On 22 Nov 2024, at 20:54, Simon Urbanek <simon.urbanek at R-project.org> wrote:
Gilberto,
On Nov 23, 2024, at 12:40 PM, Gilberto Camara <gilberto.camara at inpe.br> wrote:
Dear Simon
You state that ?R itself doesn't load OpenMP?. So why is libomp.dylib loaded in core libraries of R-4.4.2 for Mac?
It is not loaded - you can check yourself:
echo '' | R CMD sh -c 'DYLD_PRINT_LIBRARIES=1 $R_HOME/bin/exec/R' 2>&1 | grep libomp
Please see the contents of the core R libaries for R-4.2 and R-4.4.2 in my MacOS. In both cases, these are downloads straight from CRAN.
Why has libomp.dylib been included in the R core libraries?
We simply distribute it as a convenience, because some packages (like data.table) benefit hugely from OpenMP so by providing a fixed location it allows them to avoid shipping it separately and they all can use one shared version. See also https://mac.r-project.org/openmp/ for details
But as noted, I don't think this has anything to do with your problem precisely because it is not loaded by R itself.
As I said before I would appreciate is you avoided random speculations (there are many changes between R 4.2 and 4.4 especially in the packages between the two versions) as that is not productive, but instead share a reproducible example that we can look at. Any discussion is entirely pointless until you do that.
Cheers,
Simon
Best regards
Gilberto
R 4.2
? ls -lha /Library/Frameworks/R.framework/Versions/4.2/Resources/lib
total 19976
drwxrwxr-x 15 root admin 480B 22 Nov 15:28 .
drwxrwxr-x 18 root admin 576B 22 Nov 15:28 ..
-rwxrwxr-x 1 root admin 4,0M 17 Mar 2023 libR.dylib
drwxrwxr-x 3 root admin 96B 17 Mar 2023 libR.dylib.dSYM
-rwxrwxr-x 1 root admin 221K 17 Mar 2023 libRblas.0.dylib
drwxrwxr-x 3 root admin 96B 17 Mar 2023 libRblas.0.dylib.dSYM
lrwxrwxr-x 1 root admin 16B 22 Nov 15:28 libRblas.dylib -> libRblas.0.dylib
drwxrwxr-x 3 root admin 96B 17 Mar 2023 libRblas.dylib.dSYM
-rwxrwxr-x 1 root admin 151K 17 Mar 2023 libRblas.vecLib.dylib
drwxrwxr-x 3 root admin 96B 17 Mar 2023 libRblas.vecLib.dylib.dSYM
-rwxrwxr-x 1 root admin 2,2M 17 Mar 2023 libRlapack.dylib
drwxrwxr-x 3 root admin 96B 17 Mar 2023 libRlapack.dylib.dSYM
-rw-rw-r-- 1 root admin 157K 17 Mar 2023 libgcc_s.1.dylib
-rwxrwxr-x 1 root admin 2,7M 17 Mar 2023 libgfortran.5.dylib
-rwxrwxr-x 1 root admin 302K 17 Mar 2023 libquadmath.0.dylib
R 4.4.2
? ls -lha /Library/Frameworks/R.framework/Versions/4.4-x86_64/Resources/lib
total 27320
drwxrwxr-x 17 root admin 544B 22 Nov 15:03 .
drwxrwxr-x 17 root admin 544B 22 Nov 15:03 ..
-rwxrwxr-x 1 root admin 5,3M 31 Out 22:15 libR.dylib
drwxrwxr-x 3 root admin 96B 31 Out 22:14 libR.dylib.dSYM
-rwxrwxr-x 1 root admin 238K 31 Out 22:15 libRblas.0.dylib
drwxrwxr-x 3 root admin 96B 31 Out 22:14 libRblas.0.dylib.dSYM
lrwxrwxr-x 1 root admin 16B 22 Nov 15:03 libRblas.dylib -> libRblas.0.dylib
drwxrwxr-x 3 root admin 96B 31 Out 22:14 libRblas.dylib.dSYM
-rwxrwxr-x 1 root admin 151K 31 Out 22:15 libRblas.vecLib.dylib
drwxrwxr-x 3 root admin 96B 31 Out 22:14 libRblas.vecLib.dylib.dSYM
-rwxrwxr-x 1 root admin 2,3M 31 Out 22:15 libRlapack.dylib
drwxrwxr-x 3 root admin 96B 31 Out 22:14 libRlapack.dylib.dSYM
-rw-rw-r-- 1 root admin 183K 31 Out 22:15 libgcc_s.1.1.dylib
-rwxrwxr-x 1 root admin 3,3M 31 Out 22:15 libgfortran.5.dylib
-rwxrwxr-x 1 root admin 1,5M 31 Out 22:15 libomp.dylib
-rwxrwxr-x 1 root admin 376K 31 Out 22:15 libquadmath.0.dylib
drwxrwxr-x 3 root admin 96B 31 Out 22:11 pkgconfig
============================
Prof Dr Gilberto Camara
Senior Researcher
Getulio Vargas Foundation (FGV)
National Institute for Space Research (INPE), Brazil
https://gilbertocamara.org/
=============================
On 22 Nov 2024, at 19:48, Simon Urbanek <simon.urbanek at R-project.org> wrote:
Gilberto,
you still didn't provide any useful examples, and all you have are baseless speculations. R itself doesn't load OpenMP, so the problem could very well be in the other software or libraries you use - it may have nothing to do with R, but without an example we can neither tell nor help. Based on this there is nothing to solve "at the highest level of R core developers for Mac" - it has to start with you providing a working example as no one else has that problem.
