[R-SIG-Mac] R, FORTRAN and Apple Silicon

[Simon Urbanek]

Jim,

thanks for the update, yes, we are aware and we have an ADP now, too, and are assessing the issues. I got base R working using f2c as a stop-gap measure, but it doesn't pass all checks due to fp issues (not necessarily related to Fortran) so there are more moving parts to this. It would be good to be involved in the discussions, so any pointers are welcome.

Thanks,
Simon



> On Aug 25, 2020, at 2:12 AM, Jim Hester <james.f.hester using gmail.com> wrote:
> 
> I recently inquired about the progress of gcc (more specifically gfortran)
> support with Apple's new 'Apple Silicon' architecture. [0]
> 
> As of today it seems gfortran is not available natively, which has
> implications for R support on upcoming Apple hardware.
> 
> Subsequently I have been in email contact with a few people at Apple on
> their progress and wanted to update those here.
> 
> They also have some questions on what features we rely on in the R
> community that I thought some on this mailing list would have better
> insight in answering than I.
> 
> I would be happy to give the apple contacts information off-list, or CC you
> into the email thread if you would like. Alternatively I can communicate
> our answers back to them myself.
> 
> Also I have a Developer Transition Kit at home, so I can run tests if it
> would be beneficial.
> 
> Apple's response follows
> 
>> We've been working with Iain on figuring out a path forward for gfortran,
> they recently published a branch
> https://github.com/iains/gcc-darwin-arm64#readme with some early support in
> gcc. We are aware of some of the issues with FP80 vs FP64 and our last
> internal discussion was trying to ascertain how relevant and important FP80
> is in the fortran community. There are some complexities with FP80 (and
> higher) that are going to make it not very timely, even native, on Apple
> silicon. If we could set aside FP80+ as not critical it would allow us to
> focus on FP64. What are your thoughts?
> 
>> I think we are able to support technical discussions if people have
> questions and we've made hardware available to individuals as
> needed/requested as well.
> 
>> I did see those tweets when they were posted, we'd already been
> discussing Fortran with several vendors that depend on it, like Matlab.  It
> would also be good to understand example workloads that are important so we
> can take a look at them on Rosetta today, since any effort to go native
> will likely rely on some period of Rosetta as a fall back and we'd like to
> ensure thats feasible.
> 
> [0]: https://twitter.com/jimhester_/status/1292821727165194240
> 
> 	[[alternative HTML version deleted]]
> 
> _______________________________________________
> R-SIG-Mac mailing list
> R-SIG-Mac using r-project.org
> https://stat.ethz.ch/mailman/listinfo/r-sig-mac