Move executorch_call_delegate to torch.ops.higher_order.executorch_call_delegate (#1149)

Summary:
Pull Request resolved: #1149

Move torch.ops.executorch_call_delegate to torch.ops.higher_order.executorch_call_delegate. This is to better organize the namespace for all higher order operators.

Reviewed By: angelayi

Differential Revision: D51032881

fbshipit-source-id: 88b746119b47a2e0478471ac914ee107d66a9350

Author: ydwu4

backends/xnnpack/test/models/inception_v3.py

@@ -37,7 +37,7 @@ def test_fp32_ic3(self):
             .to_edge()
             .check(list(self.all_operators))
             .partition()
-            .check(["torch.ops.executorch_call_delegate"])
+            .check(["torch.ops.higher_order.executorch_call_delegate"])
             .check_not(list(self.all_operators))
             .to_executorch()
             .serialize()
@@ -58,7 +58,7 @@ def test_qs8_ic3(self):
             .to_edge()
             .check(list(ops_after_quantization))
             .partition()
-            .check(["torch.ops.executorch_call_delegate"])
+            .check(["torch.ops.higher_order.executorch_call_delegate"])
             .check_not(list(ops_after_quantization))
             .to_executorch()
             .serialize()

backends/xnnpack/test/models/inception_v4.py

@@ -35,7 +35,7 @@ def test_fp32_ic4(self):
             .to_edge()
             .check(list(self.all_operators))
             .partition()
-            .check(["torch.ops.executorch_call_delegate"])
+            .check(["torch.ops.higher_order.executorch_call_delegate"])
             .check_not(list(self.all_operators))
             .to_executorch()
             .serialize()
@@ -56,7 +56,7 @@ def test_qs8_ic4(self):
             .to_edge()
             .check(list(ops_after_quantization))
             .partition()
-            .check(["torch.ops.executorch_call_delegate"])
+            .check(["torch.ops.higher_order.executorch_call_delegate"])
             .check_not(list(ops_after_quantization))
             .to_executorch()
             .serialize()

backends/xnnpack/test/models/mobilenet_v2.py

@@ -35,7 +35,7 @@ def test_fp32_mv2(self):
             .to_edge()
             .check(list(self.all_operators))
             .partition()
-            .check(["torch.ops.executorch_call_delegate"])
+            .check(["torch.ops.higher_order.executorch_call_delegate"])
             .check_not(list(self.all_operators))
             .to_executorch()
             .serialize()
@@ -56,7 +56,7 @@ def test_qs8_mv2(self):
             .to_edge()
             .check(list(ops_after_quantization))
             .partition()
-            .check(["torch.ops.executorch_call_delegate"])
+            .check(["torch.ops.higher_order.executorch_call_delegate"])
             .check_not(list(ops_after_quantization))
             .to_executorch()
             .serialize()

backends/xnnpack/test/models/mobilenet_v3.py

@@ -37,7 +37,7 @@ def test_fp32_mv3(self):
             .to_edge()
             .check(list(self.all_operators))
             .partition()
-            .check(["torch.ops.executorch_call_delegate"])
+            .check(["torch.ops.higher_order.executorch_call_delegate"])
             .check_not(list(self.all_operators))
             .to_executorch()
             .serialize()
@@ -58,7 +58,7 @@ def test_qs8_mv3(self):
             .to_edge()
             .check(list(ops_after_quantization))
             .partition()
-            .check(["torch.ops.executorch_call_delegate"])
+            .check(["torch.ops.higher_order.executorch_call_delegate"])
             .check_not(list(ops_after_lowering))
             .to_executorch()
             .serialize()

backends/xnnpack/test/models/torchvision_vit.py

@@ -53,7 +53,7 @@ def test_fp32_vit(self):
             .to_edge()
             .check(list(self.all_operators))
             .partition()
-            .check(["torch.ops.executorch_call_delegate"])
+            .check(["torch.ops.higher_order.executorch_call_delegate"])
             .check_not(list(lowerable_xnn_operators))
             .to_executorch()
             .serialize()

