[CI][NewPM] Turn on nightly testing with the new PM set by default

[mlychkov]

Signed-off-by: Mikhail Lychkov mikhail.lychkov@intel.com

[alexbatashev]

You don't need to change main workflow file. It can be customized from outside.

[mlychkov]

You don't need to change main workflow file. It can be customized from outside.

This is great. Thanks a lot.

[bader]

@mlychkov, one of the tests hangs(?) with new PM. Please, investigate the root cause.

[mlychkov]

@mlychkov, one of the tests hangs(?) with new PM. Please, investigate the root cause.

It has been found that new PM option didn't set properly and clang was still built with legacy PM for new tests. So, the test hung on compiler with legacy PM. I'll create an issue.

[mlychkov]

Failed 'SYCL Nightly Builds / ubuntu2004_build_test_new_pm' is expected because not all patches that fix SYCL optimization pipeline using new PM have been merged yet.

[bader]

Let's fix these tests before merging this change.

[mlychkov]

Failed tests are a new regression since last testing. I will create a new issue for that.

[olegmaslovatintel]

@mlychkov could you please tell number of failing tests?

[mlychkov]

@mlychkov could you please tell number of failing tests?

Failed Tests (5):
SYCL :: check_device_code/matrix/matrix-nvptx-bf16-test.cpp
SYCL :: check_device_code/matrix/matrix-nvptx-half-float-test.cpp
SYCL :: check_device_code/matrix/matrix-nvptx-half-half-test.cpp
SYCL :: check_device_code/matrix/matrix-nvptx-int8-test.cpp
SYCL :: check_device_code/matrix/matrix-nvptx-uint8-test.cpp

These tests have been added in 5373362

Issue #5473 is raised.

[mlychkov]

Let's fix these tests before merging this change.

@bader All initial fixes have been merged into main branch. There is a new regression for new PM, but I think that it'd be better to merge this PR and just start to create issues for new regressions. Otherwise, another regression may occur while we fix current one and so on.

[bader]

I would prefer the job to pass. Is it possible to mark failing tests as "expected to fail with new pass manager"?

[JackAKirk]

I would prefer the job to pass. Is it possible to mark failing tests as "expected to fail with new pass manager"?

I've summarized what is different for the legacy pass manager case and temporarily disabled the tests here : #5475

I'll try to understand why the the variable names change for the different pass managers.

[mlychkov]

This time tests failed again due to test host connectivity issues.

[mlychkov]

@bader Restarted pre-commit has passed: http://llvm-ci-test2.intel.com:8080/blue/organizations/jenkins/SYCL_CI%2Fintel%2FLin%2FBasic_E2E/detail/Basic_E2E/761/pipeline/