pytorch

d2a90f91 - [opinfo] use dtypes instead of dtypesIfCPU (#67619)

[opinfo] use dtypes instead of dtypesIfCPU (#67619) Summary: Replace usage of `dtypesIfCPU` with `dtypes` in OpInfo class and also make it a mandatory argument. Also added DeprecationWarning on using `dtypesIfCPU` This raises a question : For an OpInfo entry, currently `dtypes` works for any external backend, `dtypesIfCPU` for CPU and `dtypesIfCUDA` and `dtypesIfROCM` for CUDA and ROCm respectively. If we merge `dtypes` and `dtypesIfCPU`, then for cases where external backend `dtypes` don't match cpu `dtypes` then it will lead to failures. Currently there are few issues (5 failures) due to this on XLA (we may add relevant skips for the same). If we agree that skip should be added, then should it be added via OpInfo using decorators mechanism or at the XLA end? I think XLA end makes more sense to me to have one source of skips. <details> <summary>XLA Fail Log</summary> ``` Nov 01 11:48:26 ====================================================================== Nov 01 11:48:26 ERROR [0.016s]: test_reference_eager_histogram_xla_float32 (__main__.TestOpInfoXLA) Nov 01 11:48:26 ---------------------------------------------------------------------- Nov 01 11:48:26 Traceback (most recent call last): Nov 01 11:48:26 File "/opt/conda/lib/python3.6/site-packages/torch/testing/_internal/common_device_type.py", line 371, in instantiated_test Nov 01 11:48:26 result = test(self, **param_kwargs) Nov 01 11:48:26 File "/opt/conda/lib/python3.6/site-packages/torch/testing/_internal/common_device_type.py", line 737, in test_wrapper Nov 01 11:48:26 return test(*args, **kwargs) Nov 01 11:48:26 File "/var/lib/jenkins/workspace/xla/test/test_ops.py", line 411, in test_reference_eager Nov 01 11:48:26 self.compare_with_eager_reference(op, sample_input) Nov 01 11:48:26 File "/var/lib/jenkins/workspace/xla/test/test_ops.py", line 397, in compare_with_eager_reference Nov 01 11:48:26 cpu_inp, cpu_args, cpu_kwargs = cpu(sample_input) Nov 01 11:48:26 File "/var/lib/jenkins/workspace/xla/test/test_ops.py", line 393, in cpu Nov 01 11:48:26 sample.args), to_cpu(sample.kwargs) Nov 01 11:48:26 File "/var/lib/jenkins/workspace/xla/test/test_ops.py", line 386, in to_cpu Nov 01 11:48:26 return {k: to_cpu(v) for k, v in x.items()} Nov 01 11:48:26 File "/var/lib/jenkins/workspace/xla/test/test_ops.py", line 386, in <dictcomp> Nov 01 11:48:26 return {k: to_cpu(v) for k, v in x.items()} Nov 01 11:48:26 File "/var/lib/jenkins/workspace/xla/test/test_ops.py", line 390, in to_cpu Nov 01 11:48:26 raise ValueError("Unknown type {0}!".format(type(x))) Nov 01 11:48:26 ValueError: Unknown type <class 'NoneType'>! Nov 01 11:48:26 Nov 01 11:48:26 ====================================================================== Nov 01 11:48:26 FAIL [0.575s]: test_reference_eager___rmatmul___xla_int64 (__main__.TestOpInfoXLA) Nov 01 11:48:26 ---------------------------------------------------------------------- Nov 01 11:48:26 Traceback (most recent call last): Nov 01 11:48:26 File "/opt/conda/lib/python3.6/site-packages/torch/testing/_internal/common_device_type.py", line 371, in instantiated_test Nov 01 11:48:26 result = test(self, **param_kwargs) Nov 01 11:48:26 File "/opt/conda/lib/python3.6/site-packages/torch/testing/_internal/common_device_type.py", line 737, in test_wrapper Nov 01 11:48:26 return test(*args, **kwargs) Nov 01 11:48:26 File "/var/lib/jenkins/workspace/xla/test/test_ops.py", line 411, in test_reference_eager Nov 01 11:48:26 self.compare_with_eager_reference(op, sample_input) Nov 01 11:48:26 File "/var/lib/jenkins/workspace/xla/test/test_ops.py", line 402, in compare_with_eager_reference Nov 01 11:48:26 self.assertEqual(actual, expected, exact_dtype=True, exact_device=False) Nov 01 11:48:26 File "/var/lib/jenkins/workspace/xla/test/pytorch_test_base.py", line 607, in assertEqual Nov 01 11:48:26 return DeviceTypeTestBase.assertEqual(self, x, y, *args, **kwargs) Nov 01 11:48:26 File "/opt/conda/lib/python3.6/site-packages/torch/testing/_internal/common_utils.py", line 1903, in assertEqual Nov 01 11:48:26 super().assertTrue(result, msg=self._get_assert_msg(msg, debug_msg=debug_msg)) Nov 01 11:48:26 AssertionError: False is not true : Tensors failed to compare as equal!With rtol=0.001 and atol=0.001, found 44 element(s) (out of 50) whose difference(s) exceeded the margin of error (including 0 nan comparisons). The greatest difference was 9.187201950435738e+18 (-9.187201950435738e+18 vs. 34.0), which occurred at index (0, 4). Nov 01 11:48:26 Nov 01 11:48:26 ====================================================================== Nov 01 11:48:26 FAIL [0.137s]: test_reference_eager_linalg_multi_dot_xla_int64 (__main__.TestOpInfoXLA) Nov 01 11:48:26 ---------------------------------------------------------------------- Nov 01 11:48:26 Traceback (most recent call last): Nov 01 11:48:26 File "/opt/conda/lib/python3.6/site-packages/torch/testing/_internal/common_device_type.py", line 371, in instantiated_test Nov 01 11:48:26 result = test(self, **param_kwargs) Nov 01 11:48:26 File "/opt/conda/lib/python3.6/site-packages/torch/testing/_internal/common_device_type.py", line 737, in test_wrapper Nov 01 11:48:26 return test(*args, **kwargs) Nov 01 11:48:26 File "/var/lib/jenkins/workspace/xla/test/test_ops.py", line 411, in test_reference_eager Nov 01 11:48:26 self.compare_with_eager_reference(op, sample_input) Nov 01 11:48:26 File "/var/lib/jenkins/workspace/xla/test/test_ops.py", line 402, in compare_with_eager_reference Nov 01 11:48:26 self.assertEqual(actual, expected, exact_dtype=True, exact_device=False) Nov 01 11:48:26 File "/var/lib/jenkins/workspace/xla/test/pytorch_test_base.py", line 607, in assertEqual Nov 01 11:48:26 return DeviceTypeTestBase.assertEqual(self, x, y, *args, **kwargs) Nov 01 11:48:26 File "/opt/conda/lib/python3.6/site-packages/torch/testing/_internal/common_utils.py", line 1903, in assertEqual Nov 01 11:48:26 super().assertTrue(result, msg=self._get_assert_msg(msg, debug_msg=debug_msg)) Nov 01 11:48:26 AssertionError: False is not true : Tensors failed to compare as equal!With rtol=0.001 and atol=0.001, found 4 element(s) (out of 4) whose difference(s) exceeded the margin of error (including 0 nan comparisons). The greatest difference was 140230883884432.0 (0.0 vs. 140230883884432.0), which occurred at index (0, 0). Nov 01 11:48:26 Nov 01 11:48:26 ====================================================================== Nov 01 11:48:26 FAIL [0.461s]: test_reference_eager_matmul_xla_int64 (__main__.TestOpInfoXLA) Nov 01 11:48:26 ---------------------------------------------------------------------- Nov 01 11:48:26 Traceback (most recent call last): Nov 01 11:48:26 File "/opt/conda/lib/python3.6/site-packages/torch/testing/_internal/common_device_type.py", line 371, in