[R-SIG-Mac] Just a thought on FAT packages

Simon Urbanek simon.urbanek at r-project.org
Tue Feb 26 15:40:32 CET 2008


Bill,

I'm not sure what you're talking about - R has separate files for each  
architecture. That approach ensures that we can choose the  
architecture at will. We don't use any 4-way executables (except for  
an extra copy of Rscript - there is a convenience fat binary, but each  
individual arch binary is kept aside, so you can still choose each  
arch). In fact, the default in the CRAN R is 32-bit for both ppc and  
Intel unless you choose 64-bit using --arch.

Cheers,
Simon

On Feb 25, 2008, at 10:44 PM, Bill Northcott wrote:

> I notice that the FAT MacOS X binaries being built at the moment are
> all 4 way fat.
>
> When a fat executable is launched, it will run as 64 bit, on a Mac
> with G5 or Core2Duo processors.  There is no way of overriding this
> behaviour.
>
> On a Core2Duo machine this is good because the x86_64 ISA is a
> considerable improvement on the old i386 one, and all applications
> will generally run faster on this architecture in 64 bit.  However, on
> the G5 the ISAs are the same for both 32 and 64 bit, only the pointer
> size changes.  So that 64 bit actually runs more slowly.  The only use
> for ppc64 is for programs that need very large memory spaces on G5
> machines with at least 3GB of DRAM.  I expect the sort of power users
> who run such code and can afford large amounts of DRAM have upgraded
> to a Mac Pro already.
>
> In short, I can see very little demand for ppc64 and including it just
> reduces performance for those running G5 Macs.
>
> Hence my suggestion to make Universal Binaries 3 way fat i386, x86_64
> and ppc.
>
> Just a thought
> Bill
>
> _______________________________________________
> R-SIG-Mac mailing list
> R-SIG-Mac at stat.math.ethz.ch
> https://stat.ethz.ch/mailman/listinfo/r-sig-mac
>
>



More information about the R-SIG-Mac mailing list