On Sun, Sep 8, 2013 at 5:42 PM, Armin Rigo <ar...@tunes.org> wrote: > Hi Alex, > > On Sun, Sep 8, 2013 at 5:33 PM, Alex Gaynor <alex.gay...@gmail.com> wrote: >> LLVM also has a link time optimization, is it on by default in LLVM, or do >> we need to benchmark with it enabled explicitly? > > The point I made in my mail was that the llvm backend is written in a > way that makes link-time optimizations unnecessary. We could also not > rely on "-flto" and instead write a single big .c file with the word > "static" added everywhere. > > > A bientôt, > > Armin.
One C file sounds bad, but we can add -ftlo and add a word "static" a bit everywhere too (I don't think we care for non-exported symbols at all). To be honest, a separate intermediate file is a very good idea (tm), for various reasons, like it would be trivial to parallelize the C-generation-from-something step. If we can make the low-level graphs a file format, we can even kinda-parallelize other steps, like JIT or GC. Cheers, fijal _______________________________________________ pypy-dev mailing list pypy-dev@python.org https://mail.python.org/mailman/listinfo/pypy-dev
