Common Issues

Installation Issues

Cannot find gmake

When the command of GNU make is make instead of gmake, please set the environmental variable

$ export GNU_MAKE=make # for bash
$ setenv GNU_MAKE make # for csh/tcsh

or set the variable when calling mm,

$ GNU_MAKE=make mm

Cannot find cublas_v2.h

For certain Linux systems, NVIDIA installer installs cublas to the system directory /usr/include and /usr/lib/x86_64-linux-gnu instead of /usr/local/cuda. In this case, please add the include and library paths to cuda.incpath and cuda.libpath in config.mm file.

Run-time Issues

Locales

Error

UnicodeDecodeError: ‘ascii’ codec can’t decode byte 0xc3 in position 18: ordinal not in range(128)

You might need to set the LANG variable,

$ export LANG=en_US.UTF-8

if en_US.UTF-8 locale is not installed, update your locale by

$ sudo apt install locales
$ sudo locale-gen --no-purge --lang en_US.UTF-8
$ sudo update-locale LANG=en_US.UTF-8 LANGUAGE

Base case name

Error

altar: bad case name: ‘patch-9’

The AlTar App cannot find the configuration file (usually) or the input file directory. Please go to the job directory with the configuration and input files, and run the App again. If the configuration is not named as theAlTarApp.pfg, you need --config=YourConfigFile.pfg option, e.g.,

linear --config=my_linear_model.pfg

Configuration Parser Error

Error

File “/opt/anaconda3/envs/altar/lib/python3.9/site-packages/pyre/parsing/Scanner.py”, line 71, in pyre_tokenize

match = stream.match(scanner=self, tokenizer=tokenizer)

This is usually due to a bad format in configuration file. For example, .pfg files do not recognize TABs; please check your file for possible TABs and replace them with SPACEs. See Pyre Config Format (.pfg) for more details.

MPI launcher error

Error

launcher = self.mpi.launcher

AttributeError: ‘NoneType’ object has no attribute ‘launcher’

This happens when AlTar cannot locate the mpirun command. It can be solved by manually setting up an mpi.pfg file. See MPI setup for more details.

Intel MKL Library

Error

Intel MKL FATAL ERROR: Cannot load libmkl_avx2.so.1 or libmkl_def.so.1.

This is due to a Conda issue with MKL libraries. The solution is to preload certain MKL libraries before running AlTar applications,

LD_PRELOAD=$CONDA_PREFIX/lib/libmkl_core.so:$CONDA_PREFIX/lib/libmkl_sequential.so altarApp --config=configFile