search

intel / llvm

Intel staging area for llvm.org contribution. Home for Intel LLVM-based projects.
Other
1.26k stars 739 forks source link

[SYCL][CUDA]Massive runfail for CUDA SYCL CTS on unrelated commit #8556

Open dm-vodopyanov opened 1 year ago

dm-vodopyanov commented 1 year ago

Post-commit failed on unrelated patch for ESIMD https://github.com/intel/llvm/commit/d2bc9d10a50acc317e727b9b6c06a44646ed9a18 (@v-klochkov please confirm that this is unrelated) - CUDA SYCL CTS run failed. This could be due to changes in https://github.com/KhronosGroup/SYCL-CTS.

Logs: https://github.com/intel/llvm/actions/runs/4348365190/jobs/7597934088

dm-vodopyanov commented 1 year ago

Looks like culprit is https://github.com/KhronosGroup/SYCL-CTS/commit/df2259b2bc17cc724920d97e19beeb09bd86aed0

dm-vodopyanov commented 1 year ago

@vladimirkhashev FYI.

bader commented 1 year ago

Looks like culprit is KhronosGroup/SYCL-CTS@df2259b

@dm-vodopyanov, I suggest disabling new test to fix post-commit. Please, keep this issue open to address the failures and re-enable the test.

dm-vodopyanov commented 1 year ago

Created https://github.com/intel/llvm/pull/8575 to disable the tests

JackAKirk commented 1 month ago

full atomic_ref functionality requires building with sm_70 specified using nvptx target flag. We only support the full ptx 6.0 memory model (since https://github.com/intel/llvm/pull/12516) and this requires sm_70 and we have confirmed that the sycl CTS has pull pass rate for atomic tests for this case. Earlier versions have an unspecified memory model that we have not fully implemented.