[Backend Tester] Add flow-specific skipped tests, skip argmin/max on Core ML due to native crashes, float16/64 on Vulkan

backends/test/suite/flow.py

@@ -1,6 +1,6 @@
 import logging
 
-from dataclasses import dataclass
+from dataclasses import dataclass, field
 from typing import Callable
 
 from executorch.backends.test.harness import Tester
@@ -35,6 +35,12 @@ class TestFlow:
     is_delegated: bool = True
     """ Indicates whether the flow is expected to generate CALL_DELEGATE nodes. """
 
+    skip_patterns: list[str] = field(default_factory=lambda: [])
+    """ Tests with names containing any substrings in this list are skipped. """
+
+    def should_skip_test(self, test_name: str) -> bool:
+        return any(pattern in test_name for pattern in self.skip_patterns)
+
 
 def all_flows() -> dict[str, TestFlow]:
     flows = []

backends/test/suite/flows/coreml.py

@@ -19,6 +19,7 @@ def _create_coreml_flow(
             CoreMLTester, minimum_deployment_target=minimum_deployment_target
         ),
         quantize=quantize,
+        skip_patterns=["test_argmin", "test_argmax"],
     )
 
 

backends/test/suite/flows/vulkan.py

@@ -20,6 +20,7 @@ def _create_vulkan_flow_base(
         tester_factory=VulkanTester,
         quantize=quantize_stage_factory is not None,
         quantize_stage_factory=quantize_stage_factory,
+        skip_patterns=["float16", "float64"], # Not supported in swiftshader
     )
 
 

backends/test/suite/generate_markdown_summary.py

@@ -12,6 +12,25 @@
 #
 
 
+def escape_for_markdown(text: str) -> str:
+    """
+    Modify a string to properly display in a markdown table cell.
+    """
+    if not text:
+        return text
+
+    # Replace newlines with <br /> tags
+    escaped = text.replace("\n", "<br />")
+
+    # Escape backslashes.
+    escaped = escaped.replace("\\", "\\\\")
+
+    # Escape pipe characters that would break table structure
+    escaped = escaped.replace("|", "\\|")
+
+    return escaped
+
+
 def generate_markdown(csv_path: str, exit_code: int = 0):  # noqa (C901)
     # Print warning if exit code is non-zero
     if exit_code != 0:
@@ -46,7 +65,7 @@ def generate_markdown(csv_path: str, exit_code: int = 0):  # noqa (C901)
 
     for row in data_rows:
         # Make a copy of the row to avoid modifying the original
-        processed_row = row.copy()
+        processed_row = [escape_for_markdown(cell) for cell in row]
 
         # Count results and collect failed tests
         if result_column_index is not None and result_column_index < len(row):
@@ -96,7 +115,8 @@ def generate_markdown(csv_path: str, exit_code: int = 0):  # noqa (C901)
     # Generate Failed Tests section
     print("# Failed Tests\n")
     if failed_tests:
-        print("| " + " | ".join(header) + " |")
+        escaped_header = [escape_for_markdown(col) for col in header]
+        print("| " + " | ".join(escaped_header) + " |")
         print("|" + "|".join(["---"] * len(header)) + "|")
         for row in failed_tests:
             print("| " + " | ".join(row) + " |")

backends/test/suite/models/__init__.py

@@ -52,6 +52,9 @@ def wrapped_test(self):
             "use_dynamic_shapes": use_dynamic_shapes,
         }
         with TestContext(test_name, test_func.__name__, flow.name, params):
+            if flow.should_skip_test(test_name):
+                raise unittest.SkipTest(f"Skipping test due to matching flow {flow.name} skip patterns")
+
             test_func(self, flow, dtype, use_dynamic_shapes)
 
     wrapped_test._name = test_func.__name__  # type: ignore

backends/test/suite/operators/__init__.py

@@ -97,6 +97,9 @@ def _make_wrapped_test(
 ):
     def wrapped_test(self):
         with TestContext(test_name, test_base_name, flow.name, params):
+            if flow.should_skip_test(test_name):
+                raise unittest.SkipTest(f"Skipping test due to matching flow {flow.name} skip patterns")
+
             test_kwargs = copy.copy(params) or {}
             test_kwargs["flow"] = flow
 

backends/test/suite/operators/test_abs.py

@@ -7,6 +7,8 @@
 # pyre-unsafe
 
 
+import unittest
+
 import torch
 from executorch.backends.test.suite.flow import TestFlow
 
