Arm backend: Do not partition noop subgraphs (#12995)

- Reject partitions that will be lowered to empty subgraphs, i.e.
containing only clones/ noop expands.
- For eye/one/zeros the graph is not really empty since it contains one
constant, and this works now. Simply move the previosuly xfailing tests
to MI/BI. (u55/ u85 still fails because of missing CPU ops)

cc @digantdesai @freddan80 @per @zingo @oscarandersson8218

Signed-off-by: Adrian Lundell <[email protected]>

Author: AdrianLundell

backends/arm/_passes/convert_expand_copy_to_repeat.py

@@ -8,12 +8,43 @@
 import logging
 from typing import cast
 
+import torch
+
 from executorch.exir.dialects._ops import ops as exir_ops
 from executorch.exir.pass_base import ExportPass
 
 logger = logging.getLogger(__name__)
 
 
+def calculate_multiples(args):
+    input_node_or_tensor = args[0]
+
+    if isinstance(input_node_or_tensor, torch.fx.node.Node):
+        input_data = input_node_or_tensor.meta["val"]
+    else:
+        input_data = input_node_or_tensor.data
+
+    input_shape = input_data.shape
+
+    multiples = cast(list[int], args[1])
+    expanded_rank = len(multiples)
+
+    # Expanded shape is 'input_shape' front-padded with ones.
+    padding = expanded_rank - len(input_shape)
+    extended_shape = [
+        input_shape[i] if i >= 0 else 1 for i in range(-padding, len(input_shape))
+    ]
+
+    # To convert expand arg to repeat arg, non-repeated dims should have
+    # multiples[dim] = 1. Passing -1 to expand arg means
+    # not changing the size of that dimension.
+    multiples = [
+        multiples[i] if multiples[i] != -1 and extended_shape[i] == 1 else 1
+        for i in range(expanded_rank)
+    ]
+    return multiples
+
+
 class ConvertExpandCopyToRepeatPass(ExportPass):
     """
     Replace expand copy with repeat since it is a repeat that can only repeat singleton dimensions.
@@ -26,23 +57,7 @@ def call_operator(self, op, args, kwargs, meta):
         if op != self.expand_copy:
             return super().call_operator(op, args, kwargs, meta)
 
-        input_shape = args[0].data.shape
-        multiples = cast(list[int], args[1])
-        expanded_rank = len(multiples)
-
-        # Expanded shape is 'input_shape' front-padded with ones.
-        padding = expanded_rank - len(input_shape)
-        extended_shape = [
-            input_shape[i] if i >= 0 else 1 for i in range(-padding, len(input_shape))
-        ]
-
-        # To convert expand arg to repeat arg, non-repeated dims should have
-        # multiples[dim] = 1. Passing -1 to expand arg means
-        # not changing the size of that dimension.
-        multiples = [
-            multiples[i] if multiples[i] != -1 and extended_shape[i] == 1 else 1
-            for i in range(expanded_rank)
-        ]
+        multiples = calculate_multiples(args)
 
         if all((x == 1 for x in multiples)):
             # All dimensions/repetitions occur only once. Remove node

backends/arm/_passes/remove_clone_pass.py

@@ -6,9 +6,13 @@
 
 # pyre-unsafe
 
+import logging
+
 from executorch.exir.dialects._ops import ops as exir_ops
 from executorch.exir.pass_base import ExportPass
 
+logger = logging.getLogger(__name__)
+
 
 class RemoveClonePass(ExportPass):
     """Remove all clones from graph_module"""
@@ -21,4 +25,10 @@ def call_operator(self, op, args, kwargs, meta):
             raise ValueError(
                 f"clone operator expects exactly one argument, got {len(args)}"
             )
+
+        if "memory_format" in kwargs:
+            logger.warning(
+                f"Removing clone with memory_format '{kwargs['memory_format']}'."
+            )
+
         return args[0]

backends/arm/operator_support/__init__.py

@@ -6,6 +6,7 @@
 # pyre-unsafe
 
 from . import (  # noqa
+    clone_support,
     convolution_support,
     embedding_support,
     ethos_u55_support,

