Since BLAS kernels are hand-coded assembly that probably wouldn't change much either.
On Mon, Aug 17, 2015 at 4:34 PM, Elliot Saba <[email protected]> wrote: > Yeah, this would maybe possibly affect things like OpenBLAS performance > (unlikely, as most of the heavy lifting code is in assembly!), or julia > compilation performance, and I don't expect large gains in any case. > -E > > On Mon, Aug 17, 2015 at 12:47 PM, Stefan Karpinski <[email protected]> > wrote: > >> Note that which GCC you use has very little impact on Julia's performance >> since Julia code is generated by LLVM, regardless of how the C code is >> compiled. >> >> On Mon, Aug 17, 2015 at 3:43 PM, Elliot Saba <[email protected]> >> wrote: >> >>> Hello Federico, I don't think that we are using any features in Julia >>> that rely on GCC newer than 4.8. Although it is possible that the newer >>> GCC releases will know new tricks and can generate slightly faster/more >>> efficient code, I don't think there is a compelling reason to use one over >>> the other. >>> -E >>> >>> On Mon, Aug 17, 2015 at 12:01 AM, Federico Calboli <[email protected]> >>> wrote: >>> >>>> Hi All, >>>> >>>> I am using gfortran from GCC: >>>> >>>> GNU Fortran (Homebrew gcc49 4.9.3 --with-fortran) 4.9.3 >>>> >>>> but I am aware GCC 4.9 is not the latest. I was wondering whether it >>>> would be useful to move up to GCC 5.2.x, to future proof myself -- when >>>> Julia 0.4 becomes the new 'stable' I plan to remove my 0.3 and rebuild from >>>> scratch, so I would have a perfect moment to switch to a different GCC... >>>> >>>> Any ideas? >>>> >>>> Cheers >>>> >>>> F >>>> >>> >>> >> >
