[CI][SYCL] Enable E2E testing for New Offloading Model

[maksimsab]

This patch has the following changes:

• --param enable_new_offload_model is added to SYCL E2E LIT test configuration
• NewOffloadModel mode run is added to sycl-linux-precommit
• NewOffloadModel mode run is added to sycl-nightly for linux and windows.

[maksimsab]

Hi @sarnex

Could you please help with reviewing this change or recommend somebody for that?

[sarnex]

I can review, thanks

[sarnex]

Also you might want to rebase this PR against HEAD if you're going to run the nightly, I merged a fix today that should fix the AOT failures, they're unrelated to this PR.

[maksimsab]

Update: I have about 500 uniquely tests failing in a separate branch in sycl-nightly run here: https://github.com/intel/llvm/actions/runs/19550681758

I am going to mark them with XFAIL or UNSUPPORTED directive in this patch.

[YuriPlyakhin]

@maksimsab , there are still some tests failing. Should they also be marked with XFAIL?

[maksimsab]

Update:

• testing enabling depending on label is done.
• SYCL :: AOT/multiple-devices.cpp is failing for linux GEN 12.
• SYCL :: Basic/out_of_order_queue_status_ext_oneapi_empty.cpp is failing for linux GEN 12. Looks like a flaky test for now.
• I still need to try to xfail some tests instead of unsupporting them.

[maksimsab]

Hi @aelovikov-intel @KornevNikita

How does it look to you now?

I have a final run of sycl-nightly here. https://github.com/intel/llvm/actions/runs/19935023822 I hope it doesn't find any failing tests.

[maksimsab]

Hi @YuriPlyakhin @aelovikov-intel

I've edited target_devices so that it tests opencl:cpu only once.

Last sycl-nightly run: https://github.com/intel/llvm/actions/runs/20067784201

I've marked 2 tests Regression/reduction_resource_leak_dw.cpp and KernelCompiler/sycl_join.cpp with UNSUPPORTED directive because I've observed flaky results for them. For reduction_resource_leak_dw.cpp: It was failing here https://github.com/intel/llvm/actions/runs/20034022855 Then being marked XFAIL it became Unexpectedly Passed. https://github.com/intel/llvm/actions/runs/20070023567

For sycl_join.cpp: It appears to be failing only on Windows Arc configuration. I can leave it as UNSUPPORTED until it investigated or try to XFAIL it by target. In sycl-nightly it is also failing in Old Offload Model.

Do you mind if we merge this PR?

[maksimsab]

So, from the latest run (Link) I see that 13 ESIMD tests report Unexpectedly Passed in run mode and 503 ESIMD tests report Unexpectedly Passed in build mode. I think, that marking them unsupported is the easiest way for now.

Also, from this run (Link) some InvokeSimd tests are also unexpectedly passed in both modes.

[sarnex]

The below three failures are unrelated to this PR, are failing in the nightly already, and will be fixed soon.

  SYCL :: Matrix/joint_matrix_bf16_fill_k_cache_arg_dim.cpp
  SYCL :: Matrix/joint_matrix_bf16_fill_k_cache_runtime_dim.cpp
  SYCL :: SubGroup/reduce.cpp

[sarnex]

Just merged the fix, so you might want to rebase on top of HEAD

[maksimsab]

@intel/llvm-gatekeepers Can we merge that please?

CI failures are unrelated.

[maksimsab]

Merged latest sycl branch to see whether the issue is resolved.

[sarnex]

I'd like to wait until CI runs again, I'll merge it if I don't see anything that seems related to this PR.

[sarnex]

@maksimsab The code formatting CI job failed and the failure looks related to this PR. Can you take a look?

[sarnex]

The failures don't seem related