kokkos / kokkos

Kokkos C++ Performance Portability Programming Ecosystem: The Programming Model - Parallel Execution and Memory Abstraction
https://kokkos.org
Other
2.02k stars 437 forks source link

Sycl unit failures with intel/2024.1.0 on PVC arch #7060

Open ndellingwood opened 5 months ago

ndellingwood commented 5 months ago

Describe the bug

We recently had intel oneapi 2024.1.0 installed on our machine with PVC GPUs, testing of the Sycl backend reported a couple failures:

Kokkos_CoreUnitTest_SYCL1B

 4/53 Test  #4: Kokkos_CoreUnitTest_SYCL1B .......................***Failed  Error regular expression found in output. Regex=[  FAILED  ]  4.63 sec
...
[ RUN      ] sycl.reducers_int8_t
/home/ndellin/kokkos/core/unit_test/TestReducers.hpp:494: Failure
Expected equality of these values:
  sum_scalar
    Which is: '\0'
  reference_sum
    Which is: '\x6' (6)
N: 2
...
[==========] 72 tests from 2 test suites ran. (4381 ms total)
[  PASSED  ] 70 tests.
[  SKIPPED ] 1 test, listed below:
[  SKIPPED ] sycl.mdspan_minimal_functional
[  FAILED  ] 1 test, listed below:
[  FAILED  ] sycl.reducers_int8_t

Kokkos_AlgorithmsUnitTest_StdSet_Team_I

48/53 Test #48: Kokkos_AlgorithmsUnitTest_StdSet_Team_I ..........***Failed  Error regular expression found in output. Regex=[  FAILED  ]  2.72 sec
...
[ RUN      ] std_algorithms_reduce_team_test.test
/home/ndellin/kokkos/algorithms/unit_tests/TestStdAlgorithmsTeamReduce.cpp:240: Failure
Expected equality of these values:
  result
    Which is: 382398
  reduceResultsView_h(i)
    Which is: 269253
...
[ RUN      ] std_algorithms_transform_reduce_team_test.test
/home/ndellin/kokkos/algorithms/unit_tests/TestStdAlgorithmsTeamTransformReduce.cpp:261: Failure
Expected equality of these values:
  result
    Which is: 132655612
  resultsView_h(i)
    Which is: 93055015
...
[==========] 4 tests from 4 test suites ran. (2658 ms total)
[  PASSED  ] 2 tests.
[  FAILED  ] 2 tests, listed below:
[  FAILED  ] std_algorithms_reduce_team_test.test
[  FAILED  ] std_algorithms_transform_reduce_team_test.test

Please include the following for a minimal reproducer SHA: 8e0d4a9234f676664346ce47b88c9b85a02c7354

Reproducer notes (Blake PV queue):

source /projects/x86-64-icelake-rocky8/spack-config/blake-setup-user-module-env.sh
module purge
module load cmake intel-oneapi-compilers/2024.1.0 intel-oneapi-dpl/2022.5.0 intel-oneapi-tbb/2021.12.0 intel-oneapi-mkl/2024.0.0

$KOKKOS_PATH/generate_makefile.bash --with-sycl --arch=INTEL_PVC --compiler=icpx --cxxflags="-fp-model=precise"
masterleinad commented 5 months ago

Seems to be fine with 2024.2.0 on the testbeds.

crtrott commented 5 months ago

So are you saying it works with something older than 2024.1.0 and it works with 2024.2.0?

masterleinad commented 5 months ago

So are you saying it works with something older than 2024.1.0 and it works with 2024.2.0?

No, it seems it works with RelWithDebInfo but fails inRelease mode.