Add deprecation warning for to_edge + to_backend workflow in CoreMLPa… (#17082)

Fixes #15960
This PR adds a deprecation warning in the CoreMLPartitioner to guide
users away from the deprecated [to_edge() +
to_backend()](cci:1://file:///c:/Users/moham/Documents/osc/executorch-contribution/backends/xnnpack/test/test_xnnpack_partitioner.py:62:4-88:61)
workflow and toward the recommended
[to_edge_transform_and_lower()](cci:1://file:///c:/Users/moham/Documents/osc/executorch-contribution/backends/xnnpack/test/test_xnnpack_partitioner.py:62:4-88:61)
flow.
## Changes
- Added
[_check_if_called_from_to_backend()](cci:1://file:///c:/Users/moham/Documents/osc/executorch-contribution/backends/apple/coreml/partition/coreml_partitioner.py:225:4-241:20)
method to detect deprecated workflow
- Modified
[partition()](cci:1://file:///c:/Users/moham/Documents/osc/executorch-contribution/backends/apple/coreml/partition/coreml_partitioner.py:243:4-292:9)
to log warning when deprecated flow is detected
- Added two unit tests to verify warning behavior
## Pattern
Following the same pattern as #13209 (XNNPACK).
## Related
This picks up the abandoned PR #15963.

cc @kimishpatel @YifanShenSZ @cymbalrush @metascroy

---------

Signed-off-by: mohammed-saalim <mohammed.saalim.k@gmail.com>
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

Author: mohammed-saalim

backends/apple/coreml/partition/coreml_partitioner.py

@@ -2,6 +2,7 @@
 #
 # Please refer to the license found in the LICENSE file in the root directory of the source tree.
 
+import inspect
 import logging
 from typing import Callable, List, Optional, Tuple
 
@@ -222,7 +223,38 @@ def __init__(
                 self.take_over_mutable_buffer
             ), "When lower_full_graph=True, you must set take_over_mutable_buffer=True"
 
+    def _check_if_called_from_to_backend(self) -> bool:
+        """
+        Check if the partition method is being called from the deprecated to_backend workflow.
+        Returns True if called from deprecated direct to_backend, False if called from to_edge_transform_and_lower.
+        """
+        stack = inspect.stack()
+
+        for frame_info in stack:
+            if frame_info.function == "to_edge_transform_and_lower":
+                return False
+
+        for frame_info in stack:
+            if frame_info.function == "to_backend":
+                filename = frame_info.filename
+                if "program/_program.py" in filename:
+                    return True
+        return False
+
     def partition(self, exported_program: ExportedProgram) -> PartitionResult:
+        """
+        Override partition to add deprecation warning when called from to_backend.
+        """
+        # Check if we're being called from the deprecated to_backend workflow
+        if self._check_if_called_from_to_backend():
+            logger.warning(
+                "\nDEPRECATION WARNING: You are using the deprecated 'to_edge() + to_backend()' workflow. "
+                "This may result in decreased performance because ExecuTorch decomposes ops (e.g., SDPA) "
+                "that CoreML has optimized implementations for. "
+                "Please consider migrating to 'to_edge_transform_and_lower()' for better performance. "
+                "See: https://docs.pytorch.org/executorch/main/backends/coreml/coreml-overview.html#using-the-core-ml-backend"
+            )
+
         # Run the CapabilityBasedPartitioner to return the largest possible
         # subgraphs containing the nodes with the tags
         logger.info("CoreMLPartitioner::partition")

backends/apple/coreml/test/test_coreml_partitioner.py

@@ -3,6 +3,8 @@
 # Please refer to the license found in the LICENSE file in the root directory of the source tree.
 
 import copy
+import io
+import logging
 import sys
 import unittest
 
@@ -336,6 +338,93 @@ def forward(self, x):
                 torch.allclose(et_outputs, eager_outputs, atol=1e-02, rtol=1e-02)
             )
 
+    def test_deprecation_warning_for_to_backend_workflow(self):
+        """
+        Test that the deprecated to_edge + to_backend workflow shows a deprecation warning.
+        """
+
+        class SimpleModel(torch.nn.Module):
+            def __init__(self):
+                super().__init__()
+                self.linear = torch.nn.Linear(10, 5)
+
+            def forward(self, x):
+                return self.linear(x)
+
+        model = SimpleModel()
+        model.eval()
+        x = torch.randn(1, 10)
+
+        exported_model = torch.export.export(model, (x,), strict=True)
+
+        # Capture log output to check for deprecation warning
+        log_capture_string = io.StringIO()
+        ch = logging.StreamHandler(log_capture_string)
+        ch.setLevel(logging.WARNING)
+
+        partitioner_logger = logging.getLogger(
+            "executorch.backends.apple.coreml.partition.coreml_partitioner"
+        )
+        partitioner_logger.addHandler(ch)
+        partitioner_logger.setLevel(logging.WARNING)
+
+        edge = executorch.exir.to_edge(
+            exported_model, compile_config=self.edge_compile_config
+        )
+        partitioner = CoreMLPartitioner()
+
+        edge.to_backend(partitioner)
+
+        log_contents = log_capture_string.getvalue()
+        self.assertIn("DEPRECATION WARNING", log_contents)
+        self.assertIn("to_edge() + to_backend()", log_contents)
+        self.assertIn("to_edge_transform_and_lower()", log_contents)
+
+        # Clean up handler
+        partitioner_logger.removeHandler(ch)
+
+    def test_no_warning_for_to_edge_transform_and_lower_workflow(self):
+        """
+        Test that the recommended to_edge_transform_and_lower workflow does NOT show a deprecation warning.
+        """
+
+        class SimpleModel(torch.nn.Module):
+            def __init__(self):
+                super().__init__()
+                self.linear = torch.nn.Linear(10, 5)
+
+            def forward(self, x):
+                return self.linear(x)
+
+        model = SimpleModel()
+        model.eval()
+        x = torch.randn(1, 10)
+
+        exported_model = torch.export.export(model, (x,), strict=True)
+
+        # Capture log output to check for deprecation warning
+        log_capture_string = io.StringIO()
+        ch = logging.StreamHandler(log_capture_string)
+        ch.setLevel(logging.WARNING)
+
+        partitioner_logger = logging.getLogger(
+            "executorch.backends.apple.coreml.partition.coreml_partitioner"
+        )
+        partitioner_logger.addHandler(ch)
+        partitioner_logger.setLevel(logging.WARNING)
+
+        partitioner = CoreMLPartitioner()
+
+        executorch.exir.to_edge_transform_and_lower(
+            exported_model, partitioner=[partitioner]
+        )
+
+        log_contents = log_capture_string.getvalue()
+        self.assertNotIn("DEPRECATION WARNING", log_contents)
+
+        # Clean up handler
+        partitioner_logger.removeHandler(ch)
+
 
 if __name__ == "__main__":
     test_runner = TestCoreMLPartitioner()
@@ -346,3 +435,5 @@ def forward(self, x):
     test_runner.test_lower_full_graph()
     # test_runner.test_symint_arg()
     test_runner.test_take_over_constant_data_false()
+    test_runner.test_deprecation_warning_for_to_backend_workflow()
+    test_runner.test_no_warning_for_to_edge_transform_and_lower_workflow()