Fix cudagraph tests blocked by feature gap in xpugraph

[Copilot]

Multiple test cases across test_backends.py, test_cudagraphs.py, test_cudagraphs_expandable_segments.py, test_structured_trace.py, test_foreach.py, and test_cuda_repro.py are failing due to the XPUGraph feature gap in torch-xpu-ops. This PR skips the affected tests and adds XPU dynamo test wrappers with proper skip annotations.

Changes Made

• test/xpu/skip_list_common.py: Updated test_foreach_xpu.py skip pattern from use_cuda_graph_True to use_xpu_graph_True (the parametrize name was renamed in the XPU test wrapper). Also added entries for 4 new dynamo XPU test files with cudagraph-related test skips.

• test/xpu/dynamo/test_backends_xpu.py (new): XPU wrapper for test/dynamo/test_backends.py that skips test_aot_cudagraphs since XPUGraph is not fully supported.

• test/xpu/dynamo/test_cudagraphs_xpu.py (new): XPU wrapper that skips all tests from test_cudagraphs.py (all depend on XPUGraph).

• test/xpu/dynamo/test_cudagraphs_expandable_segments_xpu.py (new): XPU wrapper that skips all tests for the same reason.

• test/xpu/dynamo/test_structured_trace_xpu.py (new): Full XPU-adapted version of test_structured_trace.py (based on upstream PR [XPU] Migrate 11 dynamo test cases for XPU pytorch/pytorch#169241) with test_gpugraphs (the XPU equivalent of test_cudagraphs) explicitly skipped due to XPUGraph feature gap.

Notes

• test/inductor/test_cuda_repro.py::test_cpu_index already has @skipIfXpu(msg="cudagraph is not supported on xpu") in the upstream PyTorch, so no additional change is needed for that test.

Test

In PyTorch repo (run via op_ut):

• test/xpu/dynamo/test_backends_xpu.py
• test/xpu/dynamo/test_cudagraphs_xpu.py
• test/xpu/dynamo/test_cudagraphs_expandable_segments_xpu.py
• test/xpu/dynamo/test_structured_trace_xpu.py
• test/xpu/test_foreach_xpu.py

[laifenxiawucha]

@copilot continue fix

[Copilot]

@copilot continue fix

Implemented all the fixes in commit ceae918:

1. skip_list_common.py: Updated test_foreach_xpu.py skip pattern from use_cuda_graph_True → use_xpu_graph_True (parameter was renamed in the test file)

2. test/xpu/dynamo/test_backends_xpu.py: XPU wrapper that skips test_aot_cudagraphs (requires XPUGraph, which is not fully supported)

3. test/xpu/dynamo/test_cudagraphs_xpu.py: XPU wrapper with all tests skipped (all require XPUGraph)

4. test/xpu/dynamo/test_cudagraphs_expandable_segments_xpu.py: XPU wrapper with all tests skipped

5. test/xpu/dynamo/test_structured_trace_xpu.py: Full XPU-adapted version of the upstream structured trace tests (from PR Migrate XPU UT files from Upstream PR #3593), with test_gpugraphs explicitly skipped due to XPUGraph feature gap

6. skip_list_common.py: Added all 4 new dynamo test files with appropriate skip patterns for cudagraph tests

Note: test/inductor/test_cuda_repro.py::test_cpu_index already has @skipIfXpu(msg="cudagraph is not supported on xpu") in the upstream PyTorch, so no additional change is needed.

[github-actions]

Performance outliers, please check!

• 🔴 [-1, 80%), should be regression

Category	Model	Target vs. Baseline [Eager]	Target vs. Baseline [Inductor]
timm_models_bfloat16_training	mobilenetv3_large_100	0.783104	0.738373
timm_models_bfloat16_training	vit_base_patch16_siglip_256	0.693589	0.741715
timm_models_bfloat16_training	deit_base_distilled_patch16_224	0.739197	0.752590
timm_models_bfloat16_training	nfnet_l0	0.730817	0.758981
timm_models_bfloat16_training	visformer_small	0.721568	0.759062
timm_models_bfloat16_training	dm_nfnet_f0	0.637236	0.776720
timm_models_bfloat16_training	tf_efficientnet_b0	0.808093	0.781045
timm_models_bfloat16_training	adv_inception_v3	0.787141	0.784371
timm_models_bfloat16_training	mobilenetv2_100	0.832888	0.786561
timm_models_bfloat16_training	ghostnet_100	0.861574	0.792291
timm_models_bfloat16_training	mobilevit_s	0.715427	0.799490
timm_models_bfloat16_training	beit_base_patch16_224	0.727340	0.804644

• 🟡 [80%, 90%), may be fluctuations

Category	Model	Target vs. Baseline [Eager]	Target vs. Baseline [Inductor]
timm_models_bfloat16_training	convnextv2_nano.fcmae_ft_in22k_in1k	0.805841	0.806135
timm_models_bfloat16_training	swin_base_patch4_window7_224	0.852855	0.819549
timm_models_bfloat16_training	inception_v3	0.841121	0.825677
torchbench_bfloat16_training	mnasnet1_0	1.046836	0.847807
timm_models_bfloat16_training	repvgg_a2	0.844693	0.867251
timm_models_bfloat16_training	deit_tiny_patch16_224.fb_in1k	0.878619	0.883622

[liangan1]

The related issue is just used to track the task progress, No need to fix.
@ZhaoqiongZ

1. need to add task tracking label in the issue.
2. add skills to skip the related issue.