From 07dd9e1931f10633d11cfc6fa78413f71fe07cbe Mon Sep 17 00:00:00 2001 From: kaileiyx Date: Mon, 24 Nov 2025 14:07:48 +0800 Subject: [PATCH 1/5] update skip list --- test/xpu/skip_list_common.py | 40 ++++++++++++++++-------------------- 1 file changed, 18 insertions(+), 22 deletions(-) diff --git a/test/xpu/skip_list_common.py b/test/xpu/skip_list_common.py index 9f810433bd..ea8ed29342 100644 --- a/test/xpu/skip_list_common.py +++ b/test/xpu/skip_list_common.py @@ -2,7 +2,11 @@ "test_ops_xpu.py": ( # Jiterator is only supported on CUDA and ROCm GPUs, none are available. # https://github.com/intel/torch-xpu-ops/issues/584 - "_jiterator_", + "test_cow_input_jiterator_2inputs_2outputs_xpu_float32", + "test_cow_input_jiterator_4inputs_with_extra_args_xpu_float32", + "test_cow_input_jiterator_binary_return_by_ref_xpu_float32", + "test_cow_input_jiterator_binary_xpu_float32", + "test_cow_input_jiterator_unary_xpu_float32", # OPs not supported "test_errors_dot_xpu", "test_errors_vdot_xpu", @@ -13,17 +17,23 @@ "test_python_ref_executor__refs_pow_executor_aten_xpu_complex32", "test_python_ref_executor__refs_mul_executor_aten_xpu_complex32", # https://github.com/intel/torch-xpu-ops/issues/2254 - "histogramdd", - "_vdot_", - "_dot_", - "_flash_attention_", - "_efficient_attention_", + "test_dtypes_histogramdd_xpu", + "test_cow_input_histogramdd_xpu_float32", + "test_operator_histogramdd_xpu_float32", + "test_view_replay_histogramdd_xpu_float32", + "test_neg_view_histogramdd_xpu_float64", + "test_comprehensive_vdot_xpu_complex128", + "test_comprehensive_vdot_xpu_complex64", + "test_quick_vdot_xpu_complex128", + "test_quick_vdot_xpu_complex64", + "test_variant_consistency_eager_nn_functional_scaled_dot_product_attention_xpu_float32", + "test_multiple_devices_linalg_multi_dot_xpu_int64", + "test_cow_input_linalg_multi_dot_xpu_float32", + "test_cow_input_nn_functional_scaled_dot_product_attention_xpu_float32", ), "test_binary_ufuncs_xpu.py": ( "test_fmod_remainder_by_zero_integral_xpu_int64", # zero division is an undefined behavior: different handles on different backends "test_div_rounding_numpy_xpu_float16", # Calculation error. XPU implementation uses opmath type. - # AssertionError: Jiterator is only supported on CUDA and ROCm GPUs, none are available. - "_jiterator_", ), "test_scatter_gather_ops_xpu.py": ( # AssertionError: Tensor-likes are not equal! @@ -469,11 +479,6 @@ "test_foreach_xpu.py": ( # RuntimeError: Tried to instantiate dummy base class CUDAGraph "use_cuda_graph_True", - # randomly fails - "test_parity__foreach_div_fastpath_inplace_xpu_complex128", - "test_parity__foreach_div_fastpath_outplace_xpu_complex128", - "test_parity__foreach_addcdiv_fastpath_inplace_xpu_complex128", - "test_parity__foreach_addcdiv_fastpath_outplace_xpu_complex128", ), "nn/test_convolution_xpu.py": ( # Summary: all of them are oneDNN related issues @@ -764,7 +769,6 @@ # AssertionError: Tensor-likes are not close! ; Exception: Tensor-likes are not close! "test_comprehensive_nn_functional_instance_norm_xpu_float64", # RuntimeError: Difference from float64 is larger with decomposition nll_loss_forward.default than original on output 0. - "test_comprehensive_nn_functional_nll_loss_xpu_float16", "test_comprehensive_nn_functional_pad_reflect_xpu_bfloat16", # NotImplementedError: Could not run 'aten::_flash_attention_forward' with arguments from the 'CPU' backend. "test_comprehensive_torch_ops_aten__flash_attention_forward_xpu_float16", @@ -792,14 +796,6 @@ # RuntimeError: I got this output for HasDecompTest.test_aten_core_operators: "test_aten_core_operators", "test_has_decomposition", - # AssertionError: Tensor-likes are not close! - "test_comprehensive_diff_xpu_complex128", - "test_comprehensive_ormqr_xpu_complex128", - "test_quick_var_mean_xpu_float64", - "test_comprehensive_diff_xpu_complex64", - "test_comprehensive_ormqr_xpu_complex64", - "test_quick_mean_xpu_complex128", - "test_comprehensive_grid_sampler_2d_xpu_bfloat16", ), "functorch/test_ops_xpu.py": None, "test_sparse_xpu.py": None, From 227137386cc983053e47d601fa3afa948133f171 Mon Sep 17 00:00:00 2001 From: kaileiyx Date: Mon, 24 Nov 2025 14:28:40 +0800 Subject: [PATCH 2/5] update --- test/xpu/skip_list_common.py | 4 ---- 1 file changed, 4 deletions(-) diff --git a/test/xpu/skip_list_common.py b/test/xpu/skip_list_common.py index ea8ed29342..e4cdfa66f2 100644 --- a/test/xpu/skip_list_common.py +++ b/test/xpu/skip_list_common.py @@ -22,10 +22,6 @@ "test_operator_histogramdd_xpu_float32", "test_view_replay_histogramdd_xpu_float32", "test_neg_view_histogramdd_xpu_float64", - "test_comprehensive_vdot_xpu_complex128", - "test_comprehensive_vdot_xpu_complex64", - "test_quick_vdot_xpu_complex128", - "test_quick_vdot_xpu_complex64", "test_variant_consistency_eager_nn_functional_scaled_dot_product_attention_xpu_float32", "test_multiple_devices_linalg_multi_dot_xpu_int64", "test_cow_input_linalg_multi_dot_xpu_float32", From eaf359e674c682f5d8936eb40fd72d76aaa025ec Mon Sep 17 00:00:00 2001 From: kaileiyx Date: Tue, 25 Nov 2025 09:58:29 +0800 Subject: [PATCH 3/5] update --- test/xpu/skip_list_common.py | 30 +++++++++++++++++++++++++----- 1 file changed, 25 insertions(+), 5 deletions(-) diff --git a/test/xpu/skip_list_common.py b/test/xpu/skip_list_common.py index e4cdfa66f2..1cc5ed3549 100644 --- a/test/xpu/skip_list_common.py +++ b/test/xpu/skip_list_common.py @@ -2,11 +2,7 @@ "test_ops_xpu.py": ( # Jiterator is only supported on CUDA and ROCm GPUs, none are available. # https://github.com/intel/torch-xpu-ops/issues/584 - "test_cow_input_jiterator_2inputs_2outputs_xpu_float32", - "test_cow_input_jiterator_4inputs_with_extra_args_xpu_float32", - "test_cow_input_jiterator_binary_return_by_ref_xpu_float32", - "test_cow_input_jiterator_binary_xpu_float32", - "test_cow_input_jiterator_unary_xpu_float32", + "_jiterator_" # OPs not supported "test_errors_dot_xpu", "test_errors_vdot_xpu", @@ -21,15 +17,39 @@ "test_cow_input_histogramdd_xpu_float32", "test_operator_histogramdd_xpu_float32", "test_view_replay_histogramdd_xpu_float32", + "test_noncontiguous_samples_histogramdd_xpu_float32", + "test_out_histogramdd_xpu_float32", + "test_variant_consistency_eager_histogramdd_xpu_float32", + "test_out_warning_histogramdd_xpu", "test_neg_view_histogramdd_xpu_float64", + "test_python_ref_torch_fallback__refs_vdot_xpu_complex128", + "test_python_ref_torch_fallback__refs_vdot_xpu_complex64", + "test_python_ref_executor__refs_vdot_executor_aten_xpu_complex128", + "test_python_ref_executor__refs_vdot_executor_aten_xpu_complex64", + "test_dtypes_vdot_xpu", "test_variant_consistency_eager_nn_functional_scaled_dot_product_attention_xpu_float32", "test_multiple_devices_linalg_multi_dot_xpu_int64", "test_cow_input_linalg_multi_dot_xpu_float32", "test_cow_input_nn_functional_scaled_dot_product_attention_xpu_float32", + "test_dtypes_dot_xpu", + "test_dtypes_linalg_multi_dot_xpu", + "test_dtypes_torch_ops_aten__efficient_attention_forward_xpu", + "test_fake_crossref_backward_no_amp_torch_ops_aten__efficient_attention_forward_xpu_float32", + "test_cow_input_torch_ops_aten__efficient_attention_forward_xpu_float32", + "test_out_warning_torch_ops_aten__efficient_attention_forward_xpu", + "test_variant_consistency_eager_torch_ops_aten__efficient_attention_forward_xpu_float32", + "test_fake_crossref_backward_amp_torch_ops_aten__efficient_attention_forward_xpu_float32", + "test_out_torch_ops_aten__efficient_attention_forward_xpu_float32", + "test_view_replay_torch_ops_aten__efficient_attention_forward_xpu_float32", + "test_out_torch_ops_aten__flash_attention_forward_xpu_float16", + "test_out_warning_torch_ops_aten__flash_attention_forward_xpu", + "test_dtypes_torch_ops_aten__flash_attention_forward_xpu", ), "test_binary_ufuncs_xpu.py": ( "test_fmod_remainder_by_zero_integral_xpu_int64", # zero division is an undefined behavior: different handles on different backends "test_div_rounding_numpy_xpu_float16", # Calculation error. XPU implementation uses opmath type. + # AssertionError: Jiterator is only supported on CUDA and ROCm GPUs, none are available. + "_jiterator_", ), "test_scatter_gather_ops_xpu.py": ( # AssertionError: Tensor-likes are not equal! From 7b27cc3437c74e95df08bc4aff90847e3fd288d3 Mon Sep 17 00:00:00 2001 From: kaileiyx Date: Tue, 25 Nov 2025 10:00:57 +0800 Subject: [PATCH 4/5] update --- test/xpu/skip_list_common.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/test/xpu/skip_list_common.py b/test/xpu/skip_list_common.py index 1cc5ed3549..58ab2813d3 100644 --- a/test/xpu/skip_list_common.py +++ b/test/xpu/skip_list_common.py @@ -2,7 +2,7 @@ "test_ops_xpu.py": ( # Jiterator is only supported on CUDA and ROCm GPUs, none are available. # https://github.com/intel/torch-xpu-ops/issues/584 - "_jiterator_" + "_jiterator_", # OPs not supported "test_errors_dot_xpu", "test_errors_vdot_xpu", From 14ef4bec3d92ef9c5c741a13e627101c4bf01c21 Mon Sep 17 00:00:00 2001 From: kaileiyx Date: Thu, 27 Nov 2025 15:06:31 +0800 Subject: [PATCH 5/5] remove 4 random failures --- .github/scripts/ut_result_check.sh | 4 ---- 1 file changed, 4 deletions(-) diff --git a/.github/scripts/ut_result_check.sh b/.github/scripts/ut_result_check.sh index a94007da11..baba794944 100644 --- a/.github/scripts/ut_result_check.sh +++ b/.github/scripts/ut_result_check.sh @@ -21,10 +21,6 @@ declare -A EXPECTED_CASES=( # Tests that are known to randomly pass and should be ignored when detecting new passes # These are typically flaky tests that don't indicate real improvements IGNORE_TESTS=( - "test_parity__foreach_div_fastpath_inplace_xpu_complex128" - "test_parity__foreach_div_fastpath_outplace_xpu_complex128" - "test_parity__foreach_addcdiv_fastpath_inplace_xpu_complex128" - "test_parity__foreach_addcdiv_fastpath_outplace_xpu_complex128" "test_python_ref__refs_log2_xpu_complex128" "_jiterator_" # Pattern to match all jiterator tests )