[CI] Run L0 v2 tests on PVC

[igchor]

No description provided.

[igchor]

@intel/dpcpp-devops-reviewers @intel/sycl-matrix-reviewers could you please take a look? Enabling v2 testing doesn't seem to increase testing time at all (probably the bottleneck is just in dispatching the tests).

[uditagarwal97]

@intel/dpcpp-devops-reviewers @intel/sycl-matrix-reviewers could you please take a look? Enabling v2 testing doesn't seem to increase testing time at all (probably the bottleneck is just in dispatching the tests).

I think there is an overhead:
E2E on PVC for this PR: 7m 18sec (https://github.com/intel/llvm/actions/runs/15597044541/job/44073154508?pr=18909)
E2E on PVC for last Nightly run (same PVC runner): 3m 38sec (https://github.com/intel/llvm/actions/runs/15625712272/job/44019756304)

Nevertheless, I think that we have enough PVC machines to accommodate L0 V2 testing, without it becoming the bottleneck.
Edit:- I'll wait for @sarnex to share his thoughts before approving the PR.

[igchor]

@intel/dpcpp-devops-reviewers @intel/sycl-matrix-reviewers could you please take a look? Enabling v2 testing doesn't seem to increase testing time at all (probably the bottleneck is just in dispatching the tests).

I think there is an overhead: E2E on PVC for this PR: 7m 18sec (https://github.com/intel/llvm/actions/runs/15597044541/job/44073154508?pr=18909) E2E on PVC for last Nightly run (same PVC runner): 3m 38sec (https://github.com/intel/llvm/actions/runs/15625712272/job/44019756304)

Nevertheless, I think that we have enough PVC machines to accommodate L0 V2 testing, without it becoming the bottleneck. Edit:- I'll wait for @sarnex to share his thoughts before approving the PR.

Hm, that's interesting. I looked at some other PRs and it seems the time varies quite a lot. For example, here https://github.com/intel/llvm/actions/runs/15632385788/job/44041538881?pr=18970 the Run E2E Tests takes almost exactly the same as on this PR

[uditagarwal97]

@intel/dpcpp-devops-reviewers @intel/sycl-matrix-reviewers could you please take a look? Enabling v2 testing doesn't seem to increase testing time at all (probably the bottleneck is just in dispatching the tests).

I think there is an overhead: E2E on PVC for this PR: 7m 18sec (https://github.com/intel/llvm/actions/runs/15597044541/job/44073154508?pr=18909) E2E on PVC for last Nightly run (same PVC runner): 3m 38sec (https://github.com/intel/llvm/actions/runs/15625712272/job/44019756304)
Nevertheless, I think that we have enough PVC machines to accommodate L0 V2 testing, without it becoming the bottleneck. Edit:- I'll wait for @sarnex to share his thoughts before approving the PR.

Hm, that's interesting. I looked at some other PRs and it seems the time varies quite a lot. For example, here https://github.com/intel/llvm/actions/runs/15632385788/job/44041538881?pr=18970 the Run E2E Tests takes almost exactly the same as on this PR

True, there's quite a significant variation between PVC E2E test runs among different runners, probably because of the differences in compute resources.
https://github.com/intel/llvm/actions/runs/15632385788/job/44041538881?pr=18970 uses a different runner then the CI run of this PR, so, not an apple-to-apple comparison 🍎.

[sarnex]

Sorry for the late response. I think we have a bit of room left on PVC GPU testing, so I'm willing to try it. @igchor when we merge this please make a note to check pvc e2e runtimes for a few days/a week after to make sure there is no huge slowdown. Thanks