backends/xnnpack/test/ops/add.py

@@ -34,7 +34,7 @@ def test_fp32_add(self):
             .to_edge()
             .check_count({"executorch_exir_dialects_edge__ops_aten_add_Tensor": 4})
             .partition()
-            .check_count({"torch.ops.executorch_call_delegate": 1})
+            .check_count({"torch.ops.higher_order.executorch_call_delegate": 1})
             .check_not(["executorch_exir_dialects_edge__ops_aten_add_Tensor"])
             .to_executorch()
             .serialize()
@@ -53,7 +53,7 @@ def test_qs8_add(self):
             .to_edge()
             .check_count({"executorch_exir_dialects_edge__ops_aten_add_Tensor": 4})
             .partition(Partition(partitioner=XnnpackQuantizedPartitioner))
-            .check_count({"torch.ops.executorch_call_delegate": 1})
+            .check_count({"torch.ops.higher_order.executorch_call_delegate": 1})
             .check_not(["executorch_exir_dialects_edge__ops_aten_add_Tensor"])
             .check_not(["torch.ops.quantized_decomposed"])
             .to_executorch()
@@ -80,7 +80,7 @@ def test_fp32_add_relu(self):
             .partition()
             .check_not(["executorch_exir_dialects_edge__ops_aten_add_Tensor"])
             .check_not(["executorch_exir_dialects_edge__ops_aten_relu_default"])
-            .check_count({"torch.ops.executorch_call_delegate": 1})
+            .check_count({"torch.ops.higher_order.executorch_call_delegate": 1})
             .to_executorch()
             .serialize()
             .run_method()
@@ -103,7 +103,7 @@ def test_qs8_add_relu(self):
             .check_not(["executorch_exir_dialects_edge__ops_aten_add_Tensor"])
             .check_not(["executorch_exir_dialects_edge__ops_aten_relu_default"])
             .check_not(["torch.ops.quantized_decomposed"])
-            .check_count({"torch.ops.executorch_call_delegate": 1})
+            .check_count({"torch.ops.higher_order.executorch_call_delegate": 1})
             .to_executorch()
             .serialize()
             .run_method()

backends/xnnpack/test/ops/bilinear2d.py

@@ -86,7 +86,7 @@ def test_fp32_static_resize_bilinear2d(self):
             .check(self.ops)
             .partition()
             .check_not(self.ops)
-            .check_count({"torch.ops.executorch_call_delegate": 1})
+            .check_count({"torch.ops.higher_order.executorch_call_delegate": 1})
             .to_executorch()
             .serialize()
             .run_method()
@@ -102,7 +102,7 @@ def test_fp32_static_resize_bilinear2d_with_align_cornesr(self):
             .check(self.ops)
             .partition()
             .check_not(self.ops)
-            .check_count({"torch.ops.executorch_call_delegate": 1})
+            .check_count({"torch.ops.higher_order.executorch_call_delegate": 1})
             .to_executorch()
             .serialize()
             .run_method()
@@ -127,5 +127,5 @@ def test_fp32_static_resize_bilinear2d_antialiased(self):
                     "executorch_exir_dialects_edge__ops_aten__upsample_bilinear2d_aa_default": 2
                 }
             )
-            .check_not(["torch.ops.executorch_call_delegate"])
+            .check_not(["torch.ops.higher_order.executorch_call_delegate"])
         )

backends/xnnpack/test/ops/cat.py

@@ -27,7 +27,7 @@ def test_fp32_cat(self):
             .to_edge()
             .check_count({"executorch_exir_dialects_edge__ops_aten_cat": 1})
             .partition()
-            .check_count({"torch.ops.executorch_call_delegate": 1})
+            .check_count({"torch.ops.higher_order.executorch_call_delegate": 1})
             .check_not(["executorch_exir_dialects_edge__ops_aten_cat"])
             .to_executorch()
             .serialize()
@@ -51,7 +51,7 @@ def test_fp32_cat_negative_dim(self):
             .to_edge()
             .check_count({"executorch_exir_dialects_edge__ops_aten_cat": 1})
             .partition()
-            .check_count({"torch.ops.executorch_call_delegate": 1})
+            .check_count({"torch.ops.higher_order.executorch_call_delegate": 1})
             .check_not(["executorch_exir_dialects_edge__ops_aten_cat"])
             .to_executorch()
             .serialize()

backends/xnnpack/test/ops/conv1d.py

@@ -50,7 +50,7 @@ def test_conv1d(self):
             )
             .partition()
             .check_not(["executorch_exir_dialects_edge__ops_aten_convolution_default"])
-            .check_count({"torch.ops.executorch_call_delegate": 1})
+            .check_count({"torch.ops.higher_order.executorch_call_delegate": 1})
             .to_executorch()
             .serialize()
             .run_method()