backends/arm/operator_support/clone_support.py

@@ -0,0 +1,37 @@
+# Copyright 2025 Arm Limited and/or its affiliates.
+#
+# This source code is licensed under the BSD-style license found in the
+# LICENSE file in the root directory of this source tree.
+
+import logging
+
+import torch.fx as fx
+from executorch.backends.arm.operator_support.tosa_supported_operators import (
+    register_tosa_support_check,
+    SupportedTOSAOperatorCheck,
+)
+from executorch.backends.arm.tosa_specification import TosaSpecification
+from executorch.exir.dialects._ops import ops as exir_ops
+
+logger = logging.getLogger(__name__)
+
+
+@register_tosa_support_check
+class CloneSupported(SupportedTOSAOperatorCheck):
+    targets = [exir_ops.edge.aten.clone.default]
+
+    tosa_specs = [
+        TosaSpecification.create_from_string("TOSA-1.0+INT"),
+        TosaSpecification.create_from_string("TOSA-1.0+FP"),
+    ]
+
+    def is_node_tosa_supported(
+        self, node: fx.Node, tosa_spec: TosaSpecification
+    ) -> bool:
+
+        input_node = node.args[0]
+        if not isinstance(input_node, fx.Node):
+            self.reporter.report_reject(node, "Non tensor clones are not supported")
+            return False
+
+        return True

backends/arm/operator_support/tosa_supported_operators.py

@@ -228,7 +228,6 @@ def is_node_supported(
             exir_ops.edge.aten.var.correction,
             exir_ops.edge.aten.var.dim,
             exir_ops.edge.aten.view_copy.default,
-            exir_ops.edge.aten.clone.default,
             exir_ops.edge.aten.unsqueeze_copy.default,
             exir_ops.edge.aten.squeeze_copy.dims,
             exir_ops.edge.aten.pow.Tensor_Scalar,

backends/arm/scripts/parse_test_names.py

@@ -25,6 +25,7 @@
     "unflatten.int",
     "_native_batch_norm_legit_no_training.default",
     "_native_batch_norm_legit.no_stats",
+    "alias_copy.default",
 ]
 ALL_EDGE_OPS = SAMPLE_INPUT.keys() | CUSTOM_EDGE_OPS
 

backends/arm/test/ops/test_alias_copy.py

@@ -41,7 +41,9 @@ def __init__(self):
         super().__init__()
 
     def forward(self, x: torch.Tensor):
-        return torch.alias_copy(x)
+        return (
+            torch.alias_copy(x) * 1
+        )  # Multiply by one to make sure it is partitioned.
 
 
 @common.parametrize("test_data", AliasCopy.test_data)

backends/arm/test/ops/test_clone.py

@@ -3,13 +3,9 @@
 # This source code is licensed under the BSD-style license found in the
 # LICENSE file in the root directory of this source tree.
 
-#
-# Tests the clone op which copies the data of the input tensor (possibly with new data format)
-#
 
 from typing import Tuple
 
-import pytest
 import torch
 
 from executorch.backends.arm.test import common
@@ -28,57 +24,82 @@
 input_t = Tuple[torch.Tensor]
 
 
-class Clone(torch.nn.Module):
-    """A simple module that clones an input tensor."""
+class CloneFirstArg(torch.nn.Module):
+    def forward(self, x):
+        return x.clone() + x
 
-    def forward(self, x: torch.Tensor):
-        return x.clone()
 
+class CloneSecondArg(torch.nn.Module):
+    def forward(self, x):
+        return x * x.clone()
+
+
+class CloneOutput(torch.nn.Module):
+    def forward(self, x):
+        return (x / x).clone()
+
+
+class CloneBothArgs(torch.nn.Module):
+    def forward(self, x):
+        return x.clone() + x.clone()
+
+
+class CloneAfterOtherOp(torch.nn.Module):
+    def forward(self, x):
+        x = x * 2
+        return x.clone() + x
+
+
+class CloneParallelToOtherOp(torch.nn.Module):
+    def forward(self, x):
+        return x * 2 + x.clone()
 
-test_data_suite = {
-    "ones_1D_10": lambda: (torch.ones(10),),
-    "ones_1D_50": lambda: (torch.ones(50),),
-    "rand_1D_20": lambda: (torch.rand(20),),
-    "rand_2D_10x10": lambda: (torch.rand(10, 10),),
-    "rand_3D_5x5x5": lambda: (torch.rand(5, 5, 5),),
-    "rand_4D_2x3x4x5": lambda: (torch.rand(2, 3, 4, 5),),
-    "large_tensor": lambda: (torch.rand(1000),),
-}
 
