten_kernel_deps,

codegen/templates/Functions.h

@@ -3,7 +3,7 @@
 
 #include <tuple>
 
-#include <executorch/core/kernel_types/kernel_types.h> // at::Tensor etc.
+#include <executorch/runtime/core/exec_aten/exec_aten.h> // at::Tensor etc.
 #include <executorch/codegen/macros.h> // TORCH_API
 #include <executorch/runtime/kernel/kernel_runtime_context.h>
 

codegen/templates/NativeFunctions.h

@@ -3,7 +3,7 @@
 
 #include <tuple>
 
-#include <executorch/core/kernel_types/kernel_types.h> // at::Tensor etc.
+#include <executorch/runtime/core/exec_aten/exec_aten.h> // at::Tensor etc.
 #include <executorch/codegen/macros.h> // TORCH_API
 #include <executorch/runtime/kernel/kernel_runtime_context.h>
 

codegen/templates/RegisterCodegenUnboxedKernels.cpp

@@ -1,5 +1,5 @@
-#include <executorch/core/kernel_types/kernel_types.h>
 #include <executorch/runtime/core/evalue.h>
+#include <executorch/runtime/core/exec_aten/exec_aten.h>
 #include <executorch/runtime/kernel/operator_registry.h>
 #include <executorch/runtime/platform/profiler.h>
 #include "${fn_header}" // Generated Function import headers

codegen/templates/RegisterDispatchKeyCustomOps.cpp

@@ -5,7 +5,7 @@
 #include <torch/library.h>
 #include <ATen/Tensor.h>
 
-#include <executorch/core/kernel_types/kernel_types.h>
+#include <executorch/runtime/core/exec_aten/exec_aten.h>
 $ops_headers
 
 namespace torch {

core/kernel_types/test/targets.bzl

@@ -67,7 +67,7 @@ def stride_from_dim_order(sizes: List[int], dim_order: List[bytes]) -> List[int]
     Converts dim order to stride using sizes
     e.g. if sizes = (2, 3, 4) and dim_order = (0, 1, 2) then strides = (12, 4, 1)
     while for the same size if dim_order = (0, 2, 1) then strides = (12, 1, 3)
-    See executorch/core/kernel_types/util/DimOrderUtils.h for details
+    See executorch/runtime/core/exec_aten/util/dim_order_util.h for details
     Args:
         sizes (Tuple[int]): sizes of the tensor
         dim_order (Tuple[bytes]): dim order of the tensor

core/kernel_types/testing/README.md

@@ -39,10 +39,10 @@ the types defined by PyTorch core in the `at` or `c10` namespaces. To retain
 tigher control over CPU and memory runtime behavior, Executorch reimplements
 compatible but restricted subsets of those types.
 
-[`//executorch/core/kernel_types/kernel_types.h`](https://www.internalfb.com/code/fbsource/fbcode/executorch/core/kernel_types/kernel_types.h)
+[`//executorch/runtime/core/exec_aten/exec_aten.h`](https://www.internalfb.com/code/fbsource/fbcode/executorch/runtime/core/exec_aten/exec_aten.h)
 contains the mapping between ATen/c10 types and the Executorch types. The
 Executorch types are defined in other headers in that same directory,
-[`//executorch/core/kernel_types/`](https://www.internalfb.com/code/fbsource/fbcode/executorch/core/kernel_types/).
+[`//executorch/runtime/core/portable_type/`](https://www.internalfb.com/code/fbsource/fbcode/executorch/runtime/core/portable_type/).
 
 The Executorch types are source-compatible with the ATen/c10 types; if you write
 code that works with the Executorch types, then that same code should work when
@@ -318,7 +318,7 @@ _op_target(name = "op_<name>_test")
 By default, this target will depend on
 `//executorch/kernels/portable/cpu:op_<name>`, the core Executor types, and
 some helper test utilities ([see
-headers](https://www.internalfb.com/code/fbsource/fbcode/executorch/core/kernel_types/testing/)),
+headers](https://www.internalfb.com/code/fbsource/fbcode/executorch/runtime/core/exec_aten/testing_util/)),
 but you can add additional deps if you want to.
 
 ### Create a skeleton test .cpp file
@@ -330,9 +330,9 @@ starting point:
 // Copyright (c) Meta Platforms, Inc. and affiliates.
 
 #include <executorch/kernels/test/FunctionHeaderWrapper.h> // Declares the operator
-#include <executorch/core/kernel_types/kernel_types.h>
-#include <executorch/core/kernel_types/testing/TensorFactory.h>
-#include <executorch/core/kernel_types/testing/TensorUtil.h>
+#include <executorch/runtime/core/exec_aten/exec_aten.h>
+#include <executorch/runtime/core/exec_aten/testing_util/tensor_factory.h>
+#include <executorch/runtime/core/exec_aten/testing_util/tensor_util.h>
 
 #include <gtest/gtest.h>
 

core/kernel_types/util/test/targets.bzl

@@ -1,7 +1,7 @@
 // Copyright (c) Meta Platforms, Inc. and affiliates.
 #include <cstring>
 
-#include <executorch/core/kernel_types/util/DimOrderUtils.h>
+#include <executorch/runtime/core/exec_aten/util/dim_order_util.h>
 #include <executorch/runtime/kernel/kernel_includes.h>
 
 namespace torch {

exir/tensor.py

@@ -7,8 +7,8 @@
 #include <cmath>
 #include <limits>
 
-#include <executorch/core/kernel_types/kernel_types.h>
-#include <executorch/core/kernel_types/util/ScalarTypeUtil.h>
+#include <executorch/runtime/core/exec_aten/exec_aten.h>
+#include <executorch/runtime/core/exec_aten/util/scalar_type_util.h>
 #include <executorch/runtime/core/portable_type/scalar.h>
 
 #define ET_CHECK_SCALAR_SAME_TYPE(a__, b__)                      \