instantiated_test Nov 01 11:48:26 result = test(self, **param_kwargs) Nov 01 11:48:26 File "/opt/conda/lib/python3.6/site-packages/torch/testing/_internal/common_device_type.py", line 737, in test_wrapper Nov 01 11:48:26 return test(*args, **kwargs) Nov 01 11:48:26 File "/var/lib/jenkins/workspace/xla/test/test_ops.py", line 411, in test_reference_eager Nov 01 11:48:26 self.compare_with_eager_reference(op, sample_input) Nov 01 11:48:26 File "/var/lib/jenkins/workspace/xla/test/test_ops.py", line 402, in compare_with_eager_reference Nov 01 11:48:26 self.assertEqual(actual, expected, exact_dtype=True, exact_device=False) Nov 01 11:48:26 File "/var/lib/jenkins/workspace/xla/test/pytorch_test_base.py", line 607, in assertEqual Nov 01 11:48:26 return DeviceTypeTestBase.assertEqual(self, x, y, *args, **kwargs) Nov 01 11:48:26 File "/opt/conda/lib/python3.6/site-packages/torch/testing/_internal/common_utils.py", line 1903, in assertEqual Nov 01 11:48:26 super().assertTrue(result, msg=self._get_assert_msg(msg, debug_msg=debug_msg)) Nov 01 11:48:26 AssertionError: False is not true : Tensors failed to compare as equal!With rtol=0.001 and atol=0.001, found 37 element(s) (out of 50) whose difference(s) exceeded the margin of error (including 0 nan comparisons). The greatest difference was 7.661375630332297e+18 (-7.66128151259864e+18 vs. 94117733658072.0), which occurred at index (4, 5). Nov 01 11:48:26 Nov 01 11:48:26 ====================================================================== Nov 01 11:48:26 FAIL [0.050s]: test_reference_eager_remainder_autodiffed_xla_int64 (__main__.TestOpInfoXLA) Nov 01 11:48:26 ---------------------------------------------------------------------- Nov 01 11:48:26 Traceback (most recent call last): Nov 01 11:48:26 File "/opt/conda/lib/python3.6/site-packages/torch/testing/_internal/common_device_type.py", line 371, in instantiated_test Nov 01 11:48:26 result = test(self, **param_kwargs) Nov 01 11:48:26 File "/opt/conda/lib/python3.6/site-packages/torch/testing/_internal/common_device_type.py", line 737, in test_wrapper Nov 01 11:48:26 return test(*args, **kwargs) Nov 01 11:48:26 File "/var/lib/jenkins/workspace/xla/test/test_ops.py", line 411, in test_reference_eager Nov 01 11:48:26 self.compare_with_eager_reference(op, sample_input) Nov 01 11:48:26 File "/var/lib/jenkins/workspace/xla/test/test_ops.py", line 402, in compare_with_eager_reference Nov 01 11:48:26 self.assertEqual(actual, expected, exact_dtype=True, exact_device=False) Nov 01 11:48:26 File "/var/lib/jenkins/workspace/xla/test/pytorch_test_base.py", line 607, in assertEqual Nov 01 11:48:26 return DeviceTypeTestBase.assertEqual(self, x, y, *args, **kwargs) Nov 01 11:48:26 File "/opt/conda/lib/python3.6/site-packages/torch/testing/_internal/common_utils.py", line 1903, in assertEqual Nov 01 11:48:26 super().assertTrue(result, msg=self._get_assert_msg(msg, debug_msg=debug_msg)) Nov 01 11:48:26 AssertionError: False is not true : Tensors failed to compare as equal!Attempted to compare equality of tensors with different dtypes. Got dtypes torch.int64 and torch.float32. Nov 01 11:48:26 Nov 01 11:48:26 ---------------------------------------------------------------------- ``` </details> Pull Request resolved: https://github.com/pytorch/pytorch/pull/67619 Reviewed By: ngimel Differential Revision: D32541986 Pulled By: mruberry fbshipit-source-id: 793d7d22c3ec9b4778784254ef6f9c980b4b0ce2