Skip to content

Conversation

@maksimsab
Copy link
Contributor

@maksimsab maksimsab commented Nov 13, 2025

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.

@YuriPlyakhin
Copy link
Contributor

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

@maksimsab
Copy link
Contributor Author

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.

Copy link
Contributor

@sarnex sarnex left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

lgtm overall, minor comments. i wish we didn't have so much copy and paste but i dont have a better idea that's reasonable to implement

//===----------------------------------------------------------------------===//

// XFAIL: new-offload-model
// XFAIL-TRACKER: CMPLRLLVM-61170
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

For issues in this project public GH issue should be used. Internal bug trackers are only meaningful for issues in unrelated components because having GH issue here would provide no benefit.

Comment on lines +345 to +362
- name: Remove E2E tests before NewOffloadModel run
if: ${{ inputs.e2e_binaries_new_offload_model_artifact && !cancelled() && steps.build.conclusion == 'success' }}
run: rm -rf build-e2e

- name: Build E2E tests with NewOffloadModel
if: |
inputs.e2e_binaries_new_offload_model_artifact &&
!cancelled()
&& steps.build.conclusion == 'success' &&
contains(github.event.pull_request.labels.*.name, 'new-offload-model')
uses: ./devops/actions/run-tests/e2e
with:
ref: ${{ inputs.ref || github.sha }}
testing_mode: build-only
target_devices: all
binaries_artifact: ${{ inputs.e2e_binaries_new_offload_model_artifact }}
sycl_compiler: $GITHUB_WORKSPACE/toolchain/bin/clang++
extra_lit_opts: --param sycl_build_targets="spir;nvidia;amd" --param enable_new_offload_model=True
Copy link
Contributor

@aelovikov-intel aelovikov-intel Nov 26, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This is getting too verbose. Can we fold the removal into the devops/actions/run-tests/e2e and then use a matrix job to build all the flavors without that much copy-paste? That prerequisite refactoring should be done in a separate PR, obviously.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

refactoring is posted here: #20790

packages: read
strategy:
fail-fast: false
matrix:
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Isn't this a bit excessive? Do you really see value in running on all configurations?

Copy link
Contributor Author

@maksimsab maksimsab Dec 1, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

That could be shrinked to intel only targets.

target_devices: level_zero:gpu
extra_lit_opts: --param test-preview-mode=True

- name: AMD/HIP with NewOffloadModel
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

done.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

new-offload-model Enables testing with NewOffloadModel.

Projects

None yet

Development

Successfully merging this pull request may close these issues.

5 participants