@@ -45,6 +47,7 @@ def test_abs_shapes(self, flow: TestFlow) -> None:
         # 3D tensor
         self._test_op(AbsModel(), (torch.randn(3, 4, 5),), flow)
 
+    @unittest.skip("NaN and Inf are not enforced for backends.")
     def test_abs_edge_cases(self, flow: TestFlow) -> None:
         # Test edge cases
 

backends/test/suite/operators/test_amax.py

@@ -6,6 +6,7 @@
 
 # pyre-unsafe
 
+import unittest
 from typing import List, Optional, Tuple, Union
 
 import torch
@@ -201,6 +202,7 @@ def test_amax_shapes(self, flow: TestFlow) -> None:
             flow,
         )
 
+    @unittest.skip("NaN and Inf are not enforced for backends.")
     def test_amax_edge_cases(self, flow: TestFlow) -> None:
         x = torch.tensor([[1.0, float("inf"), 3.0], [4.0, 5.0, float("inf")]])
         self._test_op(

backends/test/suite/operators/test_amin.py

@@ -6,6 +6,7 @@
 
 # pyre-unsafe
 
+import unittest
 from typing import List, Optional, Tuple, Union
 
 import torch
@@ -203,6 +204,7 @@ def test_amin_shapes(self, flow: TestFlow) -> None:
             flow,
         )
 
+    @unittest.skip("NaN and Inf are not enforced for backends.")
     def test_amin_edge_cases(self, flow: TestFlow) -> None:
         x = torch.tensor([[1.0, float("-inf"), 3.0], [4.0, 5.0, float("-inf")]])
         self._test_op(

backends/test/suite/operators/test_argmax.py

@@ -6,6 +6,7 @@
 
 # pyre-unsafe
 
+import unittest
 from typing import Optional
 
 import torch
@@ -143,6 +144,7 @@ def test_argmax_shapes(self, flow: TestFlow) -> None:
             flow,
         )
 
+    @unittest.skip("NaN and Inf are not enforced for backends.")
     def test_argmax_edge_cases(self, flow: TestFlow) -> None:
         x = torch.tensor([[1.0, float("inf"), 3.0], [4.0, 5.0, float("inf")]])
         self._test_op(

backends/test/suite/operators/test_argmin.py

@@ -6,6 +6,7 @@
 
 # pyre-unsafe
 
+import unittest
 from typing import Optional
 
 import torch
@@ -143,6 +144,7 @@ def test_argmin_shapes(self, flow: TestFlow) -> None:
             flow,
         )
 
+    @unittest.skip("NaN and Inf are not enforced for backends.")
     def test_argmin_edge_cases(self, flow: TestFlow) -> None:
         x = torch.tensor([[1.0, float("-inf"), 3.0], [4.0, 5.0, float("-inf")]])
         self._test_op(

backends/test/suite/operators/test_ceil.py

@@ -7,6 +7,8 @@
 # pyre-unsafe
 
 
+import unittest
+
 import torch
 from executorch.backends.test.suite.flow import TestFlow
 
@@ -45,6 +47,7 @@ def test_ceil_shapes(self, flow: TestFlow) -> None:
         # 3D tensor
         self._test_op(CeilModel(), (torch.randn(3, 4, 5),), flow)
 
