From 496a88f5ceb7b6fcc3db8289519d16574d67fe9b Mon Sep 17 00:00:00 2001
From: Adrian Lundell <adrian.lundell@arm.com>
Date: Fri, 27 Feb 2026 13:33:44 +0100
Subject: [PATCH 1/2] Cortex-M backend: Enable test linting

Signed-off-by: Adrian Lundell <adrian.lundell@arm.com>
Change-Id: I066358eb3033e07267593f7eb7b84905730cda02
---
 .lintrunner.toml                              |  2 --
 .../cortex_m/test/misc/test_portable_int8.py  |  4 +--
 .../cortex_m/test/misc/test_quantization.py   | 28 +++++++++----------
 .../cortex_m/test/models/test_mobilenet_v2.py |  6 ++--
 .../cortex_m/test/models/test_mobilenet_v3.py |  6 ++--
 backends/cortex_m/test/ops/test_add.py        |  6 ++--
 backends/cortex_m/test/ops/test_conv.py       | 17 ++++++-----
 .../cortex_m/test/ops/test_conv_transpose.py  |  7 +++--
 backends/cortex_m/test/ops/test_lstm.py       |  6 ++--
 backends/cortex_m/test/ops/test_max_pool2d.py |  6 ++--
 backends/cortex_m/test/ops/test_minimum.py    |  6 ++--
 backends/cortex_m/test/ops/test_mul.py        |  6 ++--
 backends/cortex_m/test/ops/test_softmax.py    |  8 +++---
 backends/cortex_m/test/ops/test_transpose.py  |  4 +--
 backends/cortex_m/test/tester.py              | 22 ++++++++++-----
 15 files changed, 70 insertions(+), 64 deletions(-)

diff --git a/.lintrunner.toml b/.lintrunner.toml
index 65e69a1f7c3..15cf0bec06b 100644
--- a/.lintrunner.toml
+++ b/.lintrunner.toml
@@ -376,7 +376,6 @@ exclude_patterns = [
     'scripts/check_binary_dependencies.py',
     'profiler/test/test_profiler_e2e.py',
     'backends/arm/test/ops/*.py',
-    'backends/cortex_m/test/**/*.py',
 ]
 command = [
     'python',
@@ -410,7 +409,6 @@ include_patterns = [
 exclude_patterns = [
     'third-party/**',
     '**/third-party/**',
-    'backends/cortex_m/test/**/*.py',
 ]
 command = [
     'python',
diff --git a/backends/cortex_m/test/misc/test_portable_int8.py b/backends/cortex_m/test/misc/test_portable_int8.py
index 8e08250bb37..82b719230eb 100644
--- a/backends/cortex_m/test/misc/test_portable_int8.py
+++ b/backends/cortex_m/test/misc/test_portable_int8.py
@@ -15,7 +15,7 @@
 from executorch.backends.arm._passes import FoldAndAnnotateQParamsPass
 from executorch.backends.arm._passes.arm_pass_utils import get_first_fake_tensor
 from executorch.backends.arm.quantizer.arm_quantizer_utils import SharedQspecQuantizer
-from executorch.backends.arm.test.common import parametrize
+from executorch.backends.arm.test.common import parametrize, xfail_type
 from executorch.backends.cortex_m.quantizer.quantizer import CortexMQuantizer
 from executorch.backends.cortex_m.test.tester import CortexMTester
 from executorch.backends.test.harness.stages import StageType
@@ -660,7 +660,7 @@ def _quantize_and_export(
     ),
 }
 
-xfails = {
+xfails: dict[str, xfail_type] = {
     "contiguous": "MLETORCH-1863: Contiguos no-op is removed in to-edge, leading to unnecessary Q-DQ-Q-DQ chain.",
     "clamp": "MLETORCH-1864: Support non-fused clamp-type activations.",
     "clamp_tensor": "MLETORCH-1864: Support non-fused clamp-type activations.",
diff --git a/backends/cortex_m/test/misc/test_quantization.py b/backends/cortex_m/test/misc/test_quantization.py
index 1f92c6ffa3f..6f27d1a6c28 100644
--- a/backends/cortex_m/test/misc/test_quantization.py
+++ b/backends/cortex_m/test/misc/test_quantization.py
@@ -6,7 +6,7 @@
 
