error while loading shared libraries: libsycl.so.5

nhoj · ‎01-11-2023

I have recently installed oneAPI Base Toolkit and the Metrics Discovery API library on Ubuntu

However the vtune-self-checker Fails on checking DPC++ application as prerequisite for GPU analyses.

The associated log (below) reports: "error while loading shared libraries: libsycl.so.5: cannot open shared object file: No such file or directory"

I note that the following files exist:

/opt/intel/oneapi/compiler/latest/linux/lib/libsycl.so -> ./libsycl.so.6
/opt/intel/oneapi/compiler/latest/linux/lib/libsycl.so.6 -> ./libsycl.so.6.0.0
/opt/intel/oneapi/compiler/latest/linux/lib/libsycl.so.6.0.0

Software versions:
Ubuntu 22.04.1 LTS
intel-basekit version (2023.0.0-25537).
VTune(TM) Profiler 2023.0.0 (build 624757)
intel-metrics-discovery-1.12.160-1_amd64.deb

How do I correct this problem?

============================================
excerpt from vtune-self-checker log.txt
----------------------------------------

Checking DPC++ application as prerequisite for GPU analyses...
Setting envirnoment variable:
SYCL_DEVICE_FILTER=opencl:gpu
Command line:
/opt/intel/oneapi/vtune/2023.0.0/bin64/self_check_apps/matrix.dpcpp/matrix.dpcpp
Stderr:
/opt/intel/oneapi/vtune/2023.0.0/bin64/self_check_apps/matrix.dpcpp/matrix.dpcpp: error while loading shared libraries: libsycl.so.5: cannot open shared object file: No such file or directory

Setting envirnoment variable:
SYCL_DEVICE_FILTER=level_zero:gpu
Command line:
/opt/intel/oneapi/vtune/2023.0.0/bin64/self_check_apps/matrix.dpcpp/matrix.dpcpp
Stderr:
/opt/intel/oneapi/vtune/2023.0.0/bin64/self_check_apps/matrix.dpcpp/matrix.dpcpp: error while loading shared libraries: libsycl.so.5: cannot open shared object file: No such file or directory

Setting envirnoment variable:
ZES_ENABLE_SYSMAN=1
Command line:
/opt/intel/oneapi/vtune/2023.0.0/bin64/self_check_apps/matrix.dpcpp/matrix.dpcpp
Stderr:
/opt/intel/oneapi/vtune/2023.0.0/bin64/self_check_apps/matrix.dpcpp/matrix.dpcpp: error while loading shared libraries: libsycl.so.5: cannot open shared object file: No such file or directory

Checking DPC++ application as prerequisite for GPU analyses: Fail
Unable to run DPC++ application on GPU connected to this system. If you are using an Intel GPU and want to verify profiling support for DPC++ applications, check these requirements:
* Install Intel(R) GPU driver.
* Install Intel(R) Level Zero GPU runtime.
* Install Intel(R) oneAPI DPC++ Runtime and set the environment.

Soumya_M_Intel · ‎01-12-2023

Hi nhoj,

Thanks for posting your concern on the forum. However, this forum is for assistance on getting started (such as registration, downloading, licensing and installing) questions regarding the Intel® oneAPI Toolkits, Intel® Parallel Studio XE and Intel® System Studio. Due to that the query is out of our scope.

However, based on your query, we are moving this topic to the 'Analyzers' community for further assistance. Our internal team will get back to you there in a timely manner.

Hope you have a nice day ahead.

Regards,

Soumya

AthiraM_Intel · ‎01-13-2023

Hi,

We ran one dpcpp sample on DevCloud with the old binary and got a similar error that you got.

The reason is oneAPI is updated to 2023.0 and in this "runtime of oneAPI 2023.0 is not backwards compatible". So, we can’t run the old binaries, in order to run the old binaries we need to rebuild those binaries again.

To resolve the issue, you need to create new binary file after the installation of oneAPI base toolkit

Please find the screenshot below:

If you have any further issue, please let us know.

Thanks

nhoj · ‎01-21-2023

Thanks Athria for your prompt response - and apologies for my tardy reply.

Sorry this has NOT solved my problem - I am new to OneAPI so I may need a little more detailed guidance if possible:

I was unable to locate file called mat_mul.cpp in my installation - but there is a number of similarly named examples matmul.cpp

:/opt/intel$ find -name mat*.cpp
./oneapi/dnnl/2023.0.0/cpu_dpcpp_gpu_dpcpp/examples/matmul.cpp
./oneapi/dnnl/2023.0.0/cpu_dpcpp_gpu_dpcpp/examples/matmul_perf.cpp
./oneapi/dnnl/2023.0.0/cpu_tbb/examples/matmul.cpp
./oneapi/dnnl/2023.0.0/cpu_tbb/examples/matmul_perf.cpp
./oneapi/dnnl/2023.0.0/cpu_gomp/examples/matmul.cpp
./oneapi/dnnl/2023.0.0/cpu_gomp/examples/matmul_perf.cpp
./oneapi/dnnl/2023.0.0/cpu_iomp/examples/matmul.cpp
./oneapi/dnnl/2023.0.0/cpu_iomp/examples/matmul_perf.cpp
./oneapi/diagnostics/2022.2.0/checkers_py/oneapi_check_offloads/matmul_offload.cpp

Also I do not have icpx installed (but I can install) - or will gcc suffice?

AthiraM_Intel · ‎01-20-2023

Hi,

We have not heard back from you. Could you please give us an update? Is your issue resolved?

Thanks

AthiraM_Intel · ‎01-25-2023

Hi,

The mat_mul.cpp sample we used is a custom file created by us. You can recompile your own sample after sourcing the oneAPI environment using the below command:

source /opt/intel/oneapi/setvars.sh --force

Now you will be able to compile your cpp file using icpx/dpcpp

icpx -fsycl <cpp file> -o <output_file_name>

>>Also I do not have icpx installed (but I can install) - or will gcc suffice?

If you source the oneAPI environment as mentioned above, you could use icpx.

Hope this helps. If you face any issue, please let us know.

Thanks