[R-pkg-devel] Compile issues on r-devel-linux-x86_64-debian-clang with OpenMP

Wed May 22 15:54:17 CEST 2024

Thank you both for your responses and help! Kurt-- your message makes a lot of sense. I'll try to debug soon and will reach out if I have more questions.

Thanks,
Michelle
________________________________
From: Kurt Hornik <Kurt.Hornik using wu.ac.at>
Sent: Wednesday, May 22, 2024 3:57 AM
To: Dirk Eddelbuettel <edd using debian.org>
Cc: Nixon, Michelle Pistner <map5672 using psu.edu>; r-package-devel using r-project.org <r-package-devel using r-project.org>; Kurt Hornik <Kurt.Hornik using wu.ac.at>
Subject: Re: [R-pkg-devel] Compile issues on r-devel-linux-x86_64-debian-clang with OpenMP

[You don't often get email from kurt.hornik using wu.ac.at. Learn why this is important at https://aka.ms/LearnAboutSenderIdentification ]

>>>>> Dirk Eddelbuettel writes:

Friends, the Debian pretest check system uses LLVM 18 and has the
corresponding OpenMP headers and libraries installed, but R is
configured not to use these.  However, the configure test in fido does

cat <<EOF > test-omp.cpp
#include <omp.h>
int main() {
  return omp_get_num_threads();
}
EOF

## Execute R CMD SHLIB.
"${R_HOME}/bin/R" CMD SHLIB test-omp.cpp >/dev/null 2>&1
if test x"$?" = x"0"; then
    AC_MSG_RESULT([yes])
    openmp_already_works="yes"
else
    AC_MSG_RESULT([no])
fi

which does not work as intended: R CMD SHLIB will happily link but the
.so will end up with an unresolved symbol:

$ nm test-omp.so | grep omp_
                 U omp_get_num_threads

Hth
-k

> Hi Michelle,

> On 21 May 2024 at 13:46, Nixon, Michelle Pistner wrote:
> | Hi all,
> |
> | I'm running into build issues for my package (fido: https://nam10.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2Fjsilve24%2Ffido&data=05%7C02%7Cmap5672%40psu.edu%7C7993944f52ae4a017d3d08dc7a34d1d6%7C7cf48d453ddb4389a9c1c115526eb52e%7C0%7C0%7C638519614500263726%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C0%7C%7C%7C&sdata=C4BI6%2FUgTjd%2FCRPzBF8NuUkPvlQkIxb2r1%2BzIqKNPpE%3D&reserved=0)<https://github.com/jsilve24/fido> on the r-devel-linux-x86_64-debian-clang system on CRAN (full check log here: https://nam10.safelinks.protection.outlook.com/?url=https%3A%2F%2Fwin-builder.r-project.org%2Fincoming_pretest%2Ffido_1.1.0_20240515_211644%2FDebian%2F00install.out&data=05%7C02%7Cmap5672%40psu.edu%7C7993944f52ae4a017d3d08dc7a34d1d6%7C7cf48d453ddb4389a9c1c115526eb52e%7C0%7C0%7C638519614500272862%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C0%7C%7C%7C&sdata=TKpQrkImemwz%2FuK7yYqCLEECU6WCCvgYwcrAId%2Bpmb0%3D&reserved=0)<https://win-builder.r-project.org/incoming_pretest/fido_1.1.0_20240515_211644/Debian/00install.out>. fido relies on several of the Rcpp packages, and I think the error is due to how OpenMP is set up in our package. The error in question states:
> |
> | "Error: package or namespace load failed for �fido� in dyn.load(file, DLLpath = DLLpath, ...):
> |  unable to load shared object '/home/hornik/tmp/R.check/r-devel-clang/Work/build/Packages/00LOCK-fido/00new/fido/libs/fido.so':
> |   /home/hornik/tmp/R.check/r-devel-clang/Work/build/Packages/00LOCK-fido/00new/fido/libs/fido.so: undefined symbol: omp_get_thread_num"
> |
> | I've had a hard time recreating the error, as I can successfully get the package to build on other systems (GitHub action results here: https://nam10.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2Fjsilve24%2Ffido%2Factions&data=05%7C02%7Cmap5672%40psu.edu%7C7993944f52ae4a017d3d08dc7a34d1d6%7C7cf48d453ddb4389a9c1c115526eb52e%7C0%7C0%7C638519614500275695%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C0%7C%7C%7C&sdata=zXCul23gWAevfEhCsfyJg8KewU8fjSuy1qZgZemNy7M%3D&reserved=0)<https://github.com/jsilve24/fido/actions> including a system using the same version of R/clang as the failing CRAN check. Looking at the logs between the two, the major difference is the lack of -fopenmp in the compiling function on the CRAN version (which is there on the r-hub check version with the same specifications):
> |
> | (From the CRAN version) clang++-18 -std=gnu++17 -shared -L/home/hornik/tmp/R-d-clang-18/lib -Wl,-O1 -o fido.so ConjugateLinearModel.o MaltipooCollapsed_LGH.o MaltipooCollapsed_Optim.o MatrixAlgebra.o PibbleCollapsed_LGH.o PibbleCollapsed_Optim.o PibbleCollapsed_Uncollapse.o PibbleCollapsed_Uncollapse_sigmaKnown.o RcppExports.o SpecialFunctions.o test_LaplaceApproximation.o test_MultDirichletBoot.o test_utils.o -L/home/hornik/tmp/R-d-clang-18/lib -lR
> |
> | My initial thought was an issue in the configure scripts (which we borrowed heavily from RcppArmadillo but made slight changes to (which is the most likely cause if there is issue here)) or that there is some mismatch somewhere as to whether or not OpenMP is available, but there isn't an obvious bug to me.
> |
> | Any guidance on how to debug would be greatly appreciated!

> I seem to recall that that machine is 'known-bad' for OpenMP due to the
> reliance on clang-18 which cannot (?) build with it.  Might be best to
> contact Kurt Hornik (CC'ed) and/or CRAN.

> Best, Dirk

> | Thanks,
> | Michelle
> |
> | Michelle Nixon, PhD
> |
> | Assistant Research Professor
> | College of Information Sciences and Technology
> | The Pennsylvania State University
> |
> |     [[alternative HTML version deleted]]
> |
> | ______________________________________________
> | R-package-devel using r-project.org mailing list
> | https://nam10.safelinks.protection.outlook.com/?url=https%3A%2F%2Fstat.ethz.ch%2Fmailman%2Flistinfo%2Fr-package-devel&data=05%7C02%7Cmap5672%40psu.edu%7C7993944f52ae4a017d3d08dc7a34d1d6%7C7cf48d453ddb4389a9c1c115526eb52e%7C0%7C0%7C638519614500278321%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C0%7C%7C%7C&sdata=LdxZWjZQvd1slPS4Tc8GkeKeV5OyoU5YL2NraN6chuY%3D&reserved=0<https://stat.ethz.ch/mailman/listinfo/r-package-devel>

> --
> dirk.eddelbuettel.com | @eddelbuettel | edd using debian.org

	[[alternative HTML version deleted]]