 import torch
 from executorch.backends.arm._passes.arm_pass_utils import get_first_fake_tensor
-from executorch.backends.arm.test.common import parametrize
+from executorch.backends.arm.test.common import parametrize, xfail_type
 from executorch.backends.cortex_m.test.tester import (
     CortexMTester,
     McuTestCase,
@@ -141,8 +141,8 @@ def forward(self, x, y):
 class SharedQspecInputForkXConstant(torch.nn.Module):
     """Shared qspec cluster with an input fork with left input as global constant."""
 
-    ops_before_transforms = {}
-    ops_after_transforms = {}
+    ops_before_transforms: dict[str, int] = {}
+    ops_after_transforms: dict[str, int] = {}
     constant = torch.tensor(5.0)
 
     def forward(self, x):
@@ -152,8 +152,8 @@ def forward(self, x):
 class SharedQspecInputForkYConstant(torch.nn.Module):
     """Shared qspec cluster with an input fork with left input as local constant."""
 
-    ops_before_transforms = {}
-    ops_after_transforms = {}
+    ops_before_transforms: dict[str, int] = {}
+    ops_after_transforms: dict[str, int] = {}
 
     def forward(self, x):
         return torch.maximum(x, torch.tensor(5.0))
@@ -259,8 +259,8 @@ def forward(self, x):
 
 
 class SharedQspecSurroundedQuantizedOpConstant(torch.nn.Module):
-    ops_before_transforms = {}
-    ops_after_transforms = {}
+    ops_before_transforms: dict[str, int] = {}
+    ops_after_transforms: dict[str, int] = {}
 
     def forward(self, x):
         x1 = torch.clone(x)
@@ -270,16 +270,16 @@ def forward(self, x):
 
 
 class SharedQspecSub(torch.nn.Module):
-    ops_before_transforms = {}
-    ops_after_transforms = {}
+    ops_before_transforms: dict[str, int] = {}
+    ops_after_transforms: dict[str, int] = {}
 
     def forward(self, x, y):
         return torch.clone(x - y)
 
 
 class SharedQspecCompetingQspecs(torch.nn.Module):
-    ops_before_transforms = {}
-    ops_after_transforms = {}
+    ops_before_transforms: dict[str, int] = {}
+    ops_after_transforms: dict[str, int] = {}
 
     def __init__(self):
         super().__init__()
@@ -299,8 +299,8 @@ def forward(self, x):
 
 
 class SharedQspecNoQspecs(torch.nn.Module):
-    ops_before_transforms = {}
-    ops_after_transforms = {}
+    ops_before_transforms: dict[str, int] = {}
+    ops_after_transforms: dict[str, int] = {}
 
     def forward(self, x):
         z = torch.clone(x - x)
@@ -358,7 +358,7 @@ def forward(self, x):
     ),
 }
 
-xfails = {
+xfails: dict[str, xfail_type] = {
     "surrounded_quantized_op_constant": "Numerical error since the add is forced to have non-correct qparams.",
 }
 
diff --git a/backends/cortex_m/test/models/test_mobilenet_v2.py b/backends/cortex_m/test/models/test_mobilenet_v2.py
index 1be721d5b71..5874a09d598 100644
--- a/backends/cortex_m/test/models/test_mobilenet_v2.py
+++ b/backends/cortex_m/test/models/test_mobilenet_v2.py
@@ -10,7 +10,7 @@
 
 from executorch.backends.cortex_m.test.tester import CortexMTester, McuTestCase
 from executorch.backends.test.harness.stages import StageType
-from torchvision import models
+from torchvision import models  # type: ignore[import-untyped]
 
