[R-pkg-devel] why doesn't CRAN add -fopenmp to SHLIB_OPENMP_CXXFLAGS for Mac OS?

[Mark van der Loo]

It is not a question of whether the CRAN maintainers _could_ do it.
Technically anything's possible[1]. But, as is nicely pointed out[2]

    "The time of the volunteers is CRAN’s most precious resource[...]"

If you ask me, the voluntary CRAN team is already supporting an amazing
range of architectures, which must be a huge job to maintain. Especially
since R and all its packages should behave consistently across all those
architectures.

I agree it is a pity that clang does not support openmp, but that's just
the way it is right now. (note: I use openmp in several packages).

You could contact the CRAN team and ask if they would be prepared to add
the gcc toolchain as an option for OSX. Or, you can contact the clang team
for openmp support.

Other options:
- Switch to C11/C++11 which has (lower-level) paralellization in the
standard and should be supported by any compiler that claims compliance.
- Use higher-level paralellization, such as offered by the parallel or
partools

Best,
Mark

[1] Well, one problem is: how would install.packages know what binary to
choose: the one compiled by clang or the one compiled by gcc?

[2] https://cran.r-project.org/web/packages/policies.html





Op vr 28 aug. 2015 om 15:26 schreef Joshua N Pritikin <jpritikin at pobox.com>:

> For example, mvnfast CRAN binaries could take advantage of OpenMP.
>
> The OpenMx team has been building Mac OS binaries with openmp enabled
> for years. We compile with gcc and bundle the necessary openmp enabled
> libraries with the binary. Why couldn't CRAN do this?
>
> --
> Joshua N. Pritikin
> Department of Psychology
> University of Virginia
> 485 McCormick Rd, Gilmer Hall Room 102
> Charlottesville, VA 22904
> http://people.virginia.edu/~jnp3bc
>
> ______________________________________________
> R-package-devel at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-package-devel
>

	[[alternative HTML version deleted]]