Arm backend: Add support for fill_.Scalar (pytorch#14501)

Adds support for fill_.Scalar. fill_.scalar is decomposed to full_like
which is already supported by Arm backend.


Signed-off-by: Oscar Andersson <[email protected]>
Co-authored-by: Teo Bergkvist <[email protected]>

Author: oscarandersson8218

backends/arm/quantizer/quantization_annotator.py

@@ -360,6 +360,7 @@ def _match_pattern(
     torch.ops.aten.max_pool2d.default,
     torch.ops.aten.full.default,
     torch.ops.aten.full,
+    torch.ops.aten.fill_.Scalar,
     torch.ops.aten.flatten.using_ints,
     torch.ops.aten.dropout.default,
     torch.ops.aten.dropout_.default,
@@ -625,6 +626,7 @@ def annotate_graph(  # type: ignore[return]
             torch.ops.aten.full_like.default,
             torch.ops.aten.full.default,
             torch.ops.aten.full,
+            torch.ops.aten.fill_.Scalar,
             torch.ops.aten.scalar_tensor.default,
         ]:
             node.kwargs = {}

backends/arm/test/ops/test_fill_scalar.py

@@ -0,0 +1,108 @@
+# 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 Tuple
+
+import torch
+from executorch.backends.arm.test import common
+from executorch.backends.arm.test.tester.test_pipeline import (
+    EthosU55PipelineINT,
+    EthosU85PipelineINT,
+    TosaPipelineFP,
+    TosaPipelineINT,
+    VgfPipeline,
+)
+
+aten_op = "torch.ops.aten.fill_.Scalar"
+exir_op = "executorch_exir_dialects_edge__ops_aten_full_like_default"
+
+input_t1 = Tuple[torch.Tensor]
+
+test_data_suite = {
+    "ones_float": [torch.ones(2, 3), 5.0],
+    "ones_int": [torch.ones(2, 3), -3],
+}
+
+
+class FillScalar(torch.nn.Module):
+    def __init__(self):
+        super().__init__()
+
+    def forward(self, y: torch.Tensor, fill_value: int | float):
+        mask = torch.full_like(y, 0)
+        mask.fill_(fill_value)
+        return mask * y
+
+
+@common.parametrize("test_data", test_data_suite)
+def test_fill_scalar_tosa_FP(test_data: Tuple):
+    pipeline = TosaPipelineFP[input_t1](
+        FillScalar(),
+        (*test_data,),
+        aten_op=aten_op,
+        exir_op=exir_op,
+    )
+    pipeline.run()
+
+
+@common.parametrize("test_data", test_data_suite)
+def test_fill_scalar_tosa_INT(test_data: Tuple):
+    pipeline = TosaPipelineINT[input_t1](
+        FillScalar(),
+        (*test_data,),
+        aten_op=aten_op,
+        exir_op=exir_op,
+    )
+    pipeline.run()
+
+
+@common.XfailIfNoCorstone300
+@common.parametrize("test_data", test_data_suite)
+def test_fill_scalar_u55_INT(test_data: Tuple):
+    pipeline = EthosU55PipelineINT[input_t1](
+        FillScalar(),
+        (*test_data,),
+        aten_ops=[aten_op],
+        exir_ops=exir_op,
+    )
+    pipeline.run()
+
+
+@common.XfailIfNoCorstone320
+@common.parametrize("test_data", test_data_suite)
+def test_fill_scalar_u85_INT(test_data: Tuple):
+    pipeline = EthosU85PipelineINT[input_t1](
+        FillScalar(),
+        (*test_data,),
+        aten_ops=[aten_op],
+        exir_ops=exir_op,
+    )
+    pipeline.run()
+
+
+@common.parametrize("test_data", test_data_suite)
+@common.SkipIfNoModelConverter
+def test_fill_scalar_vgf_FP(test_data: input_t1):
+    pipeline = VgfPipeline[input_t1](
+        FillScalar(),
+        (*test_data,),
+        aten_op,
+        exir_op,
+        tosa_version="TOSA-1.0+FP",
+    )
+    pipeline.run()
+
+
+@common.parametrize("test_data", test_data_suite)
+@common.SkipIfNoModelConverter
+def test_fill_scalar_vgf_INT(test_data: input_t1):
+    pipeline = VgfPipeline[input_t1](
+        FillScalar(),
+        (*test_data,),
+        aten_op,
+        exir_op,
+        tosa_version="TOSA-1.0+INT",
+    )
+    pipeline.run()