 
 ops_before_transforms: dict[str, int] = {
@@ -56,7 +56,7 @@
 
 @parametrize("test_case", test_cases)
 def test_dialect_mv2(test_case):
-    inputs = test_case.example_inputs()
+    inputs = test_case.get_example_inputs()
     tester = CortexMTester(test_case.model, inputs)
     tester.test_dialect(
         ops_before_transforms,
@@ -78,7 +78,7 @@ def test_dialect_mv2(test_case):
     strict=False,
 )
 def test_implementation_mv2(test_case):
-    inputs = test_case.example_inputs()
+    inputs = test_case.get_example_inputs()
     tester = CortexMTester(test_case.model, inputs)
     tester.test_implementation(
         qtol=10,
diff --git a/backends/cortex_m/test/models/test_mobilenet_v3.py b/backends/cortex_m/test/models/test_mobilenet_v3.py
index b778b94c3a7..08633d54dd6 100644
--- a/backends/cortex_m/test/models/test_mobilenet_v3.py
+++ b/backends/cortex_m/test/models/test_mobilenet_v3.py
@@ -9,7 +9,7 @@
 
 from executorch.backends.cortex_m.test.tester import CortexMTester, McuTestCase
 from executorch.backends.test.harness.stages import StageType
-from torchvision import models
+from torchvision import models  # type: ignore[import-untyped]
 
 
 ops_before_transforms: dict[str, int] = {
@@ -65,7 +65,7 @@
     strict=False,
 )
 def test_dialect_mv3(test_case):
-    inputs = test_case.example_inputs()
+    inputs = test_case.get_example_inputs()
     tester = CortexMTester(test_case.model, inputs)
     tester.test_dialect(
         ops_before_transforms,
@@ -89,7 +89,7 @@ def test_dialect_mv3(test_case):
     strict=False,
 )
 def test_implementation_mv3(test_case):
-    inputs = test_case.example_inputs()
+    inputs = test_case.get_example_inputs()
     tester = CortexMTester(test_case.model, inputs)
     tester.test_implementation(
         qtol=20,
diff --git a/backends/cortex_m/test/ops/test_add.py b/backends/cortex_m/test/ops/test_add.py
index b776ac91002..5adc77ce4aa 100644
--- a/backends/cortex_m/test/ops/test_add.py
+++ b/backends/cortex_m/test/ops/test_add.py
@@ -5,7 +5,7 @@
 
 
 import torch
-from executorch.backends.arm.test.common import parametrize
+from executorch.backends.arm.test.common import parametrize, xfail_type
 from executorch.backends.cortex_m.test.tester import (
     CortexMTester,
     McuTestCase,
@@ -152,10 +152,10 @@ class CortexMAlphaAdd(ModelAlpha):
 }
 
 
-xfails_implementation = {
+xfails_implementation: dict[str, xfail_type] = {
     "alpha": "Expecting kwargs for aten op IR to be empty - alpha arg not supported.",
 }
-xfails_dialect = xfails_implementation | {
+xfails_dialect: dict[str, xfail_type] = xfails_implementation | {
     # Cortex-M quantizer will not quantize additions that require broadcasting
     # leading to the add op not being replaced by a cortex-m specific implementation
     "broadcast_1": "Broadcasting is not supported in Cortex-M backend",
diff --git a/backends/cortex_m/test/ops/test_conv.py b/backends/cortex_m/test/ops/test_conv.py
index 5c970b5512b..5750ccf3bdb 100644
--- a/backends/cortex_m/test/ops/test_conv.py
+++ b/backends/cortex_m/test/ops/test_conv.py
@@ -1,11 +1,11 @@
-# Copyright 2025 Arm Limited and/or its affiliates.
+# Copyright 2025-2026 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 torch
-from executorch.backends.arm.test.common import parametrize
+from executorch.backends.arm.test.common import parametrize, xfail_type
 from executorch.backends.cortex_m.test.tester import (
     CortexMTester,
     McuTestCase,
@@ -14,8 +14,8 @@
 
 
 class CortexMConv1D(torch.nn.Module):
-    ops_before_transforms = {}
-    ops_after_transforms = {}
+    ops_before_transforms: dict[str, int] = {}
+    ops_after_transforms: dict[str, int] = {}
 
     def __init__(self, *args, **kwargs):
         super().__init__()
@@ -72,9 +72,8 @@ def forward(self, x):
 
 
 class CortexMConv3D(torch.nn.Module):
-    ops_before_transforms = {}
-
-    ops_after_transforms = {}
+    ops_before_transforms: dict[str, int] = {}
+    ops_after_transforms: dict[str, int] = {}
 
     def __init__(self, *args, **kwargs):
         super().__init__()
@@ -313,7 +312,7 @@ def forward(self, x):
 }
 