+    @unittest.skip("NaN and Inf are not enforced for backends.")
     def test_ceil_edge_cases(self, flow: TestFlow) -> None:
         # Test edge cases
 

backends/test/suite/operators/test_clamp.py

@@ -7,6 +7,8 @@
 # pyre-unsafe
 
 
+import unittest
+
 import torch
 from executorch.backends.test.suite.flow import TestFlow
 
@@ -56,6 +58,7 @@ def test_clamp_shapes(self, flow: TestFlow) -> None:
         # 3D tensor
         self._test_op(model, (torch.randn(3, 4, 5),), flow)
 
+    @unittest.skip("NaN and Inf are not enforced for backends.")
     def test_clamp_edge_cases(self, flow: TestFlow) -> None:
         # Test edge cases
 

backends/test/suite/operators/test_elu.py

@@ -7,6 +7,8 @@
 # pyre-unsafe
 
 
+import unittest
+
 import torch
 from executorch.backends.test.suite.flow import TestFlow
 
@@ -42,5 +44,6 @@ def test_elu_f32_multi_dim(self, flow: TestFlow) -> None:
     def test_elu_f32_alpha(self, flow: TestFlow) -> None:
         self._test_op(Model(alpha=0.5), (torch.randn(3, 4, 5),), flow)
 
+    @unittest.skip("In place activations aren't properly defunctionalized yet.")
     def test_elu_f32_inplace(self, flow: TestFlow) -> None:
         self._test_op(Model(inplace=True), (torch.randn(3, 4, 5),), flow)

backends/test/suite/operators/test_exp.py

@@ -7,6 +7,8 @@
 # pyre-unsafe
 
 
+import unittest
+
 import torch
 from executorch.backends.test.suite.flow import TestFlow
 
@@ -46,6 +48,7 @@ def test_exp_shapes(self, flow: TestFlow) -> None:
         # 3D tensor
         self._test_op(ExpModel(), (torch.randn(3, 4, 5),), flow)
 
+    @unittest.skip("NaN and Inf are not enforced for backends.")
     def test_exp_edge_cases(self, flow: TestFlow) -> None:
         # Test edge cases
 

backends/test/suite/operators/test_floor.py

@@ -7,6 +7,8 @@
 # pyre-unsafe
 
 
+import unittest
+
 import torch
 from executorch.backends.test.suite.flow import TestFlow
 
@@ -42,6 +44,7 @@ def test_floor_shapes(self, flow: TestFlow) -> None:
         # 3D tensor
         self._test_op(FloorModel(), (torch.randn(3, 4, 5),), flow)
 
+    @unittest.skip("NaN and Inf are not enforced for backends.")
     def test_floor_edge_cases(self, flow: TestFlow) -> None:
         # Test edge cases
 

backends/test/suite/operators/test_floor_divide.py

@@ -6,6 +6,8 @@
 
 # pyre-unsafe
 
+import unittest
+
 import torch
 from executorch.backends.test.suite.flow import TestFlow
 
@@ -178,6 +180,7 @@ def test_floor_divide_values(self, flow: TestFlow) -> None:
         y = torch.tensor([-2.0]).expand_as(x).clone()
         self._test_op(model, (x, y), flow, generate_random_test_inputs=False)
 
+    @unittest.skip("NaN and Inf are not enforced for backends.")
     def test_floor_divide_edge_cases(self, flow: TestFlow) -> None:
         # Test edge cases
         model = FloorDivideModel()

backends/test/suite/operators/test_hardsigmoid.py

@@ -7,6 +7,8 @@
 # pyre-unsafe
 
 
+import unittest
+
 import torch
 from executorch.backends.test.suite.flow import TestFlow
 
@@ -38,6 +40,7 @@ def test_hardsigmoid_f32_single_dim(self, flow: TestFlow) -> None:
     def test_hardsigmoid_f32_multi_dim(self, flow: TestFlow) -> None:
         self._test_op(Model(), (torch.randn(2, 3, 4, 5),), flow)
 
