Arm backend: Add is_node_supported checks for 4 ops (#8209)

Erik-Lundell · web-flow · commit 8ec08f9bdee9 · 2025-02-06T17:14:58.000+01:00
Remove unnecessary asserts from op_sigmoid and op_log
Add is_node_supported checks for 4 ops
For convolution, maxpool2d, avgpool2d, and sum.
The checks mostly target hardware constraints on
Ethos-U55, though convolution also checks for some
unsupported behavior.

Signed-off-by: Erik Lundell &lt;erik.lundell@arm.com&gt;
diff --git a/backends/arm/operator_support/__init__.py b/backends/arm/operator_support/__init__.py
@@ -1,8 +1,15 @@
-# Copyright 2024 Arm Limited and/or its affiliates.
+# Copyright 2024-2025 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.
 
 # pyre-unsafe
 
-from . import right_shift_support, to_copy_support, tosa_supported_operators  # noqa
+from . import (  # noqa
+    convolution_support,
+    pool_2d_support,
+    reduce_sum_support,
+    right_shift_support,
+    to_copy_support,
+    tosa_supported_operators,
+)
diff --git a/backends/arm/operator_support/convolution_support.py b/backends/arm/operator_support/convolution_support.py
@@ -0,0 +1,99 @@
+# 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.
+
+from typing import cast
+
+import torch
+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 Tosa_0_80, TosaSpecification
+from executorch.exir.dialects._ops import ops as exir_ops
+
+
+@register_tosa_support_check
+class ConvolutionSupported(SupportedTOSAOperatorCheck):
+    targets = [exir_ops.edge.aten.convolution.default]
+
+    tosa_specs = [
+        TosaSpecification.create_from_string("TOSA-0.80+BI"),
+        TosaSpecification.create_from_string("TOSA-0.80+MI"),
+    ]
+
+    def is_node_supported(self, node: fx.Node, tosa_spec: TosaSpecification):
+
+        # Not implemented
+        transposed = cast(bool, node.args[6])
+        output_padding = cast(list[int], node.args[7])
+        if transposed:
+            return False
+
+        for pad in output_padding:
+            if pad != 0:
+                return False
+
+        # Hardware specific constraints
+        if not (isinstance(tosa_spec, Tosa_0_80) and tosa_spec.is_U55_subset):
+            return True
+        else:
+            return self._is_node_supported_u55(node)
+
+    def _is_node_supported_u55(self, node: fx.Node):
+        """Hardware constraints for Ethos-U-55 case, Vela 4.2.0 (25.02 release)"""
+
+        shape_in = cast(torch.Tensor, node.all_input_nodes[0].meta["val"]).shape
+        shape_out = node.meta["val"].shape
+        kernel = cast(fx.Node, node.args[1]).meta["val"].shape
+        group = cast(int, node.args[8])
+
+        C_in = shape_in[1]
+        C_out = shape_out[1]
+        if (C_in == group) and (C_out % C_in) == 0:
+            # Depthwise convolution
+            for dim in shape_in[1:]:
+                if not 1 <= dim <= 65536:
+                    return False
+        else:
+            # Convolution
+            if not 1 <= C_in <= 65536:
+                return False
+
+        kernel_w = kernel[2]
+        kernel_h = kernel[3] if len(kernel) > 3 else 1
+        # Kernel condition misses constraint on sum of absolute weights
+        if not 1 <= kernel_h <= 64 or not 1 <= kernel_w * kernel_h <= 4096:
+            return False
+
+        if not self._stride_condition(node):
+            return False
+
+        return True
+
+    def _stride_condition(self, node: fx.Node) -> bool:
+        """This condition is somewhat complex but boils down
+        to not supporting stride > 3, unless we have some special conditions.
