[BE] Upgrade oneAPI XPU bundle to 2026.0.0

[mengfei25]

Summary

• upgrade Linux XPU package pins to the 2026.0.0 support stack
• upgrade Windows build package pins and installer defaults to 2026.0.0
• add Linux oneAPI 2026.0.0 offline installer setup in the build workflow
• pin Windows mkl-static and mkl-include to 2026.0.0

Dependency

• depends on [BE] Upgrade XPU support package to 2026.0 pytorch/pytorch#182003

Validation

• checked workflow/script diagnostics in the workspace
• validated edited workflow files report no errors

[Copilot]

Pull request overview

Skill files read: .github/skills/xpu-ops-pr-review/SKILL.md.

This PR updates the CI build/test environment to target Intel oneAPI/XPU 2026.0.0 across Linux and Windows, aligning the repository’s build configuration with the new support stack.

Changes:

• Bump Windows oneAPI-related package pins (and MKL static/include) to 2026.0.0 in the Windows build workflow.
• Add a Linux CI step to install the oneAPI 2026.0.0 offline installer and export XPU_ONEAPI_PATH.
• Update the Windows oneAPI installer script to support selecting 2026.0 via XPU_VERSION.

Reviewed changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated 4 comments.

Show a summary per file

File	Description
.github/workflows/_windows_ut.yml	Removes oneAPI install/pin logic from UT workflow, shifting reliance to runner provisioning.
.github/workflows/_windows_build.yml	Updates Windows build pins/defaults to oneAPI 2026.0 and pins mkl-static/mkl-include to 2026.0.0.
.github/workflows/_linux_build.yml	Adds oneAPI 2026.0.0 offline installer setup and exports XPU_ONEAPI_PATH.
.github/scripts/install_xpu.bat	Updates installer selection logic to allow 2026.0 when XPU_VERSION=2026.0.
.github/scripts/build.sh	Updates Linux-side pinned package versions to 2026.0.0 (and related components).

[Copilot]

Pull request overview

Copilot reviewed 5 out of 5 changed files in this pull request and generated 3 comments.

[chuanqi129]

@copilot New UT failures detected in op_ut tests (10 total). Please check these new failures and analyze whether they are caused by the PR changes.

Important: Do NOT update the plan in the PR description directly. Use reply comments to update the status.

Job log: https://github.com/intel/torch-xpu-ops/actions/runs/26199758244
Full new failure report: https://github.com/intel/torch-xpu-ops/actions/runs/26199758244/artifacts/7127914720

New failures:

Category	Class name	Test name	Status	Message
op_ut	third_party.torch-xpu-ops.test.xpu.dynamo.test_functions_xpu.DefaultsTests	test_zip_strict	failed	torch._dynamo.exc.Unsupported: Observed exception
op_ut	third_party.torch-xpu-ops.test.xpu.functorch.test_control_flow_xpu.TestControlFlowTraced	test_while_loop_autograd_simple	failed	AssertionError: 'clas[152 chars] zeros: "i64[]" = torch.ops.aten.zeros.def[3514 chars]1)\n' != 'clas[152 chars] clone: "f32[3, 3]" = torch.ops.aten.clone[3636 chars]1)\n'
op_ut		third_party.torch-xpu-ops.test.xpu.functorch.test_ops_xpu	error	ImportError: cannot import name 'skipOps' from 'common_utils' (/__w/torch-xpu-ops/torch-xpu-ops/pytorch/third_party/torch-xpu-ops/test/xpu/../../../../test/functorch/common_utils.py)
op_ut		third_party.torch-xpu-ops.test.xpu.functorch.test_vmap_xpu	error	ImportError: cannot import name 'skipOps' from 'common_utils' (/__w/torch-xpu-ops/torch-xpu-ops/pytorch/third_party/torch-xpu-ops/test/xpu/../../../../test/functorch/common_utils.py)
op_ut	third_party.torch-xpu-ops.test.xpu.quantization.core.test_quantized_tensor_xpu.TestQuantizedTensorXPU	test_qtensor_sub_byte_aligned_cols_xpu	failed	RuntimeError: setStorage: sizes [1, 4], strides [4, 1], storage offset 0, and itemsize 1 requiring a storage size of 4 are out of bounds for storage of size 2

... and 5 more failures. See full list in artifact.

