Pearu Peterson

@Sivannnnnn when starting up omniscdb server with `--udf`, does it report anything related to compiling the UDFs for CUDA? If it does, pay attention to the error messages and try...

Try to locate libdevice.10.bc in your system and define CUDA_HOME env such that ``` $CUDA_HOME/nvvm/libdevice/libdevice.10.bc ``` points to the file. Then re-start the server.

> @pearu Yes, I have this file, Can you provide a full path to it here? > how can I associate this file with omniscidb, See https://github.com/heavyai/heavydb/issues/728#issuecomment-1084163109 > I can't...

The content of ``` export CUDA_HOME="/usr/local/cuda-11.6/" export CUDA_PATH="/usr/local/cuda-11.6/bin/" export PATH="$CUDA_PATH:$PATH" ``` is suspicious. Try: ``` export CUDA_HOME="/usr/local/cuda-11.6/" export CUDA_PATH="/usr/local/cuda-11.6/" export PATH="$CUDA_PATH/bin:$PATH" ```

What version of the clang compiler are you using? ``` clang++ --version ```

Thanks for the data! Notice that ``` /opt/omnisci/bin/omnisci_server --config /var/lib/omnisci/omnisci.conf ``` does not compile load-time C++ UDFs and the `error: cannot find libdevice for sm_60...` may have different/unrelated origins. What...

> Because I specified the UDF in the conf, for example Notice that udf is commented out in the conf file. Try enabling it. For the sake of sanity, could...

Btw, what omniscidb version are you using: ``` /opt/omnisci/bin/omnisci_server --config /var/lib/omnisci/omnisci.conf --version ``` ?

> @pearu the only to don't get the warning is using cuda 10.2. > The funny thing is running the clang with the same parameters used by the server you...

> I am not sure that replacing `asyncio.coroutine` with `types.coroutine` would not break the compatibility of Python 3.4 or lower version. I will test it later. It will likely break...