 
-xfails_dialect = {
+xfails_dialect: dict[str, xfail_type] = {
     "conv2d_dilation": "NotImplementedError: 'slow_conv_dilated<>' not implemented for 'Int'",
     "conv1d": "Currently not supported.",
     "conv2d_nchw": "Currently not supported.",
@@ -330,7 +329,7 @@ def test_dialect_conv2d(test_case):
     )
 
 
-xfails_implementation = {
+xfails_implementation: dict[str, xfail_type] = {
     "conv1d": "Currently not supported.",
     "conv3d": "Currently not supported.",
 }
diff --git a/backends/cortex_m/test/ops/test_conv_transpose.py b/backends/cortex_m/test/ops/test_conv_transpose.py
index 9e93076179b..7a91c5e1b6b 100644
--- a/backends/cortex_m/test/ops/test_conv_transpose.py
+++ b/backends/cortex_m/test/ops/test_conv_transpose.py
@@ -1,12 +1,13 @@
 # Copyright (c) Meta Platforms, Inc. and affiliates.
 # All rights reserved.
+# Copyright 2025-2026 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 torch
-from executorch.backends.arm.test.common import parametrize
+from executorch.backends.arm.test.common import parametrize, xfail_type
 from executorch.backends.cortex_m.test.tester import (
     CortexMTester,
     McuTestCase,
@@ -220,7 +221,7 @@ def forward(self, x):
     ),
 }
 
-xfails_dialect = {
+xfails_dialect: dict[str, xfail_type] = {
     # Grouped convolutions not supported by CMSIS-NN - rejected during quantization
     "conv_transpose2d_groups_2": "Grouped transpose conv not supported by CMSIS-NN",
     "conv_transpose2d_depthwise": "Depthwise transpose conv not supported by CMSIS-NN",
@@ -248,7 +249,7 @@ def test_dialect_conv_transpose2d(test_case):
 # Implementation xfails: empty because unsupported configurations are now
 # rejected at AOT time by the quantizer filter, so they fall back to portable
 # ops and work correctly. Only xfails_dialect needs to track these.
-xfails_implementation = {}
+xfails_implementation: dict[str, xfail_type] = {}
 
 
 @parametrize("test_case", test_cases, xfails=xfails_implementation)
diff --git a/backends/cortex_m/test/ops/test_lstm.py b/backends/cortex_m/test/ops/test_lstm.py
index 60d7aba4271..ae9d17762f8 100644
--- a/backends/cortex_m/test/ops/test_lstm.py
+++ b/backends/cortex_m/test/ops/test_lstm.py
@@ -1,4 +1,4 @@
-# Copyright 2025 Arm Limited and/or its affiliates.
+# Copyright 2025-2026 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.
@@ -30,7 +30,7 @@ class CortexMLSTM(torch.nn.Module):
         "executorch_exir_dialects_edge__ops_aten_cat_default": 1,
     }
 
-    ops_after_transforms = {}
+    ops_after_transforms: dict[str, int] = {}
 
     def __init__(self, input_size: int = 4, hidden_size: int = 3) -> None:
         super().__init__()
@@ -59,7 +59,7 @@ class CortexMQuantizableLSTM(torch.nn.Module):
         "executorch_exir_dialects_edge__ops_quantized_decomposed_quantize_per_tensor_default": 27,
     }
 
-    ops_after_transforms = {}
+    ops_after_transforms: dict[str, int] = {}
 
     def __init__(self, input_size: int = 4, hidden_size: int = 3) -> None:
         super().__init__()
diff --git a/backends/cortex_m/test/ops/test_max_pool2d.py b/backends/cortex_m/test/ops/test_max_pool2d.py
index f27cec7a296..0ec3d721c22 100644
--- a/backends/cortex_m/test/ops/test_max_pool2d.py
+++ b/backends/cortex_m/test/ops/test_max_pool2d.py
@@ -5,7 +5,7 @@
 
 
 import torch
-from executorch.backends.arm.test.common import parametrize
+from executorch.backends.arm.test.common import parametrize, xfail_type
 from executorch.backends.cortex_m.test.tester import (
     CortexMTester,
     McuTestCase,
@@ -82,10 +82,10 @@ def forward(self, x: torch.Tensor) -> torch.Tensor:
     ),
 }
 