+        This condition is a simplified, relaxed version of the hardware constraint,
+        since the actual constraint requires information not available
+        here (without a lot of work).
+
+        This means that we might accept ops that are not actually supported.
+        """
+        strides = cast(list[int], node.args[3])
+        has_padding = any(pad > 0 for pad in cast(list[int], node.args[4]))
+        dilations = cast(list[int], node.args[5])
+        if len(dilations) == 1:
+            dilations = [dilations[0]] * 2
+        if len(strides) == 1:
+            strides = [strides[0]] * 2
+
+        for stride, dilation in zip(strides, dilations):
+            stride_condition = 1 <= stride <= 3
+            dilation_condition = (not has_padding) and (dilation == 1)
+            if (not stride_condition) and (not dilation_condition):
+                return False
+
+        return True
diff --git a/backends/arm/operator_support/pool_2d_support.py b/backends/arm/operator_support/pool_2d_support.py
@@ -0,0 +1,85 @@
+# 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.
+
+from typing import cast
+
+import torch
+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 Tosa_0_80, TosaSpecification
+from executorch.exir.dialects._ops import ops as exir_ops
+
+
+def kernel_check(kernel: tuple[int, int]) -> bool:
+    if not (1 <= kernel[0] * kernel[1] <= 65536):
+        return False
+    return 1 <= kernel[1] <= 256
+
+
+def stride_check(strides: tuple[int, int]) -> bool:
+    return all(1 <= stride <= 3 for stride in strides)
+
+
+def dim_check(shape=torch.Size) -> bool:
+    check = shape[0] == 1
+    for dim in shape:
+        check &= 1 <= dim <= 65536
+    return check
+
+
+@register_tosa_support_check
+class AvgPool2dSupported(SupportedTOSAOperatorCheck):
+    targets = [
+        exir_ops.edge.aten.avg_pool2d.default,
+    ]
+
+    tosa_specs = [
+        TosaSpecification.create_from_string("TOSA-0.80+BI"),
+        TosaSpecification.create_from_string("TOSA-0.80+MI"),
+    ]
+
+    def is_node_supported(self, node: fx.Node, tosa_spec: TosaSpecification):
+        if not (isinstance(tosa_spec, Tosa_0_80) and tosa_spec.is_U55_subset):
+            return True
+
+        # U55 case, Vela 4.2.0 (25.02 release)
+        shape = cast(torch.Tensor, node.all_input_nodes[0].meta["val"]).shape
+        kernel = cast(tuple[int, int], node.args[1])
+        stride = cast(tuple[int, int], node.args[2])
+        if len(node.args) > 3:
+            # Padding case
+            if not all(1 <= k <= 8 for k in kernel):
+                return False
+        else:
+            if not kernel_check(kernel):
+                return False
+
+        return dim_check(shape) and stride_check(stride)
+
+
+@register_tosa_support_check
+class MaxPool2dSupported(SupportedTOSAOperatorCheck):
+    targets = [
+        exir_ops.edge.aten.max_pool2d_with_indices.default,
+    ]
+
+    tosa_specs = [
+        TosaSpecification.create_from_string("TOSA-0.80+BI"),
+        TosaSpecification.create_from_string("TOSA-0.80+MI"),
+    ]
+
+    def is_node_supported(self, node: fx.Node, tosa_spec: TosaSpecification):
+        if not (isinstance(tosa_spec, Tosa_0_80) and tosa_spec.is_U55_subset):
+            return True
+
+        # U55 case, Vela 4.2.0 (25.02 release)
+        shape = cast(torch.Tensor, node.all_input_nodes[0].meta["val"]).shape
+        kernel = cast(tuple[int, int], node.args[1])
+        stride = cast(tuple[int, int], node.args[2])
+
+        return kernel_check(kernel) and dim_check(shape) and stride_check(stride)
