gcc and xgetbv

[Rainer Müller]

On 2013-01-25 16:13, Michael Dickens wrote:
> Hi Vincent - Thanks for your reply; it seems like that thread from a
> while back said pretty much what you're now saying.  So, what I'm trying
> to do is create a configure-time test which first tries to compile a
> "xgetbv" instruction, and if that succeeds then it tries to execute the
> binary.  If the first fails then the compiler does not support AVX; if
> the first succeeds but the second fails then the CPU does not support
> AVX.  In this way, we can rule in our out using AVX during
> configuration, and set some variables accordingly to be used during
> building.  This seems like the most reliable way to go, instead of
> relying on the compiler version alone (which is what we were doing), or
> just compiling (which clang does OK). - MLD

This will still become a problem if we are going to distribute this
software as a pre-compiled binary, as the machine running the buildbot
might have support for AVX instructions while the end-user might not
have that.

Do we have any policy for this yet?
Does the buildbot use any specific -march= setting?

Rainer