Baunsgaard commented on PR #1843:
URL: https://github.com/apache/systemds/pull/1843#issuecomment-1720975168

   > @Baunsgaard @j143 Hi, after some intensive trying using **gitpod** I could 
conclude following things:
   > 
   >     1. **libmkl_rt.so cannot open the file: no such file or directory** 
this error was because ldconfig could not find the exported LD_LIBRARY_PATH. I 
could solve this problem by adding a **.conf file** with 
**/opt/intel/oneapi/mkl/latest/lib/intel64** in **/etc/ld.so.conf.d** directory 
and reloading **ldconfig.**
   > 
   >     2. Next Thanks to @j143 I stumbled across his video about Intel MKL 
installation on ubuntu. If I would do it just like he mentioned using a 
**offline package installer** which has CLI based UI and requires user 
interaction ( _I could not find a way to do this using shell script_), 
everything would work fine after doing what I mentioned in 1.
   >        If we were to use **apt package installer** instead of offline one, 
somehow the include folder containing all the necessary header file like 
"mkl.h" is missing in mkl root folder. I tried using "locate mkl.h" no result.
   >        I could not find the solution to this.
   >        Next option would be to install whole intel-basekit which contains 
MKL as well. It is unfortunately too large ~13 GB I think and could have led to 
the not enough space error.
   > 
   >     3. Regarding "mkl_dnn.h" .I have mentioned here [[SYSTEMDS-3546] Push 
down image pre-processing to blas/mkl #1843 
(comment)](https://github.com/apache/systemds/pull/1843#issuecomment-1718879255),
 if we download it unzip it, we have 2 folders in it include and lib, then copy 
"mkl_dnn.h" and "mkl_dnn_types.h"(second name is something similar) for its 
include folder to /opt/intel/oneapi/mkl/latest/include (if present) and the 
libs to /opt/intel/oneapi/mkl/latest/lib/intel64. Using this trick I could 
compile the cpp files with "mkl_dnn.h"header file as well. **But not 
recommended it was just an experiment.**
   > 
   >     4. Nevertheless, I could run the benchmarks using both mkl and 
openblas, also DML implementations using -stats flag in gitpod. Although I 
could not install mkl properly in GitHub test docker container. I will process 
the results and post it.
   > 
   > 
   > @Baunsgaard should I comment out "mkl test case" so that we can merge? or 
how should I proceed now?
   
   lets go with the Blas test then and add some TODO that reference this PR 
about the MKL tests, then I will try to make it work at a later time, since i 
need it for other things.


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]

Reply via email to