Increase timeout for skipped_ut job and add first slow test case.

[BBBela]

Fix for: #3106

Overview

This PR increases timeout for skipped_ut job.
It also adds first very long test case to the skip_dict. This way it is not being run in the every pull workflow, but is launched in every Nightly workflow with increased timeout.

Problem description

The problem was that the test case:
op_ut,third_party.torch-xpu-ops.test.xpu.test_decomp_xpu.TestDecompXPU,test_quick_core_backward_baddbmm_xpu_float64
was failing in the CI because the timeout mechanism was stopping the worker before test case succeeded. Currently timeout for this test case was set to 10 minutes, while it takes much more than that (about 24 minutes) to succeed.

Changes

1. Increase timeout for skipped_ut job from 10 minutes to 60 minutes.
2. Add first discovered very long test case to the skip_dict.

[Copilot]

Pull request overview

This PR introduces a dedicated CI lane to run extremely slow XPU unit tests (e.g., >10 minutes) separately from the default op_ut run, so normal PR pipelines aren’t blocked by long-running cases.

Changes:

• Refactors test/xpu/skip_list_common.py to separate “tests to launch” (launch_list) from “tests to skip” (just_skip_dict), while keeping skip_dict in the legacy shape for compatibility.
• Adds slow_dict + launcher support (--test-cases slow) to run only very slow test cases in a dedicated job while excluding them from default selected runs.
• Updates CI workflows/actions to add an op_ut_long scope and include it in the nightly scheduled UT matrix with an increased pytest timeout.

Reviewed changes

Copilot reviewed 6 out of 6 changed files in this pull request and generated 5 comments.

Show a summary per file

File	Description
test/xpu/skip_list_common.py	Introduces launch_list, just_skip_dict, and slow_dict, and builds compatibility skip_dict + a combined exclusion dict.
test/xpu/run_test_with_skip.py	Adds slow selection mode and updates selected behavior to exclude slow tests via the combined exclusion dict.
.github/workflows/nightly_ondemand.yml	Adds op_ut_long to the scheduled UT matrix.
.github/workflows/_linux_ut.yml	Documents op_ut_long as a valid UT scope input.
.github/scripts/ut_result_check.sh	Adds op_ut_long to the suite dispatcher / available suites list.
.github/actions/linux-uttest/action.yml	Adds an op_ut_long step that runs --test-cases slow and increases pytest timeout for that step.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Pull request overview

Copilot reviewed 8 out of 8 changed files in this pull request and generated 3 comments.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Pull request overview

Copilot reviewed 8 out of 8 changed files in this pull request and generated 2 comments.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[BBBela]

Hello @mengfei25, @RUIJIEZHONG66166,
could you please look at this PR and review when you have time? Thanks!
That is just a proposal to resolve problem described in the linked issue #3106

[Copilot]

Pull request overview

Copilot reviewed 2 out of 2 changed files in this pull request and generated 2 comments.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[mengfei25]

LGTM, and please launch a test to check whether your changes works or not

[BBBela]

I confirm that in the On-demand workflow from this PR the mentioned test case is being launched and is passing:
https://github.com/intel/torch-xpu-ops/actions/runs/24334802146/job/71235318294#step:4:87

[BBBela]

I also confirm that in this PRs pull workflow, the test case
test_quick_core_backward_baddbmm_xpu_float64
is being correctly skipped and not run.
So to sum up - the behavior is correct and as expected. The added test case is skipped in pull workflow but launched in the Nightly skipped_ut job with increased timeout and it pass.

[BBBela]

/merge

[chuanqi129]

✅ PR has been successfully merged by @BBBela.