diff --git a/backends/arm/operator_support/reduce_sum_support.py b/backends/arm/operator_support/reduce_sum_support.py
@@ -0,0 +1,51 @@
+# 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.
+
+from typing import cast
+
+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 Tosa_0_80, TosaSpecification
+from executorch.exir.dialects._ops import ops as exir_ops
+
+
+@register_tosa_support_check
+class SumSupported(SupportedTOSAOperatorCheck):
+    targets = [exir_ops.edge.aten.sum.dim_IntList]
+
+    tosa_specs = [
+        TosaSpecification.create_from_string("TOSA-0.80+BI"),
+        TosaSpecification.create_from_string("TOSA-0.80+MI"),
+    ]
+
+    def is_node_supported(self, node: fx.Node, tosa_spec: TosaSpecification):
+        if not (isinstance(tosa_spec, Tosa_0_80) and tosa_spec.is_U55_subset):
+            return True
+
+        # U55 case, Vela 4.2.0 (25.02 release)
+        input_shape = node.all_input_nodes[0].meta["val"].shape
+        dim_list = cast(list[int], node.args[1])
+        dim_list = [dim % len(input_shape) for dim in dim_list]
+
+        for dim in dim_list:
+            if not 1 <= input_shape[dim] <= 65536:
+                return False
+
+            # We can't be certain of which dim is the last in memory yet,
+            # Always go for stricter condition.
+            pre_R_product = 1.0
+            for length in input_shape[:dim]:
+                pre_R_product *= length
+            post_R_product = 1.0
+            for length in input_shape[dim + 1 :]:
+                post_R_product *= length
+            if not 1 <= pre_R_product <= 65536:
+                return False
+            if not 1 <= post_R_product <= 65536:
+                return False
+        return True
diff --git a/backends/arm/operator_support/tosa_supported_operators.py b/backends/arm/operator_support/tosa_supported_operators.py
@@ -82,7 +82,6 @@ def is_node_supported(self, submodules, node: fx.Node) -> bool:
             exir_ops.edge.aten.hardsigmoid.default,
             exir_ops.edge.aten.hardtanh.default,
             exir_ops.edge.aten.hardswish.default,
-            exir_ops.edge.aten.convolution.default,
             exir_ops.edge.aten.div.Tensor,
             exir_ops.edge.aten.eq.Tensor,
             exir_ops.edge.aten.exp.default,
@@ -97,8 +96,6 @@ def is_node_supported(self, submodules, node: fx.Node) -> bool:
             exir_ops.edge.aten.mul.Tensor,
             exir_ops.edge.aten._native_batch_norm_legit_no_training.default,
             exir_ops.edge.aten.native_layer_norm.default,
-            exir_ops.edge.aten.avg_pool2d.default,
-            exir_ops.edge.aten.max_pool2d_with_indices.default,
             exir_ops.edge.aten.sigmoid.default,
             exir_ops.edge.aten.mean.dim,
             exir_ops.edge.aten.mm.default,
@@ -113,7 +110,6 @@ def is_node_supported(self, submodules, node: fx.Node) -> bool:
             exir_ops.edge.aten._log_softmax.default,
             exir_ops.edge.aten.slice_copy.Tensor,
             exir_ops.edge.aten.sub.Tensor,
-            exir_ops.edge.aten.sum.dim_IntList,
             exir_ops.edge.aten.tanh.default,
             exir_ops.edge.aten.upsample_nearest2d.vec,
             exir_ops.edge.aten.var.correction,
diff --git a/backends/arm/operators/op_log.py b/backends/arm/operators/op_log.py
@@ -36,7 +36,6 @@ def define_node(
         output: TosaArg,
     ) -> None:
         assert len(node.all_input_nodes) == 1
-        assert len(node.users) == 1
         assert inputs[0].dtype == output.dtype == ts.DType.FP32
 
