Installation of R-4.3.1 with intel 2022

Note that 'intel 2022' is a bit vague.  The current version is 2023.1.0,
and that has both the 'classic' (icc/icpc/ifort which it seems you used)
and new (icx/ixpx/ifx) compilers -- the former are said to be going to
be discontinued later this year.  R did not know about ifx so did not
build with the new set.

The parts of the manual Tomas referred to were about the old compilers:
the manual has now been re-written in R-devel and R-patched to major on
the newer ones.  We have patched the code to work with both old and new
compilers, pending a more thorough investigation of matherr.

In our experiments Intel Fortran only worked in conjunction with oneAPI
MKL -- see the manual.  It seems ifx is still under development, so it
should pay to use only the latest versions.

This is the first report on Intel compilers since 2015, so they are
rather low priority for the R developers.


On 21/06/2023 08:10, Tomas Kalibera wrote:

On 6/20/23 18:47, Giuseppe Cal? wrote:

Hi all,
I have the issue:

icc -std=c99 -std=gnu11 -I../../src/extra -I../../src/extra/xdr -I.
-I../../src/include -I../../src/include  -I/usr/local/include
-I../../src/nmath -DHAVE_CONFIG_H   -fopenmp -fpic  -g -O3 -wd188 -ip
-mp  -c eval.c -o eval.o
arithmetic.c(66): warning #274: declaration is not visible outside of
function
   int matherr(struct exception *exc)
                      ^

arithmetic.c(68): error: pointer to incomplete class type is not allowed
       switch (exc->type) {
               ^

arithmetic.c(69): error: identifier "DOMAIN" is undefined
       case DOMAIN:
            ^

arithmetic.c(70): error: identifier "SING" is undefined
       case SING:
            ^

arithmetic.c(73): error: identifier "OVERFLOW" is undefined
       case OVERFLOW:
            ^

arithmetic.c(76): error: identifier "UNDERFLOW" is undefined
       case UNDERFLOW:
            ^

arithmetic.c(77): error: pointer to incomplete class type is not allowed
       exc->retval = 0.0;

icc -std=c99 -std=gnu11 -I../../src/extra -I../../src/extra/xdr -I.
-I../../src/include -I../../src/include  -I/usr/local/include
-I../../src/nmath -DHAVE_CONFIG_H   -fopenmp -fpic  -g -O3 -wd188 -ip
-mp  -c flexiblas.c -o flexiblas.o
icc: command line remark #10148: option '-mp' not supported
compilation aborted for arithmetic.c (code 2)
make[3]: *** [../../Makeconf:129: arithmetic.o] Error 2
make[3]: *** Waiting for unfinished jobs....
icc: command line remark #10148: option '-mp' not supported
make[3]: Leaving directory '/opt/sources/R/R-4.3.1-intel21/src/main'
make[2]: *** [Makefile:140: R] Error 2
make[2]: Leaving directory '/opt/sources/R/R-4.3.1-intel21/src/main'
make[1]: *** [Makefile:28: R] Error 1
make[1]: Leaving directory '/opt/sources/R/R-4.3.1-intel21/src'
make: *** [Makefile:62: R] Error 1

with oneapi-2022.1.0/compiler-rt/2022.1.0;
oneapi-2022.1.0/mkl/2022.1.0 while building R-4.3.1 on redhat 8.4
glibc2.28-189

I followed a workaround proposed:

https://community.intel.com/t5/Intel-C-Compiler/Error-when-compiling-R-from-source-code-ubuntu-18-04/m-p/1176401/thread-id/36575
<
https://community.intel.com/t5/Intel-C-Compiler/Error-when-compiling-R-from-source-code-ubuntu-18-04/m-p/1176401/thread-id/36575

Deactivate  HAVE_MATHERR macro in src/include/config.h

Hi Giuseppe,

thanks for the report. Undefining HAVE_MATHERR seems a valid work-around
to me, based on reading the thread above and the sources.

We could improve this in R, if keeping this code, at least improve the
configure check so that it also tests for the presence of the macros.

Using this workaroud I get R with:

BLAS/LAPACK:

/opt/intel/oneapi_2022.2.0/mkl/2022.1.0/lib/intel64/libmkl_intel_lp64.so.2;
LAPACK version 3.9.0

is correct?

Is these a way to avoid arithmetic issue?

My configure is:

module load intel-2021.6.0/2021.6.0 oneapi-2022.1.0/mkl
MKL="-L${MKLROOT}/lib/intel64 -lmkl_gf_lp64 -lmkl_core
-lmkl_gnu_thread -dl -fopenmp"
export CC="icc -std=c99"
export CFLAGS="-g -O3 -wd188 -ip -mp"
export FC=ifort
export FLAGS="-g -O3 -mp"
export CXX=icpc
export CXXFLAGS="-g -O3 -mp"
SHLIB_CXXLD=icpc
export MKL_INTERFACE_LAYER=GNU,LP64
export MKL_THREADING_LAYER=GNU
./configure --prefix=/opt/intel-2021.6.0/R/4.3.1 --with-blas="$MKL"
--with-lapack  --enable-memory-profiling --enable-BLAS-shlib
--enable-R-shlib --enable-R-static-lib --with-pcre2

AFAIK, neither icc nor MKL is regularly tested with R/CRAN packages, so
the risk of running into some issues is somewhat higher than for say GCC
and the reference BLAS/LAPACK.

Some hints on using icc and MKL can be found in the R Admin manual,
https://cran.r-project.org/doc/manuals/r-release/R-admin.html. Unless
you have done that already, you might want to check your configuration
against those, I didn't spot any obvious issue. If you find any other
problem, please report, so that it could be fixed or the hints updated.

Thanks,
Tomas

Thanks a lot,
Giuseppe.

???????????????????????????????????????
Giuseppe Cal?

Fondazione CMCC
Centro Euro-Mediterraneo sui Cambiamenti Climatici presso Complesso
Ecotekne
Universit? del Salento - Strada Prov.le Lecce - Monteroni 73100 Lecce

IT

http://www.cmcc.it <http://www.cmcc.it/>
https://goo.gl/maps/wtahPDbNVen

mobile: (+39) 3208190020
email: giuseppe.calo at cmcc.it

Le informazioni contenute in questo messaggio di posta elettronica e
negli allegati se presenti sono riservate e confidenziali: ne ?
vietata la diffusione in qualsiasi modo o forma (GDPR 2016/679).
Qualora lei non fosse il destinatario del messaggio, la invito a non
diffonderlo e ad eliminarlo dandone gentilmente comunicazione al
mittente.

The information included in this e-mail and any attachments are
confidential and may also be privileged (GDPR 2016/679).
If you are not the correct recipient, you are kindly requested to
notify the sender immediately, to cancel it and not disclose the
contents to any other person.





    [[alternative HTML version deleted]]

--
Brian D. Ripley,                  ripley at stats.ox.ac.uk
Emeritus Professor of Applied Statistics, University of Oxford