-xfails_max_pool2d = {
+xfails_max_pool2d: dict[str, xfail_type] = {
     "maxpool_2x2_indices": (
         "Indices output not supported; quantizer does not handle getitem on max_pool2d_with_indices.",
-        (NotImplementedError, AssertionError, RuntimeError, Exception),
+        Exception,
     ),
 }
 
diff --git a/backends/cortex_m/test/ops/test_minimum.py b/backends/cortex_m/test/ops/test_minimum.py
index 633ccdbf483..6a520194f46 100644
--- a/backends/cortex_m/test/ops/test_minimum.py
+++ b/backends/cortex_m/test/ops/test_minimum.py
@@ -1,11 +1,11 @@
-# Copyright 2025 Arm Limited and/or its affiliates.
+# Copyright 2025-2026 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 torch
-from executorch.backends.arm.test.common import parametrize
+from executorch.backends.arm.test.common import parametrize, xfail_type
 from executorch.backends.cortex_m.test.tester import (
     CortexMTester,
     McuTestCase,
@@ -87,7 +87,7 @@ def forward(self, x, y):
 }
 
 
-xfails = {}
+xfails: dict[str, xfail_type] = {}
 
 
 @parametrize("test_case", test_cases, xfails=xfails)
diff --git a/backends/cortex_m/test/ops/test_mul.py b/backends/cortex_m/test/ops/test_mul.py
index 88dd904eb6e..47c5304d83a 100644
--- a/backends/cortex_m/test/ops/test_mul.py
+++ b/backends/cortex_m/test/ops/test_mul.py
@@ -1,11 +1,11 @@
-# Copyright 2025 Arm Limited and/or its affiliates.
+# Copyright 2025-2026 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 torch
-from executorch.backends.arm.test.common import parametrize
+from executorch.backends.arm.test.common import parametrize, xfail_type
 from executorch.backends.cortex_m.test.tester import (
     CortexMTester,
     McuTestCase,
@@ -127,7 +127,7 @@ class CortexMTensorMul(Model):
 }
 
 
-xfail_cases_dialect = {
+xfail_cases_dialect: dict[str, xfail_type] = {
     # Cortex-M quantizer will not quantize multiplicaitons that require broadcasting
     # leading to the mul op not being replaced by a cortex-m specific implementation
     "broadcast_1": "Broadcasting is not supported in Cortex-M backend",
diff --git a/backends/cortex_m/test/ops/test_softmax.py b/backends/cortex_m/test/ops/test_softmax.py
index 49c8c3ecabc..ab490b3ff81 100644
--- a/backends/cortex_m/test/ops/test_softmax.py
+++ b/backends/cortex_m/test/ops/test_softmax.py
@@ -1,11 +1,11 @@
-# Copyright 2025 Arm Limited and/or its affiliates.
+# Copyright 2025-2026 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 torch
-from executorch.backends.arm.test.common import parametrize
+from executorch.backends.arm.test.common import parametrize, xfail_type
 from executorch.backends.cortex_m.test.tester import (
     CortexMTester,
     McuTestCase,
@@ -54,13 +54,13 @@ def forward(self, x: torch.Tensor) -> torch.Tensor:
 }
 