         tosa_graph.addOperator(TosaOp.Op().LOG, [inputs[0].name], [output.name])
diff --git a/backends/arm/operators/op_sigmoid.py b/backends/arm/operators/op_sigmoid.py
@@ -37,7 +37,6 @@ def define_node(
     ) -> None:
 
         assert len(node.all_input_nodes) == 1
-        assert len(node.users) == 1
         assert inputs[0].dtype == output.dtype == ts.DType.FP32
 
         tosa_graph.addOperator(TosaOp.Op().SIGMOID, [inputs[0].name], [output.name])
diff --git a/backends/arm/test/ops/test_avg_pool.py b/backends/arm/test/ops/test_avg_pool.py
@@ -172,3 +172,36 @@ def test_avgpool2d_tosa_u85_BI(
             common.get_u85_compile_spec(),
             (test_data,),
         )
+
+    reject_data_suite = [
+        (AvgPool2d(1, 1, 0), torch.rand(2, 5, 5, 5)),
+        (AvgPool2d((2, 9), 1, 1), torch.rand(1, 16, 5, 32)),
+        (AvgPool2d(1, 4, 0), torch.rand(1, 10, 10, 10)),
+        (AvgPool2d((1, 257), 1, 0), torch.rand(1, 16, 5, 300)),
+        (AvgPool2d((800, 90), 1, 0), torch.rand(1, 16, 850, 100)),
+    ]
+
+    @parameterized.expand(reject_data_suite)
+    def test_reject_avgpool2d_u55_BI(
+        self,
+        module: torch.nn.Module,
+        test_data: torch.tensor,
+    ):
+        compile_spec = common.get_u55_compile_spec()
+        tosa_spec = TosaSpecification.create_from_compilespecs(compile_spec)
+        quantizer = ArmQuantizer(tosa_spec).set_io(get_symmetric_quantization_config())
+
+        (
+            ArmTester(
+                module,
+                example_inputs=(test_data,),
+                compile_spec=compile_spec,
+            )
+            .quantize(Quantize(quantizer, get_symmetric_quantization_config()))
+            .export()
+            .check_count({"torch.ops.aten.avg_pool2d.default": 1})
+            .check(["torch.ops.quantized_decomposed"])
+            .to_edge_transform_and_lower()
+            .check(["executorch_exir_dialects_edge__ops_aten_avg_pool2d_default"])
+            .check_count({"torch.ops.higher_order.executorch_call_delegate": 0})
+        )
diff --git a/backends/arm/test/ops/test_conv2d.py b/backends/arm/test/ops/test_conv2d.py
@@ -9,6 +9,7 @@
 
 import torch
 from executorch.backends.arm.test import common
+from executorch.backends.arm.test.tester.arm_tester import ArmTester
 from executorch.backends.arm.test.tester.test_pipeline import (
     EthosU55PipelineBI,
     EthosU85PipelineBI,
@@ -406,3 +407,57 @@ def test_conv2d_u85_BI_on_fvp(test_module):
         test_module, test_module.get_inputs(), aten_op, exir_op, run_on_fvp=True
     )
     pipeline.run()
+
+
+reject_suite = {
+    "large_stride": Conv2d(
+        in_channels=1,
+        out_channels=1,
+        kernel_size=(2, 4),
+        stride=(2, 4),
+        padding=1,
+        width=10,
+        height=14,
+        batches=1,
+    ),
+    "large_kernel_height": Conv2d(
+        in_channels=1,
+        out_channels=1,
+        kernel_size=(2, 65),
+        stride=(1, 1),
+        padding=0,
+        width=70,
+        height=70,
+        batches=1,
+    ),
+    "large_kernel": Conv2d(
+        in_channels=1,
+        out_channels=1,
+        kernel_size=(70, 60),
+        stride=(1,),
+        padding=0,
+        width=80,
+        height=80,
+        batches=1,
+    ),
+}
+
+
+@common.parametrize("module", reject_suite)
+def test_reject_conv2d_u55_BI(
+    module: Conv2d,
+):
+    (
+        ArmTester(
+            module,
+            example_inputs=module.get_inputs(),
+            compile_spec=common.get_u55_compile_spec(),
+        )
+        .quantize()
+        .export()
+        .check_count({"torch.ops.aten.conv2d.default": 1})
+        .check(["torch.ops.quantized_decomposed"])
+        .to_edge_transform_and_lower()
+        .check(["executorch_exir_dialects_edge__ops_aten_convolution_default"])
+        .check_count({"torch.ops.higher_order.executorch_call_delegate": 0})
+    )
diff --git a/backends/arm/test/ops/test_max_pool.py b/backends/arm/test/ops/test_max_pool.py
diff --git a/backends/arm/test/ops/test_sum.py b/backends/arm/test/ops/test_sum.py