Instructions:

1. Download the full new failure report artifact for complete error messages
2. Analyze whether the failures are caused by PR changes
3. If caused by PR changes, fix those failures on this PR branch directly
4. Add failing test cases with pytest commands in the PR description for the reproducer job to use
5. Use reply comments to report analysis status - do NOT modify the PR description plan directly

[Copilot]

@copilot New UT failures detected in op_ut tests (10 total). Please check these new failures and analyze whether they are caused by the PR changes.

**Important: Do NOT update the plan in the PR d...

Analyzed run 26199758244 and downloaded both artifacts (7127914720, 7127915439) on commit 6b5d8ae.

Conclusion: these 10 new op_ut failures are not caused by this PR’s oneAPI/workflow pin changes. They are test/framework drift issues from current PyTorch test/API state (e.g. missing skipOps/skip imports, frozen dataclass config mutation, Dynamo behavior change for zip(strict=True), XPU device-properties attr mismatch, quantized load/save storage shape mismatch).

Reproducer commands for the reproducer job:

• PYTORCH_TEST_WITH_SLOW=1 python test/xpu/dynamo/test_compiler_bisector_xpu.py TestCompilerBisector.test_pre_grad
• PYTORCH_TEST_WITH_SLOW=1 python test/xpu/dynamo/test_functions_xpu.py FunctionTests.test_get_device_properties_tensor_device
• PYTORCH_TEST_WITH_SLOW=1 python test/xpu/dynamo/test_functions_xpu.py DefaultsTests.test_zip_strict
• PYTORCH_TEST_WITH_SLOW=1 python test/xpu/functorch/test_control_flow_xpu.py TestControlFlowTraced.test_while_loop_autograd_simple
• PYTORCH_TEST_WITH_SLOW=1 python test/xpu/functorch/test_ops_xpu.py
• PYTORCH_TEST_WITH_SLOW=1 python test/xpu/functorch/test_vmap_xpu.py
• PYTORCH_TEST_WITH_SLOW=1 python test/xpu/quantization/core/test_quantized_tensor_xpu.py TestQuantizedTensorXPU.test_qtensor_sub_byte_aligned_cols_xpu
• PYTORCH_TEST_WITH_SLOW=1 python test/xpu/quantization/core/test_quantized_tensor_xpu.py TestQuantizedTensorXPU.test_qtensor_sub_byte_not_aligned_cols_xpu
• PYTORCH_TEST_WITH_SLOW=1 python test/xpu/quantization/core/test_quantized_tensor_xpu.py TestQuantizedTensorXPU.test_qtensor_per_channel_load_save_xpu
• PYTORCH_TEST_WITH_SLOW=1 python test/xpu/dynamo/test_aot_autograd_cache_xpu.py AOTAutogradCachePicklerTests.test_identical_graphs_and_configs
• PYTORCH_TEST_WITH_SLOW=1 python test/xpu/dynamo/test_aot_autograd_cache_xpu.py AOTAutogradCachePicklerTests.test_different_configs
• PYTORCH_TEST_WITH_SLOW=1 python test/xpu/test_decomp_xpu.py

I did not apply fixes in this PR branch since root causes are outside the oneAPI upgrade scope.

[chuanqi129]

@copilot New UT failures detected in xpu_distributed tests (0 total). Please check these new failures and analyze whether they are caused by the PR changes.

Important: Do NOT update the plan in the PR description directly. Use reply comments to update the status.

Job log: https://github.com/intel/torch-xpu-ops/actions/runs/26199758244
Full new failure report: https://github.com/intel/torch-xpu-ops/actions/runs/26199758244/artifacts/7130455401

New failures:

Instructions:

1. Download the full new failure report artifact for complete error messages
2. Analyze whether the failures are caused by PR changes
3. If caused by PR changes, fix those failures on this PR branch directly
4. Add failing test cases with pytest commands in the PR description for the reproducer job to use
5. Use reply comments to report analysis status - do NOT modify the PR description plan directly