Add unit test for waterfalling errors (#73)

No op

Author: PaliC

test/test_adverse_cases.py

@@ -0,0 +1,51 @@
+import pytest
+from BackendBench.torchbench_suite import TorchBenchOpTest
+from BackendBench.eval import eval_one_op
+import BackendBench.backends as backends
+import torch
+
+
+class TestAdaptiveAvgPool2dBackward:
+    # todo: @jiannanWang unskip this test
+    @pytest.mark.skip(reason="Not ready for testing yet as it'd brick the gpu")
+    def test_adaptive_avg_pool2d_backward_gpu(self):
+        """Test on GPU with eval_one_op."""
+        op_test_should_error = TorchBenchOpTest(
+            "aten._adaptive_avg_pool2d_backward.default",
+            ["((T([512, 4096, 56, 56], f16), T([512, 4096, 56, 56], f16)), {})"],
+            None,
+        )
+
+        op_test_should_succeed = TorchBenchOpTest(
+            "aten.addmm.default",
+            ["((T([14, 14], f32), T([14, 14], f32), T([14, 14], f32)), {})"],
+            None,
+        )
+
+        # run test that should brick the gpu due to an illegal memory access
+        backend = backends.AtenBackend()
+        with pytest.raises(RuntimeError):
+            _, _ = eval_one_op(
+                op_test_should_error.op,
+                backend[op_test_should_error.op],
+                list(op_test_should_error.correctness_tests),
+                list(op_test_should_error.performance_tests),
+            )
+
+        # add these in case code changes in eval_one_op. There shouldn't be any errors here
+        torch.cuda.synchronize()
+        torch.cuda.empty_cache()
+
+        # tests that a simple op works afterwards to make sure we recover after an illegal memory access
+        correctness, _ = eval_one_op(
+            op_test_should_succeed.op,
+            backend[op_test_should_succeed.op],
+            list(op_test_should_succeed.correctness_tests),
+            list(op_test_should_succeed.performance_tests),
+        )
+
+        assert correctness == 1.0
+
+
+if __name__ == "__main__":
+    pytest.main([__file__, "-v", "-s"])