intel / llvm

Intel staging area for llvm.org contribution. Home for Intel LLVM-based projects.
Other
1.23k stars 734 forks source link

Tests disabled during Intel Arc GPU CI enabling #11434

Closed aelovikov-intel closed 3 weeks ago

aelovikov-intel commented 1 year ago

Refer to the https://github.com/intel/llvm/pull/11414 for the list of disabled tests.

Preliminary CI runs that drove selection of such tests: https://github.com/intel/llvm/actions/runs/6411226892/job/17406603528?pr=11414 https://github.com/intel/llvm/actions/runs/6412058021/job/17409025370?pr=11414

As UNSUPPORTED:
********************
Timed Out Tests (7):
  SYCL :: Graph/Explicit/buffer_copy_target2host.cpp
  SYCL :: Graph/Explicit/buffer_copy_target2host_2d.cpp
  SYCL :: Graph/Explicit/buffer_copy_target2host_offset.cpp
  SYCL :: Graph/RecordReplay/buffer_copy_target2host.cpp
  SYCL :: Graph/RecordReplay/buffer_copy_target2host_2d.cpp
  SYCL :: Graph/RecordReplay/buffer_copy_target2host_offset.cpp
  SYCL :: Plugin/enqueue-arg-order-image.cpp

Also, as UNSUPPORTED ESIMD/slm_block_load_store.cpp, but this one is being tracked in https://github.com/intel/llvm/issues/11358.

As XFAIL:
********************
Failed Tests (26):
  SYCL :: Basic/accessor/empty_zero_dim_accessor.cpp
  SYCL :: DeviceLib/string_test.cpp
  SYCL :: Graph/Explicit/basic_buffer.cpp
  SYCL :: Graph/Explicit/buffer_copy.cpp
  SYCL :: Graph/Explicit/buffer_copy_2d.cpp
  SYCL :: Graph/Explicit/buffer_copy_host2target.cpp
  SYCL :: Graph/Explicit/buffer_copy_host2target_2d.cpp
  SYCL :: Graph/Explicit/buffer_copy_host2target_offset.cpp
  SYCL :: Graph/Explicit/buffer_copy_offsets.cpp
  SYCL :: Graph/Explicit/event_status_querying.cpp
  SYCL :: Graph/Explicit/temp_buffer_reinterpret.cpp
  SYCL :: Graph/Explicit/usm_copy.cpp
  SYCL :: Graph/RecordReplay/basic_buffer.cpp
  SYCL :: Graph/RecordReplay/buffer_copy.cpp
  SYCL :: Graph/RecordReplay/buffer_copy_2d.cpp
  SYCL :: Graph/RecordReplay/buffer_copy_host2target.cpp
  SYCL :: Graph/RecordReplay/buffer_copy_host2target_2d.cpp
  SYCL :: Graph/RecordReplay/buffer_copy_host2target_offset.cpp
  SYCL :: Graph/RecordReplay/buffer_copy_offsets.cpp
  SYCL :: Graph/RecordReplay/event_status_querying.cpp
  SYCL :: Graph/RecordReplay/temp_buffer_reinterpret.cpp
  SYCL :: Graph/RecordReplay/usm_copy.cpp
  SYCL :: Graph/RecordReplay/usm_copy_in_order.cpp
  SYCL :: GroupAlgorithm/SYCL2020/sort.cpp
  SYCL :: Plugin/level_zero_usm_residency.cpp
  SYCL :: USM/usm_leak_check.cpp
Bensuo commented 1 year ago

There is a fix in https://github.com/intel/llvm/pull/11133 which may resolve some of these errors, revolving around incorrect scheduler dependencies for graph commands on devices without unified memory. Many of the errors do not look related to that problem though (which was difficult to replicate in the first place), so I'm not 100% sure it would fix in this case.

steffenlarsen commented 1 year ago

sycl/test-e2e/GroupAlgorithm/SYCL2020/sort.cpp was fixed with https://github.com/intel/llvm/pull/11442 (https://github.com/intel/llvm/pull/11534 enables it.)

steffenlarsen commented 10 months ago

DeviceLib/string_test.cpp seems to suddenly be passing in post-commit. Due to temporary downtime on post-commit builds it is difficult to say which PR made the difference, though it was likely https://github.com/intel/llvm/pull/12101. https://github.com/intel/llvm/pull/12149 reenables it.

sarnex commented 10 months ago

I reenabled string_test.cpp in https://github.com/intel/llvm/commit/abed3463dd0a8ae812b54bdc054414433c62119f, me and Steffen made the same PR but never found each other :)

aelovikov-intel commented 5 months ago

https://github.com/search?q=repo%3Aintel%2Fllvm%20https%3A%2F%2Fgithub.com%2Fintel%2Fllvm%2Fissues%2F11434&type=code shows the only test that remains disabled is https://github.com/intel/llvm/blob/sycl/sycl/test-e2e/Plugin/enqueue-arg-order-image.cpp

aelovikov-intel commented 4 months ago

Still fails with a timeout: https://github.com/intel/llvm/actions/runs/9521663127/job/26249596457. There is an internal bug report against underlying GPU RT.