[SYCL][Driver] Add off value for device code split option

[Fznamznon]

Signed-off-by: Mariya Podchishchaeva [email protected]

[Fznamznon]

Sorry, I forgot the test for #853 . My bad :(

[Fznamznon]

Does anybody know why all the sycl driver tests are xfailed on windows? Even windows-specific test.

llvm/clang/test/Driver/sycl-offload-win.c

Line 101 in a339d4c

// XFAIL: windows-msvc

My test doesn’t seem failing with split-specific problem.

[AGindinson]

Does anybody know why all the sycl driver tests are xfailed on windows? Even windows-specific test.

llvm/clang/test/Driver/sycl-offload-win.c

Line 101 in a339d4c

// XFAIL: windows-msvc

My test doesn’t seem failing with split-specific problem.

I can't analyze this properly now, but it seems related to some Windows-unfriendly path regexp.

Definitely not caused by your patch, but I'd like some input from @mdtoguchi or @v-klochkov before we blissfully add the same TODO + XFAIL and commit the test.

[Fznamznon]

Definitely not caused by your patch, but I'd like some input from @mdtoguchi or @v-klochkov before we blissfully add the same TODO + XFAIL and commit the test.

I wanted to check that all tools is called properly when device code split is requested. So I took existing test cases and extended these test cases for device code split case. The problem happens in lines not related to split, so another option is reduce test cases to make them check only split-affected jobs.

[AGindinson]

The problem happens in lines not related to split, so another option is reduce test cases to make them check only split-affected jobs.

Given the feature complexity, I'd personally steer away from reducing coverage. The least risky option seems to be XFAIL: windows-msvc, and a later follow-up with either making the test "split-only" or analyzing + fixing the Windows problem, depending on others' opinions.

[Fznamznon]

The problem happens in lines not related to split, so another option is reduce test cases to make them check only split-affected jobs.

Given the feature complexity, I'd personally steer away from reducing coverage. The least risky option seems to be XFAIL: windows-msvc, and a later follow-up with either making the test "split-only" or analyzing + fixing the Windows problem, depending on others' opinions.

How about add XFAIL and create a GitHub issue?

[AGindinson]

How about add XFAIL and create a GitHub issue?

That's a good idea, I'll create one.

[Fznamznon]

How about add XFAIL and create a GitHub issue?

That's a good idea, I'll create one.

Okay then, I XFAILed the test.

[bader]

LGTM. Thanks!