Name: mkl_wrapper_for_non_CDH
Owner: Intel-bigdata
Description: null
Created: 2017-05-23 09:51:26.0
Updated: 2018-03-29 23:50:32.0
Pushed: 2017-05-23 10:05:24.0
Homepage: null
Size: 247
Language: null
GitHub Committers
User | Most Recent Commit | # Commits |
---|
Other Committers
User | Most Recent Commit | # Commits |
---|
Assume the MKL installation location is /opt/intel/mkl (you cannot just copy libmkl_rt.so).
Copy mkl_wrapper.jar to /opt/intel/mkl/wrapper/mkl_wrapper.jar
Copy mkl_wrapper.so to /opt/intel/mkl_wrapper/mkl_wrapper.so
By default, the MKL wrapper will search /opt/intel/mkl/wrapper to find mkl_wrapper.so. If you put it on other path, please use -Dcom.intel.mkl.wrapper=/path/to/find/mkl_wrapper.so.
Spark configuration. (on each node):
Create: /etc/ld.so.conf.d/mkl_blas.conf
Input: /opt/intel/mkl/lib/intel64 (your MKL library Dir) to mkl_blas.conf
Run: ldconfig
A simple sanity check to see if native MKLBLAS/LAPACK libs can be properly loaded:
bin/spark-shell –driver-class-path $CLASSPATH:/opt/intel/mkl/wrapper/mkl_wrapper.jar
scala> import com.github.fommil.netlib.BLAS;
scala> System.setProperty(“com.github.fommil.netlib.BLAS”, “com.intel.mkl.MKLBLAS”)
scala> System.out.println(BLAS.getInstance().getClass().getName());
OUTPUT: com.intel.mkl.MKLBLAS
scala> import com.github.fommil.netlib.LAPACK;
scala> System.setProperty(“com.github.fommil.netlib.LAPACK”, “com.intel.mkl.MKLLAPACK”)
scala> System.out.println(LAPACK.getInstance().getClass().getName());
OUTPUT: com.intel.mkl.MKLLAPACK