+    @unittest.skip("In place activations aren't properly defunctionalized yet.")
     def test_hardsigmoid_f32_inplace(self, flow: TestFlow) -> None:
         self._test_op(Model(inplace=True), (torch.randn(3, 4, 5),), flow)
 

backends/test/suite/operators/test_hardswish.py

@@ -7,6 +7,8 @@
 # pyre-unsafe
 
 
+import unittest
+
 import torch
 from executorch.backends.test.suite.flow import TestFlow
 
@@ -38,6 +40,7 @@ def test_hardswish_f32_single_dim(self, flow: TestFlow) -> None:
     def test_hardswish_f32_multi_dim(self, flow: TestFlow) -> None:
         self._test_op(Model(), (torch.randn(2, 3, 4, 5),), flow)
 
+    @unittest.skip("In place activations aren't properly defunctionalized yet.")
     def test_hardswish_f32_inplace(self, flow: TestFlow) -> None:
         self._test_op(Model(inplace=True), (torch.randn(3, 4, 5),), flow)
 

backends/test/suite/operators/test_hardtanh.py

@@ -7,6 +7,8 @@
 # pyre-unsafe
 
 
+import unittest
+
 import torch
 from executorch.backends.test.suite.flow import TestFlow
 
@@ -45,6 +47,7 @@ def test_hardtanh_f32_multi_dim(self, flow: TestFlow) -> None:
     def test_hardtanh_f32_custom_range(self, flow: TestFlow) -> None:
         self._test_op(Model(min_val=-2.0, max_val=2.0), (torch.randn(3, 4, 5),), flow)
 
+    @unittest.skip("In place activations aren't properly defunctionalized yet.")
     def test_hardtanh_f32_inplace(self, flow: TestFlow) -> None:
         self._test_op(Model(inplace=True), (torch.randn(3, 4, 5),), flow)
 

backends/test/suite/operators/test_leaky_relu.py

@@ -7,6 +7,8 @@
 # pyre-unsafe
 
 
+import unittest
+
 import torch
 from executorch.backends.test.suite.flow import TestFlow
 
@@ -44,6 +46,7 @@ def test_leaky_relu_f32_multi_dim(self, flow: TestFlow) -> None:
     def test_leaky_relu_f32_custom_slope(self, flow: TestFlow) -> None:
         self._test_op(Model(negative_slope=0.1), (torch.randn(3, 4, 5),), flow)
 
+    @unittest.skip("In place activations aren't properly defunctionalized yet.")
     def test_leaky_relu_f32_inplace(self, flow: TestFlow) -> None:
         self._test_op(Model(inplace=True), (torch.randn(3, 4, 5),), flow)
 

backends/test/suite/operators/test_log.py

@@ -7,6 +7,8 @@
 # pyre-unsafe
 
 
+import unittest
+
 import torch
 from executorch.backends.test.suite.flow import TestFlow
 
@@ -46,6 +48,7 @@ def test_log_shapes(self, flow: TestFlow) -> None:
         # 3D tensor
         self._test_op(LogModel(), (torch.rand(3, 4, 5) + 0.01,), flow)
 
+    @unittest.skip("NaN and Inf are not enforced for backends.")
     def test_log_edge_cases(self, flow: TestFlow) -> None:
         # Test edge cases
         # Tensor with infinity

backends/test/suite/operators/test_log10.py

@@ -7,6 +7,8 @@
 # pyre-unsafe
 
 
+import unittest
+
 import torch
 from executorch.backends.test.suite.flow import TestFlow
 
@@ -46,6 +48,7 @@ def test_log10_shapes(self, flow: TestFlow) -> None:
         # 3D tensor
         self._test_op(Log10Model(), (torch.rand(3, 4, 5) + 0.01,), flow)
 
+    @unittest.skip("NaN and Inf are not enforced for backends.")
     def test_log10_edge_cases(self, flow: TestFlow) -> None:
         # Test edge cases
         # Tensor with infinity