backends/xnnpack/test/ops/conv2d.py

@@ -76,7 +76,7 @@ def _test(
             )
             .partition()
             .check_not(["executorch_exir_dialects_edge__ops_aten_convolution_default"])
-            .check_count({"torch.ops.executorch_call_delegate": 1})
+            .check_count({"torch.ops.higher_order.executorch_call_delegate": 1})
             .to_executorch()
             .serialize()
             .run_method()

docs/source/tutorial-xnnpack-delegate-lowering.md

@@ -48,12 +48,12 @@ GraphModule(
 
 def forward(self, arg314_1):
     lowered_module_0 = self.lowered_module_0
-    executorch_call_delegate = torch.ops.executorch_call_delegate(lowered_module_0, arg314_1);  lowered_module_0 = arg314_1 = None
+    executorch_call_delegate = torch.ops.higher_order.executorch_call_delegate(lowered_module_0, arg314_1);  lowered_module_0 = arg314_1 = None
     getitem = executorch_call_delegate[0];  executorch_call_delegate = None
     aten_view_copy_default = executorch_exir_dialects_edge__ops_aten_view_copy_default(getitem, [1, 1280]);  getitem = None
     aten_clone_default = executorch_exir_dialects_edge__ops_aten_clone_default(aten_view_copy_default);  aten_view_copy_default = None
     lowered_module_1 = self.lowered_module_1
-    executorch_call_delegate_1 = torch.ops.executorch_call_delegate(lowered_module_1, aten_clone_default);  lowered_module_1 = aten_clone_default = None
+    executorch_call_delegate_1 = torch.ops.higher_order.executorch_call_delegate(lowered_module_1, aten_clone_default);  lowered_module_1 = aten_clone_default = None
     getitem_1 = executorch_call_delegate_1[0];  executorch_call_delegate_1 = None
     return (getitem_1,)
 ```

docs/source/tutorials_source/export-to-executorch-tutorial.py

@@ -471,7 +471,7 @@ def forward(self, x):
 print(exported_program.graph_module.lowered_module_0.original_module)
 
 ######################################################################
-# Notice that there is now a ``torch.ops.executorch_call_delegate`` node in the
+# Notice that there is now a ``torch.ops.higher_order.executorch_call_delegate`` node in the
 # graph, which is calling ``lowered_module_0``. Additionally, the contents of
 # ``lowered_module_0`` are the same as the ``lowered_module`` we created
 # previously.
@@ -513,7 +513,7 @@ def f(a, x, b):
 print(delegated_program.graph_module.lowered_module_1.original_module)
 
 ######################################################################
-# Notice that there are now 2 ``torch.ops.executorch_call_delegate`` nodes in the
+# Notice that there are now 2 ``torch.ops.higher_order.executorch_call_delegate`` nodes in the
 # graph, one containing the operations `add, mul` and the other containing the
 # operations `mul, add`.
 #

exir/backend/test/backend_with_compiler_demo.py

@@ -62,7 +62,7 @@ def forward(self, x):
     graph():
         %arg0_1 : [#users=2] = placeholder[target=arg0_1]
         %lowered_module_0 : [#users=1] = get_attr[target=lowered_module_0]
-        %executorch_call_delegate : [#users=1] = call_function[target=torch.ops.executorch_call_delegate](args = (%lowered_module_0, forward, %arg0_1), kwargs = {})
+        %executorch_call_delegate : [#users=1] = call_function[target=torch.ops.higher_order.executorch_call_delegate](args = (%lowered_module_0, forward, %arg0_1), kwargs = {})
         return [executorch_call_delegate]
 
     Args:

exir/backend/test/test_backends.py

@@ -171,7 +171,9 @@ def forward(self, x):
 
         # Check that there exists a call_delegate, representing the call to the
         # delegated function
-        FileCheck().check("torch.ops.executorch_call_delegate").run(graph_module.code)
+        FileCheck().check("torch.ops.higher_order.executorch_call_delegate").run(
+            graph_module.code
+        )
         lowered_submodules = get_lowered_submodules(graph_module)
         self.assertEqual(len(lowered_submodules), 1)
 
@@ -386,7 +388,9 @@ def forward(self, x):
 
         # Check that there exists a call_delegate op, representing the call to the
         # delegated function
-        FileCheck().check("torch.ops.executorch_call_delegate").run(graph_module.code)
+        FileCheck().check("torch.ops.higher_order.executorch_call_delegate").run(
+            graph_module.code
+        )
 
         for node in graph_module.graph.nodes:
