Finding the machine CPU arch

[Rainer Müller]

On 2013-03-06 16:19, Ryan Schmidt wrote:
> Atlas seems to be one of those performance-critical ports that could
> benefit by being compiled for the user's specific processor instead
> of using a pre-compiled binary that might have been build for a
> slightly different processor. To that end, Vincent is going to be
> clearing archive_sites in the next update to the atlas port. Users
> could still manually transport an archive from one machine to another
> and thereby possibly get suboptimal performance, but I think we're
> assuming users won't do that.

Couldn't we just select variants like +sse, +sse2, etc. based on the
current architecture?

The following command can be used to check which features with the
matching command line flags will be selected by -march=native on the
current hardware:
  gcc-mp-4.7 -Q --help=target -march=native

I know that other software like OpenSSL builds different versions
("engines") of performance critical functions for different CPU
instruction sets and selects the appropriate code to run. Doesn't atlas
use anything similar like this? How do other binary distributions handle
atlas in this regard (e.g. Debian/Ubuntu)?

Rainer