 
-xfail_cases_dialect = {
+xfail_cases_dialect: dict[str, xfail_type] = {
     "dim_not_last": (
         "Softmax stays in ATen when dim isn’t the channel dimension, so dialect expectations fail",
         Exception,
     ),
 }
-xfail_cases_impl = {
+xfail_cases_impl: dict[str, xfail_type] = {
     "dim_not_last": (
         "Softmax on Cortex-M currently supports only the last dimension",
         Exception,
diff --git a/backends/cortex_m/test/ops/test_transpose.py b/backends/cortex_m/test/ops/test_transpose.py
index de16c2f81ad..978cea1ec0d 100644
--- a/backends/cortex_m/test/ops/test_transpose.py
+++ b/backends/cortex_m/test/ops/test_transpose.py
@@ -1,4 +1,4 @@
-# Copyright 2025 Arm Limited and/or its affiliates.
+# Copyright 2025-2026 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.
@@ -69,7 +69,7 @@ def forward(self, x):
     ),
     "permute_rank_1": McuTestCase(
         CortexMPermute((0,)),
-        (ramp_tensor(10, 100, (3)),),
+        (ramp_tensor(10, 100, (3,)),),
     ),
     "transpose_1_2": McuTestCase(
         CortexMTranspose(1, 2),
diff --git a/backends/cortex_m/test/tester.py b/backends/cortex_m/test/tester.py
index 1b7e623ed00..984e79d5b3a 100644
--- a/backends/cortex_m/test/tester.py
+++ b/backends/cortex_m/test/tester.py
@@ -5,7 +5,7 @@
 
 
 from dataclasses import dataclass
-from typing import Any
+from typing import Any, Callable
 
 import torch
 from executorch.backends.arm.test.common import get_u55_compile_spec
@@ -73,7 +73,11 @@ def __init__(self):
 
 class CortexMTester(TesterBase):
     def __init__(self, module, example_inputs):
-        super().__init__(module, example_inputs, cortex_m_stage_classes)
+        if callable(example_inputs):
+            resolved_example_inputs = example_inputs()
+        else:
+            resolved_example_inputs = example_inputs
+        super().__init__(module, resolved_example_inputs, cortex_m_stage_classes)
 
     def test_dialect(
         self,
@@ -124,10 +128,14 @@ def test_implementation(self, qtol=0, calibration_samples=None):
 @dataclass
 class McuTestCase:
     model: torch.nn.Module
-    example_inputs: tuple[Any, ...]
+    example_inputs: tuple[Any, ...] | Callable[[], tuple[Any, ...]]
+
+    def get_example_inputs(self) -> tuple[Any, ...]:
+        if callable(self.example_inputs):
+            return self.example_inputs()
+        return self.example_inputs
 
 
-def ramp_tensor(start: int, end: int, shape: tuple[int, ...]) -> torch.Tensor:
-    return torch.linspace(start, end, steps=torch.prod(torch.tensor(shape))).reshape(
-        shape
-    )
+def ramp_tensor(start: float, end: float, shape: tuple[int, ...]) -> torch.Tensor:
+    steps = int(torch.prod(torch.tensor(shape)).item())
+    return torch.linspace(start, end, steps=steps).reshape(shape)

From a0fd147a7d3a86e1575bcf3f28b8531794254cc0 Mon Sep 17 00:00:00 2001
From: Adrian Lundell <adrian.lundell@arm.com>
Date: Fri, 27 Mar 2026 09:49:32 +0100
Subject: [PATCH 2/2] Cortex-M backend: Add xfail type

Signed-off-by: Adrian Lundell <adrian.lundell@arm.com>
Change-Id: I6eaa632041cf566ee38b3b3ad06346ae5f48c7c1
---
 backends/cortex_m/test/models/test_nn_modules.py | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/backends/cortex_m/test/models/test_nn_modules.py b/backends/cortex_m/test/models/test_nn_modules.py
index f016f94e7d0..77aa07e04c9 100644
--- a/backends/cortex_m/test/models/test_nn_modules.py
+++ b/backends/cortex_m/test/models/test_nn_modules.py
@@ -24,7 +24,7 @@
 """
 
 import torch
-from executorch.backends.arm.test.common import parametrize
+from executorch.backends.arm.test.common import parametrize, xfail_type
 from executorch.backends.cortex_m.test.tester import (
     CortexMTester,
     McuTestCase,
@@ -188,7 +188,7 @@ def forward(self, x):
     ),
 }
 
-xfails = {
+xfails: dict[str, xfail_type] = {
     "conv_add_relu": "Activation fusion does not support relu after add",
 }