The testcase test_graph_unit_dnnl_layer_norm_usm_cpu fails

[yurivict]

Summary

----------------------------------------------------------

113/125 Testing: test_graph_unit_dnnl_layer_norm_usm_cpu
113/125 Test: test_graph_unit_dnnl_layer_norm_usm_cpu
Command: "/usr/ports/math/onednn/work/.build/tests/gtests/graph/unit/test_graph_unit" "--gtest_filter=test_layer_norm_*" "--engine=cpu"
Directory: /usr/ports/math/onednn/work/.build/tests/gtests/graph/unit
"test_graph_unit_dnnl_layer_norm_usm_cpu" start time: Jul 27 00:19 PDT
Output:
----------------------------------------------------------
Note: Google Test filter = test_layer_norm_*:-*_GPU*
[==========] Running 6 tests from 2 test suites.
[----------] Global test environment set-up.
[----------] 5 tests from test_layer_norm_execute
[ RUN      ] test_layer_norm_execute.LayernormNDimCheck
<end of output>
Test time =   0.17 sec
----------------------------------------------------------
Test Failed. 
"test_graph_unit_dnnl_layer_norm_usm_cpu" end time: Jul 27 00:19 PDT
"test_graph_unit_dnnl_layer_norm_usm_cpu" time elapsed: 00:00:00
----------------------------------------------------------

Version

Report oneDNN version and githash. Version information is printed to stdout in verbose mode.

Environment

• CPU make and model (try lscpu; if your lscpu does not list CPU flags, try running cat /proc/cpuinfo | grep flags | sort -u)
• OS version FreeBSD 14.1
• Compiler version clang-18
• Version: 3.5.2

[TaoLv]

@rongzha1 Could you please take a look on this?

[rongzha1]

thanks for your report, could you show me the build log and build script？ And what platform did you use? I can't reproduce it in my env on SRF and GNR.

[yurivict]

here is the full log.

[yurivict]

cmake args: -DDNNL_CPU_RUNTIME=OMP -DCMAKE_C_COMPILER:STRING="cc" -DCMAKE_CXX_COMPILER:STRING="c++" -DCMAKE_C_FLAGS:STRING="-O2 -pipe -fstack-protector-strong -isystem /usr/local/include -fno-strict-aliasing " -DCMAKE_C_FLAGS_DEBUG:STRING="-O2 -pipe -fstack-protector-strong -isystem /usr/local/include -fno-strict-aliasing " -DCMAKE_C_FLAGS_RELEASE:STRING="-O2 -pipe -fstack-protector-strong -isystem /usr/local/include -fno-strict-aliasing -DNDEBUG" -DCMAKE_CXX_FLAGS:STRING="-O3 -pipe -isystem /usr/local/include " -DCMAKE_CXX_FLAGS_DEBUG:STRING="-O3 -pipe -isystem /usr/local/include " -DCMAKE_CXX_FLAGS_RELEASE:STRING="-O3 -pipe -isystem /usr/local/include -DNDEBUG" -DCMAKE_EXE_LINKER_FLAGS:STRING=" -fstack-protector-strong -L/usr/local/lib " -DCMAKE_MODULE_LINKER_FLAGS:STRING=" -fstack-protector-strong -L/usr/local/lib " -DCMAKE_SHARED_LINKER_FLAGS:STRING=" -fstack-protector-strong -L/usr/local/lib " -DCMAKE_INSTALL_PREFIX:PATH="/usr/local" -DCMAKE_AUTOGEN_PARALLEL:STRING="7" -DCMAKE_BUILD_TYPE:STRING="Release" -DTHREADS_HAVE_PTHREAD_ARG:BOOL=YES -DCMAKE_INSTALL_RPATH_USE_LINK_PATH=YES -DCMAKE_VERBOSE_MAKEFILE:BOOL=ON -DFETCHCONTENT_FULLY_DISCONNECTED:BOOL=ON -DDNNL_BUILD_TESTS:BOOL=OFF -DDNNL_BUILD_EXAMPLES:BOOL=OFF -GNinja

[rongzha1]

Hi, @yurivict . I use clang-18 and same camke options( except -DDNNL_BUILD_TESTS:BOOL=ON for test_layernorm* cases). but still can't reproduce the issue. log as attached picture shows.

I guess it may be a corner bug that only occurs with a specific toolchain or OS. I don't have a FreeBSD. Could you help gdb it (I'll fully support you) or can I get the platform I will debug it myself?

[TaoLv]

Hi @yurivict, recently we fixed a bug in the unit test (d815a479118d884e65151be79f5abc1bdd52146c). Could you please check if you still see this issue in your environment? Thanks.