Cheers,
Simon
On Nov 23, 2024, at 7:30 AM, Gilberto Camara <gilberto.camara at inpe.br> wrote:
Dear Peter and Simon
I may have found the cause of the crashes in R-4.4.2 in the Intel MacOs environment. In R-4.4.2 for MacOS, the ?libomp.dylib? is loaded from the CRAN version. Comparing with earlier version, this is a new addition to the core R libraries. The OpenMP library is not provided in R-4.2.
It so happens that the ?torch? package is also including the an OpenMP library (?libiomp5.dylib?) which is causing the crash. In previous R versions, this behavoir did not cause problems, since R did not provide the OpenMP library by default.
Thus, the problem needs to be solved at the highest level of R core developers for Mac. Please bear in mind that the ?torch? package is essential for running deep learning algorithms in R. I hope you can find a solution that preserves the use of ?torch? in R. I am copying the message to Daniel Falbel, who is the maintainer of the ?torch? package.
Your support in solving this problem is highly appreciated.
Best regards
Gilberto
============================
Prof Dr Gilberto Camara
Senior Researcher
Getulio Vargas Foundation (FGV)
National Institute for Space Research (INPE), Brazil
https://gilbertocamara.org/
=============================
On 22 Nov 2024, at 14:18, Gilberto Camara <gilberto.camara at inpe.br> wrote:
Dear Peter
Many thanks for your help. I ran R in a terminal, something I admit I had not done before.
It crashed and produced the following message:
===
OMP: Error #15: Initializing libiomp5.dylib, but found libomp.dylib already initialized.
OMP: Hint This means that multiple copies of the OpenMP runtime have been linked into the program. That is dangerous, since it can degrade performance or cause incorrect results. The best thing to do is to ensure that only a single OpenMP runtime is linked into the process, e.g. by avoiding static linking of the OpenMP runtime in any library. As an unsafe, unsupported, undocumented workaround you can set the environment variable KMP_DUPLICATE_LIB_OK=TRUE to allow the program to continue to execute, but that may cause crashes or silently produce incorrect results. For more information, please see http://www.intel.com/software/products/support/.
===
I am using Rcpp Armadillo, and I am following its guidelines for C++ compilation. The ?Makevars? file is produced by running
usethis::use_rcpp_armadillo()
which produces the following Makevars file:
PKG_CXXFLAGS = $(SHLIB_OPENMP_CXXFLAGS)
PKG_LIBS = $(SHLIB_OPENMP_CXXFLAGS) $(LAPACK_LIBS) $(BLAS_LIBS) $(FLIBS)
I removed the Makevars file, compiled my package from scratch but the error continues. Any help on how to solve this problem would be most appreciated.
Best regards
Gilberto
============================
Prof Dr Gilberto Camara
Senior Researcher
Getulio Vargas Foundation (FGV)
National Institute for Space Research (INPE), Brazil
https://gilbertocamara.org/
=============================
On 22 Nov 2024, at 12:08, peter dalgaard <pdalgd at gmail.com> wrote:
A couple of questions:
- Is it the GUI or R itself that crashes? (i.e, can you run R in Terminal an still crash it? or even RStudio?)
- Is it a Sequoia issue as such? I'm not seeing issues on Monterey/4.4.1 (I'm a little superstitious about upgrading production machines mid-semester and some of my machines are too old for Sequoia.) The source tarballs 4.4.2 were built and tested on Intel/Monterey.
- pd
On 21 Nov 2024, at 16:07 , Gilberto Camara <gilberto.camara at inpe.br> wrote:
Dear Simon
Since your message last week, I have been trying to reproduce the errors I am finding with R-4.4.2 in an Intel MacMini to build a minimum testable example. I am not succeeding in doing so.
The error is a total collapse of R and occurs in a random fashion. Sometimes calling another package (e.g. ?xgboost?) produces the error. Sometimes making a simple operation in a data table leads to error.
The problem affects only Intel-based MacMinis with R-4.4.2. Running MacMinis with R-4.2.3 work will. All is also well with R-4.4.2 in Macs with ARM, in Windows and in Lunix/Ubuntu and Linux/Fedora.
One issue I noticed is that the default C++ compiler in MacOS Sequoia is compatible with c++-17, while the one used by CRAN is version c++-14.00.
Do you have any ideas on how to proceed? Any test data set I could use?
Many thanks
Gilberto
============================
Prof Dr Gilberto Camara
Senior Researcher
Getulio Vargas Foundation (FGV)
National Institute for Space Research (INPE), Brazil
https://gilbertocamara.org/
=============================
On 12 Nov 2024, at 18:53, Simon Urbanek <simon.urbanek at R-project.org> wrote:
Gilberto,
please read https://www.r-project.org/bugs.html first, in particular about how to report a problem. Just saying "cannot execute my scripts" is not helpful at all - please provide exact output, how it differs between the versions etc. For example, the problem may be in your code or packages used - we have no way of knowing without the necessary additional information.
Thanks,
Simon
On Nov 13, 2024, at 9:44 AM, Gilberto Camara <gilberto.camara at inpe.br> wrote:
Dear R-SIG-MAC
I have problems with R-4.4.2 in the latest Mac OS Sequoia (15.1) version in a Mac mini with an Intel chip. R-4.4.2 cannot execute my scripts. No such problems occur with R-4.4.2 in MacBook with ARM chip.
I went back to R-4.2.3 and all is well in my environment (MacMini, Intel, MacOS X 15.1).
I will try to provide an MWE to help those of you who know a lot about Mac OS X.
Best
Gilberto
============================
Prof Dr Gilberto Camara
Senior Researcher
Getulio Vargas Foundation (FGV)
National Institute for Space Research (INPE), Brazil
https://gilbertocamara.org/
=============================