Good afternoon Marco, Gunnar:
I followed the instructions containing in the Gunnar e-mail (those contained in the https://stackoverflow.com/questions/38090206/linking-intels-math-kernel-library-mkl-to-r-on-windows), and it works really fine!! I reproduce them below, because I think it is a very useful way to upgrade the current R version. Just in case it was useful, I could share the last MRO version (4.0.2).
Thank you very much for your contributions,
Fernando.
De: statet-users <statet-users-bounces@xxxxxxxxxxx> En nombre de Hofmann, Marco via statet-users
Enviado el: jueves, 20 de julio de 2023 18:59
Para: Gunnar Deuring <gdeuring@xxxxxxxxxxx>; 'Statet user discussions' <statet-users@xxxxxxxxxxx>
CC: Hofmann, Marco <Marco.Hofmann@xxxxxxxx>
Asunto: Re: [statet-users] Problems Updating the R environment index
Hello Gunnar,
I don’t know if really all files are needed and the solution to replace 2 files by 47 is likely not the most elegant one.
For windows users it is a more or less easy way to use an optimised BLAS/LAPACK which should probably also work in the (near) future..
Meanwhile, it seems there is also a way to use OpenBLAS with R on windows without the need to build R. But I have not tried this.
https://github.com/david-cortes/R-openblas-in-windows
Kind regards
Marco
Hi Marco,
thanks for your contribution, the Intel oneAPI MKL is likely the future way to go.
I also had started out with the Intel oneAPI MKL, however, first it was unclear, which one of the oneMKL packages to download (Stand Alone Version, Offline Installer is fine), then and a drag to find the relevant files since the location had changed from the description I was following, and further, the amount of files (47) and their size (~ 2 GB), made it somewhat cumbersome to implement, or to recommend and share that procedure. It would be nice to know, if really all those dll-files are relevant for use with R, do you have any idea?
I found the MRO approach slicker and way easier to follow, at least once you let go of the MRO packages functionality. But with MRO being discontinued and new processors being developed it may be just a matter of time until MRO is not the “better” option anymore.
Spread the knowledge (MRO and Intel oneAPI), since the Intel MKL is free now, it is a great improvement over the native R BLAS/LAPACK.
Kind regards,
Gunnar
Von: Hofmann, Marco [mailto:Marco.Hofmann@xxxxxxxx]
Gesendet: Mittwoch, 19. Juli 2023 15:07
An: Statet user discussions
Cc: Gunnar Deuring
Betreff: AW: [EXTERN] Re: [statet-users] Problems Updating the R environment index
Hello Fernando,
in addition to the answer of Gunnar, I can confirm that the second solution in the thread Gunnar mentioned, starting with “Alternatively, if you don't like copying...” does also work and that there is no need to adapt the Rprofile.site file.
In short, install the Intel oneAPI MKL (I choosed the standalone version):
https://www.intel.com/content/www/us/en/developer/tools/oneapi/onemkl.html
The path to the compiler and mkl files has changed meanwhile and can be found here:
C:\Program Files (x86)\Intel\oneAPI\compiler\latest\windows\redist\intel64_win\compiler
C:\Program Files (x86)\Intel\oneAPI\mkl\latest\redist\intel64
The rest (copying and renaming) is described in the thread.
But I would like to mention that the increase in speed depends strongly on the type of calculation. As I don`t need to calculate cross-products of matrices or something like that, I experienced a strong speed increase in benchmark tests but not in my every-day applications. But give it a try!
Marco
Hi Fernando,
I just took a look at the computer I’m currently working on. I hadn’t modified the original CRAN Rprofile.site there and only copied the files
Rlapack.dll
Rblas.dll
libiomp5md.dll
to R…\bin\x64
I remember now, it was complicated and I didn’t know, what the packages such as “RevoScaleR” or “RevoUtilsMath” exactly are doing, since I didn’t use those ever before.
The Intel MK libraries are working nonetheless very well without them. Try a benchmark before and after, it’s stunning!
If you would like to use the MRO Rprofile.site, at least adapt the line
r["CRAN"] <- RevoUtils::getRevoRepos()
options(repos="https://stat.ethz.ch/CRAN/")
to a package repo of your preference.
Kind regards,
Gunnar
Thank you very much for your e-mail, Gunnar:
Yes, it is a possibility if I cannot run it properly. Please, could you be so kind to send me your Rprofile.site file in order to use it as example to adap it to use the MKL libraries in the standard R?
Thank you very much again for your kind support,
Hi Fernando,
I can’t help you with the error, but if you are using that outdated Microsoft R Open version 4.0.2 only to enjoy their integrated Intel MKL multithreaded BLAS libraries, I would like to point out to you that you can extract those from MRO and use them also with newer versions of CRAN R, which possibly don’t have compatibility issues with up-to-date versions Eclipse or StatET.
As a starting point I’d suggest the steps described here:
https://stackoverflow.com/questions/38090206/linking-intels-math-kernel-library-mkl-to-r-on-windows
(Highest rated Answer “Easier solution than having to recompile R against …”)
I basically copied the relevant files from the MRO installation to the R…\bin\x64 folder, replacing the native Rblas.dll and Rlapack.dll (rename to keep, just in case!), and copied some MRO-specific packages to my library, however, the latter may be less relevant. I further adapted the Rprofile.site, mainly to adjust the url to the CRAN package repository (I also removed all references to the MRO functions, which I don’t use).
Kind regards,
Gunnar
Hi everybody:
I have installed Microsoft R Open version 4.0.2, and I am trying to run it together with eclipse 2023-06. I have installed in MRO rj_4.0.1-4.zip and rj.gd_4.0.1-3.zip. When I launch MRO from eclipse all seems to work fine; message “[INFO ] Updating the R environment index started...” appears as usual; but after 5 to 10 minutes, the message “Error in textConnection("rval", "w", local = TRUE) : all connections are in use” appears, and I have to kill the process, as it can see in the figure below.
Please Can give me someone any idea to solve this annoying problem??
Thank you very much, and best regards,
Fernando.
