Enable CoreML by default on macOS wheel builds (#9483)

### Summary
Context: #9481

* Include the `executorchcoreml` pybinding in the builds
* Remove separate installation option
* Turn on CoreML by default for macOS builds
* Add a dependency on coremltools for macOS

### Test plan

CI

```
$ rm -rf cmake-out pip-out dist && ./install_executorch.sh
$ ./examples/models/llama/install_requirements.sh
$ .ci/scripts/test_llama.sh -model stories110M -build_tool cmake -dtype fp32 -mode coreml
$ .ci/scripts/test_llama.sh -model stories110M -build_tool cmake -dtype fp32 -mode xnnpack+custom+quantize_kv
```

cc @larryliu0820 @lucylq

Author: jathu

backends/apple/coreml/compiler/coreml_preprocess.py

@@ -16,8 +16,8 @@
 
 import coremltools as ct
 import coremltools.optimize as cto
-import executorchcoreml
 
+from executorch.backends.apple.coreml import executorchcoreml
 from executorch.exir.backend.backend_details import (
     BackendDetails,
     ExportedProgram,

backends/apple/coreml/runtime/inmemoryfs/setup.py

@@ -10,6 +10,10 @@ SCRIPT_DIR_PATH="$(
     pwd -P
 )"
 
+# TODO(jathu): remove the need to fetch coremltools to build deps for coreml_executor_runner.
+# Keep this version in sync with: pyproject.toml
+COREMLTOOLS_VERSION="8.1"
+
 red=`tput setaf 1`
 green=`tput setaf 2`
 
@@ -24,7 +28,7 @@ rm -rf "$COREML_DIR_PATH/third-party"
 mkdir "$COREML_DIR_PATH/third-party"
 
 echo "${green}ExecuTorch: Cloning coremltools."
-git clone --depth 1 --branch 8.1 "https://github.com/apple/coremltools.git" $COREMLTOOLS_DIR_PATH
+git clone --depth 1 --branch "${COREMLTOOLS_VERSION}" "https://github.com/apple/coremltools.git" $COREMLTOOLS_DIR_PATH
 cd $COREMLTOOLS_DIR_PATH
 
 STATUS=$?
@@ -43,16 +47,7 @@ fi
 
 mkdir "$COREMLTOOLS_DIR_PATH/build"
 cmake -S "$COREMLTOOLS_DIR_PATH" -B "$COREMLTOOLS_DIR_PATH/build"
-cmake --build "$COREMLTOOLS_DIR_PATH/build" --parallel
-
-echo "${green}ExecuTorch: Installing coremltools."
-pip install "$COREMLTOOLS_DIR_PATH"
-
-STATUS=$?
-if [ $STATUS -ne 0 ]; then
-    echo "${red}ExecuTorch: Failed to install coremltools."
-    exit 1
-fi
+cmake --build "$COREMLTOOLS_DIR_PATH/build" --parallel --target mlmodel
 
 echo "${green}ExecuTorch: Cloning nlohmann."
 git clone https://github.com/nlohmann/json.git "$COREML_DIR_PATH/third-party/nlohmann_json"
@@ -62,8 +57,6 @@ if [ $STATUS -ne 0 ]; then
     exit 1
 fi
 
-sh "$COREML_DIR_PATH/scripts/install_inmemoryfs.sh"
-
 echo "${green}ExecuTorch: Copying protobuf files."
 mkdir -p "$COREML_DIR_PATH/runtime/sdk/format/"
 cp -rf "$PROTOBUF_FILES_DIR_PATH" "$COREML_DIR_PATH/runtime/sdk/format/"

backends/apple/coreml/scripts/install_inmemoryfs.sh

@@ -10,12 +10,9 @@
 
 from typing import List, Optional
 
-import executorchcoreml
-
+from executorch.backends.apple.coreml import executorchcoreml
 from executorch.backends.apple.coreml.compiler import CoreMLBackend
-
 from executorch.exir._serialize._program import deserialize_pte_binary
-
 from executorch.exir.schema import (
     BackendDelegate,
     BackendDelegateDataReference,

backends/apple/coreml/scripts/install_requirements.sh

@@ -67,6 +67,8 @@ dependencies=[
   "sympy",
   "tabulate",
   "typing-extensions",
+  # Keep this version in sync with: ./backends/apple/coreml/scripts/install_requirements.sh
+  "coremltools==8.1; platform_system == 'Darwin'",
 ]
 
 [project.urls]

examples/apple/coreml/scripts/extract_coreml_models.py

@@ -84,6 +84,10 @@ def _cmake_args_defines() -> Dict[str, str]:
     return result
 
 
+def _is_macos() -> bool:
+    return sys.platform == "darwin"
+
+
 class ShouldBuild:
     """Indicates whether to build various components."""
 
@@ -125,7 +129,7 @@ def pybindings(cls) -> bool:
 
     @classmethod
     def coreml(cls) -> bool:
-        return cls._is_cmake_arg_enabled("EXECUTORCH_BUILD_COREML", default=False)
+        return cls._is_cmake_arg_enabled("EXECUTORCH_BUILD_COREML", default=_is_macos())
 
     @classmethod
     def mps(cls) -> bool:
@@ -720,8 +724,14 @@ def run(self):
                 "-DEXECUTORCH_BUILD_KERNELS_QUANTIZED=ON",  # add quantized ops to pybindings.
                 "-DEXECUTORCH_BUILD_KERNELS_QUANTIZED_AOT=ON",
             ]
+
             if ShouldBuild.training():
                 build_args += ["--target", "_training_lib"]
+
+            if ShouldBuild.coreml():
+                cmake_args += ["-DEXECUTORCH_BUILD_COREML=ON"]
+                build_args += ["--target", "executorchcoreml"]
+
             build_args += ["--target", "portable_lib"]
             # To link backends into the portable_lib target, callers should
             # add entries like `-DEXECUTORCH_BUILD_XNNPACK=ON` to the CMAKE_ARGS
@@ -855,6 +865,14 @@ def get_ext_modules() -> List[Extension]:
                     "executorch.extension.training.pybindings._training_lib",
                 )
             )
+        if ShouldBuild.coreml():
+            ext_modules.append(
+                BuiltExtension(
+                    src="executorchcoreml.*",
+                    src_dir="backends/apple/coreml",
+                    modpath="executorch.backends.apple.coreml.executorchcoreml",
+                )
+            )
     if ShouldBuild.llama_custom_ops():
         ext_modules.append(
             BuiltFile(