Thanks for your replies. I wasn't aware of `CFLAGS` and it works great.
I think it makes sense to enable the SIMD-instructions with #27103 <https://trac.sagemath.org/ticket/27103> no matter if all of sage is compiled with `-march=native` by default. All it really takes to get the intrinsics working is to recompile the sources for CombinatorialPolyhedron. I think one could mention this option in the docstring of those function, which profit from it and hope that users who really want the performance read the docstring. Is it acceptable to change the setup file to compile with `-march=native` as a default e.g. if GCC >= 5.1 or clang >= 6.0 (and of course SAGE_FAT_BINARIES is not set)? (GCC 5.1 fixed a bug, so this is important, clang 6.0 is a wild guess, I have no clue when clang added march=native and if they ever fixed an important bug). I added a branch with my very naive approach of doing it in setup.py to #27122 <https://trac.sagemath.org/ticket/27122>. Am Dienstag, 22. Oktober 2019 13:55:18 UTC+2 schrieb Michael Orlitzky: > > On 10/22/19 7:41 AM, Dima Pasechnik wrote: > >> > >> Set CFLAGS="-march=native" in your shell before running "make", and > then > >> pray that the sage packages respect that variable. > > > > many packages would ignore it, and C++ and Fortran modules won't be > > affected by this, either. > > Well there's also CXXFLAGS and FCFLAGS for those. > > If packages don't respect them, then it's an upstream or sage bug, but > we're all probably better off concentrating on getting sage to use our > distro packages where we've already fixed those bugs. > -- You received this message because you are subscribed to the Google Groups "sage-devel" group. To unsubscribe from this group and stop receiving emails from it, send an email to sage-devel+unsubscr...@googlegroups.com. To view this discussion on the web visit https://groups.google.com/d/msgid/sage-devel/caec034f-d3c9-4c42-8248-44a33727eda9%40googlegroups.com.