+delegated_clones = {
+    "clone_first_arg": lambda: (CloneFirstArg, (torch.rand(1, 2, 3, 4),)),
+    "clone_second_arg": lambda: (CloneSecondArg, (torch.rand(1, 2, 3, 4),)),
+    "clone_output": lambda: (CloneOutput, (torch.rand(1, 2, 3, 4),)),
+    "clone_both_args": lambda: (CloneBothArgs, (torch.rand(1, 2, 3, 4),)),
+    "clone_after_other_op": lambda: (CloneAfterOtherOp, (torch.rand(1, 2, 3, 4),)),
+    "clone_parallel_to_other_op": lambda: (
+        CloneParallelToOtherOp,
+        (torch.rand(1, 2, 3, 4),),
+    ),
+}
 
-@common.parametrize("test_data", test_data_suite)
-def test_clone_tosa_FP(test_data: Tuple[torch.Tensor]):
 
+@common.parametrize("input_data", delegated_clones)
+def test_clone_tosa_FP(input_data):
+    module, input_tensor = input_data()
     pipeline = TosaPipelineFP[input_t](
-        Clone(),
-        test_data(),
-        aten_op,
-        exir_op,
+        module(),
+        input_tensor,
+        [],
     )
-
     pipeline.run()
 
 
-@common.parametrize("test_data", test_data_suite)
-def test_clone_tosa_INT(test_data):
+@common.parametrize("input_data", delegated_clones)
+def test_clone_tosa_INT(input_data):
+    module, input_tensor = input_data()
+
     pipeline = TosaPipelineINT[input_t](
-        Clone(),
-        test_data(),
+        module(),
+        input_tensor,
         aten_op,
         exir_op,
     )
     pipeline.run()
 
 
-@common.parametrize("test_data", test_data_suite)
+@common.parametrize("input_data", delegated_clones)
 @common.XfailIfNoCorstone300
-@pytest.mark.xfail(
-    reason="Empty subgraph leads to Vela compilation failure. See: https://jira.arm.com/browse/MLBEDSW-10477"
-)
-def test_clone_u55_INT(test_data):
+def test_clone_u55_INT(input_data):
+    module, input_tensor = input_data()
+
     pipeline = EthosU55PipelineINT[input_t](
-        Clone(),
-        test_data(),
+        module(),
+        input_tensor,
         aten_op,
         exir_op,
         run_on_fvp=True,
@@ -87,15 +108,14 @@ def test_clone_u55_INT(test_data):
     pipeline.run()
 
 
-@common.parametrize("test_data", test_data_suite)
+@common.parametrize("input_data", delegated_clones)
 @common.XfailIfNoCorstone320
-@pytest.mark.xfail(
-    reason="Empty subgraph leads to Vela compilation failure. See: https://jira.arm.com/browse/MLBEDSW-10477"
-)
-def test_clone_u85_INT(test_data):
+def test_clone_u85_INT(input_data):
+    module, input_tensor = input_data()
+
     pipeline = EthosU85PipelineINT[input_t](
-        Clone(),
-        test_data(),
+        module(),
+        input_tensor,
         aten_op,
         exir_op,
         run_on_fvp=True,
@@ -104,27 +124,23 @@ def test_clone_u85_INT(test_data):
     pipeline.run()
 
 
-@common.parametrize("test_data", test_data_suite)
+@common.parametrize("test_data", delegated_clones)
 @common.SkipIfNoModelConverter
-@pytest.mark.xfail(
-    reason="Empty subgraph leads to Vela compilation failure. See: https://jira.arm.com/browse/MLBEDSW-10477"
-)
 def test_clone_vgf_FP(test_data):
+    module, input_tensor = test_data()
     pipeline = VgfPipeline[input_t](
-        Clone(), test_data(), aten_op, exir_op, tosa_version="TOSA-1.0+FP"
+        module(), input_tensor, aten_op, exir_op, tosa_version="TOSA-1.0+FP"
     )
     pipeline.run()
 
 
-@common.parametrize("test_data", test_data_suite)
+@common.parametrize("test_data", delegated_clones)
 @common.SkipIfNoModelConverter
-@pytest.mark.xfail(
-    reason="Empty subgraph leads to Vela compilation failure. See: https://jira.arm.com/browse/MLBEDSW-10477"
-)
 def test_clone_vgf_INT(test_data):
+    module, input_tensor = test_data()
     pipeline = VgfPipeline[input_t](
-        Clone(),
-        test_data(),
+        module(),
+        input_tensor,
         aten_op,
         exir_op,
         tosa_version="TOSA-1.0+INT",