search

intel / llvm

Intel staging area for llvm.org contribution. Home for Intel LLVM-based projects.
Other
1.21k stars 728 forks source link

[IGC-DEV] hardware_dispatch.cpp failing with igc-dev from igc CI #14774

Open jsji opened 1 month ago

jsji commented 1 month ago

Describe the bug

sycl/test-e2e/ESIMD/hardware_dispatch.cpp is failing with dev-igc , so we XFAIL it first for now in https://github.com/intel/llvm/pull/14761 .

The issue is when we use igc-dev build, hardware_dispatch.cpp will fail. If we build the igc locally using the same commit, it will pass. And when they build the same commit in the release that we use in normal testing, it also passed. Looks like there are some differences in how igc is build in CI and built in release that is causing the issue.

To reproduce

remove the XFAIL and run test using igc_dev env.

Environment

Additional context

No response

jsji commented 1 month ago

FYI. @intel/dpcpp-esimd-reviewers

sarnex commented 1 month ago

@jsji My understanding is we believe the problem is in the way the igc-dev packages we use in CI are built and not a product issue. I asked the IGC team to investigate this before and they refused because they do not support the packages we are using. Not sure if there is anything for the ESIMD team to do here.

jsji commented 1 month ago

@jsji My understanding is we believe the problem is in the way the igc-dev packages we use in CI are built and not a product issue. I asked the IGC team to investigate this before and they refused because they do not support the packages we are using. Not sure if there is anything for the ESIMD team to do here.

Yes. Understand. So this is more of a FYI to ESIMD team that this test is XFAILed in igc-dev CI testing. I think this will be a limitation for near future. We may spend time working with IGC CI team to investigate later once we have more resource to do so.