             if node.op == "call_function" and node.target == executorch_call_delegate:

exir/backend/test/test_backends_lifted.py

@@ -198,7 +198,9 @@ def forward(self, x):
 
         # Check that there exists a call_delegate, representing the call to the
         # delegated function
-        FileCheck().check("torch.ops.executorch_call_delegate").run(graph_module.code)
+        FileCheck().check("torch.ops.higher_order.executorch_call_delegate").run(
+            graph_module.code
+        )
         lowered_submodules = get_lowered_submodules(graph_module)
         self.assertEqual(len(lowered_submodules), 1)
 
@@ -415,7 +417,9 @@ def forward(self, x):
 
         # Check that there exists a call_delegate op, representing the call to the
         # delegated function
-        FileCheck().check("torch.ops.executorch_call_delegate").run(graph_module.code)
+        FileCheck().check("torch.ops.higher_order.executorch_call_delegate").run(
+            graph_module.code
+        )
 
         for node in graph_module.graph.nodes:
             if node.op == "call_function" and node.target == executorch_call_delegate:

exir/delegate.py

@@ -24,9 +24,7 @@
 from torch.utils._pytree import tree_flatten
 
 
-executorch_call_delegate = HigherOrderOperator(
-    "executorch_call_delegate", _deprecated_global_ns=True
-)
+executorch_call_delegate = HigherOrderOperator("executorch_call_delegate")
 # pyre-ignore
 executorch_call_delegate.fallthrough(torch._C.DispatchKey.PythonDispatcher)
 # pyre-ignore

exir/tests/test_delegate.py

@@ -44,7 +44,7 @@ def g(x: torch.Tensor, y: torch.Tensor) -> torch.Tensor:
         )
 
         def f(x: torch.Tensor, y: torch.Tensor) -> torch.Tensor:
-            return torch.ops.executorch_call_delegate(lowered_module, x, y)
+            return torch.ops.higher_order.executorch_call_delegate(lowered_module, x, y)
 
         orig_res = f(*inputs)
         gm = exir.capture(
@@ -55,7 +55,7 @@ def f(x: torch.Tensor, y: torch.Tensor) -> torch.Tensor:
             ),
         )
         FileCheck().check("lowered_module_0").check(
-            "torch.ops.executorch_call_delegate"
+            "torch.ops.higher_order.executorch_call_delegate"
         ).run(gm.exported_program.graph_module.code)
         self.assertTrue(torch.allclose(orig_res, gm(*inputs)))
 
@@ -77,7 +77,7 @@ def test_to_backend(self) -> None:
         # Check that there exists a call_delegate, representing the call to the
         # delegated function
         FileCheck().check("lowered_module_0").check(
-            "torch.ops.executorch_call_delegate"
+            "torch.ops.higher_order.executorch_call_delegate"
         ).run(graph_module.code)
 
         # Check that there does not exist an add node (from the non-delegated
@@ -90,7 +90,7 @@ def test_to_backend(self) -> None:
         for node in graph_module.graph.nodes:
             if (
                 node.op == "call_function"
-                and node.target == torch.ops.executorch_call_delegate
+                and node.target == torch.ops.higher_order.executorch_call_delegate
             ):
                 # Check that the first argument is the lowered backend module
                 # (which we got from a getattr)

exir/tests/test_quant_lowering_custom_backend_pass.py

@@ -594,7 +594,7 @@ def test(self) -> None:
         # Check the toplevel graph
         FileCheck().check(
             "executorch_exir_dialects_edge__ops_quantized_decomposed_quantize_per_tensor_default"
-        ).check("torch.ops.executorch_call_delegate").check(
+        ).check("torch.ops.higher_order.executorch_call_delegate").check(
             "executorch_exir_dialects_edge__ops_quantized_decomposed_dequantize_per_tensor_default"
         ).check(
             "executorch_exir_dialects_edge__ops_quantized_decomposed_dequantize_per_tensor_default"

exir/verification/arg_validator.py

@@ -18,7 +18,7 @@ class RunHigherOrderOperatorError(Exception):
     """
     Raised when an we try to run delegate or other HigherOrderOperator in a graph module.
     E.g., %executorch_call_delegate : [#users=1] = call_function[
-        target=torch.ops.executorch_call_delegate](args = (%lowered_module_0, %arg0_1), kwargs = {})
+        target=torch.ops.higher_order.executorch_call_delegate](args = (%lowered_module_0, %arg0_1), kwargs = {})
     """
 
     def __init__(self, message: str) -> None: